Information in this document is provided in connection with Intel® products. No license, express or implied, by estoppel or otherwise, to any intellectual
property rights is granted by this document. Except as provided in Intel’s Terms and Conditions of Sale for such products, Intel assumes no liability
whatsoever, and Intel disclaims any express or implied warranty, relating to sale and/or use of Intel products including liability or warranties relating to
fitness for a particular purpose, merchantability, or infringement of any patent, copyright or other intellectual property right. Intel products are not
intended for use in medical, life saving, or life sustaining applications.
Intel may make changes to specifications and product descriptions at any time, without notice.
Designers must not rely on the absence or characteristics of any features or instructions marked "reserved" or "undefined." Intel reserves these for
future definition and shall have no responsibility whatsoever for conflicts or incompatibilities arising from future changes to them.
®
The Intel
Current characterized errata are available on request.
Contact your local Intel sales office or your distributor to obtain the latest specifications and before placing your product order.
I
Implementations of the I
Corporation.
Alert on LAN is a result of the Intel-IBM Advanced Manageability Alliance and a trademark of IBM
Copies of documents which have an ordering number and are referenced in this document, or other Intel literature, may be obtained from:
Intel, Pentium II, Pentium III, Celeron, and MMX are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States
and other countries.
*Other names and brands may be claimed as the property of others.
Table 32. Power Delivery Terminology............................................................................141
Table 33. Power Sequencing Timing Definitions .............................................................151
Table 34. Recommendations For Unused AGP Port ......................................................159
R
®
10 Intel
815 Chipset Platform Design Guide
R
Revision History
Rev. No. Description Date
-001 Initial Release. April 2001
®
815 Chipset Platform Design Guide 11
Intel
R
This page is intentionally left blank.
®
12 Intel
815 Chipset Platform Design Guide
Introduction
R
1 Introduction
This design guid e organizes Int el’s design reco mmendat ions for the Intel® 815 chipset platform for
use with the Universal Socket 370. In addition to providing motherboard design recommendations
(e.g., layout and routing guid elines), this document also addresses system design issues
(e.g., thermal requirements).
This document contains design recommendations, board schematics, debug recommendatio ns, and
a system checklist. These design guidelines have been developed to ensure maximum flexibility
for board designers, while reducing the risk of board-related issues.
Board designers can use the schematics in Appendix A: Customer Reference Board (CRB) as a
reference. While the schematics cover specific designs, the core schematics will remain the same
for most Intel 815 chipset designs for use with the Universal Socket 370. Consult the debug
recommendations when debugging your design. However, these debug recommendations should
be understood before completing board design, to ensure that the debug port, in addition to other
debug features, are implemented correctly.
The Intel 815 chipset platform supports the following processors:
®
• Intel
• Intel
Pentium® III processor based on 0.18 micron technology (CPUID = 068xh).
®
Celeron™ processor based on 0.18 micron technology (CPUID = 068xh). This applies
to Celeron 533A MHz and ≥566 MHz processors
• Future 0.13 micron socket 370 processors
Note: The system bus speed supported by the design is based on the capabilities of the processor,
chipset, and clock driver.
Note: The Intel 815 chipset for use with the universal socket 370 is not compatible with the
®
Intel
Pentium® II processor (CPUID = 066xh) 370-pin socket.
®
815 Chipset Platform Design Guide 13
Intel
Introduction
1.1 Terminology
This section describes some of the terms used in this document. Additional power delivery term
definitions are provided at the beginning of Chapter 12, Power Delivery.
Term Description
R
Aggressor A network that transmits a coupled signal to another network is called the
Aggressor A network that transmits a coupled signal to another network is called the
AGP Accelerated Graphics Port
AGTL/AGTL+ Refers to processor bus signals that are implemented using either Assisted
Bus Agent A component or group of components that, when combined, represent a single
Crosstalk The reception on a victim network of a signal imposed by aggressor network(s)
GMCH Graphics and Memory Controller Hub. A component of the Intel® 815 chipset
aggressor network.
aggressor network.
Gunning Transceiver Logic (AGTL+) or its lower voltage variant (AGTL),
depending on which processor is being used.
load on the AGTL+ bus.
through inductive and capacitive coupling between the networks.
• Backward Crosstalk–coupling that creates a signal in a victim network that
travels in the opposite direction as the aggressor’s signal.
• Forward Crosstalk–coupling that creates a signal in a victim network that travels
in the same direction as the aggressor’s signal.
• Even Mode Crosstalk–coupling from single or multiple aggressors when all the
aggressors switch in the same direction that the victim is switching.
• Odd Mode Crosstalk–coupling from single or multiple aggressors when all the
aggressors switch in the opposite direction that the victim is switching.
platform for use with the Universal Socket 370
ICH Intel® 82801AA I/O Controller Hub component.
ISI Inter-symbol interference is the effect of a previous signal (or transition) on the
Network Length The distance between agent 0 pins and the agent pins at the far end of the bus.
Pad The electrical contact point of a semiconductor die to the package substrate. A
Pin The contact point of a component package to the traces on a substrate such as
14 Intel
interconnect delay. For example, when a signal is transmitted down a line and the
reflections due to the transition have not completely dissipated, the following data
transition launched onto the bus is affected. ISI is dependent upon frequency,
time delay of the line, and the reflection coefficient at the driver and receiver. ISI
can impact both timing and signal integrity.
pad is only observable in simulation.
the motherboard. Signal quality and timings can be measured at the pin.
®
815 Chipset Platform Design Guide
Introduction
R
Term Description
Ringback The voltage that a signal rings back to after achieving its maximum absolute
Setup Window The time between the beginning of Setup to Clock (T
SSO Simultaneous Switching Output (SSO) Effects refers to the difference in electrical
Stub The branch from the bus trunk terminating at the pad of an agent.
System Bus The system bus is the processor bus.
Trunk The main connection, excluding interconnect branches, from one end agent pad
Undershoot Minimum voltage observed for a signal to extend below VSS at the device pad.
Universal Socket 370 Refers to the Intel 815 chipset using the “universal” PGA370 socket. In general,
value. Ringback may be due to reflections, driver oscillations, or other
transmission line phenomena.
) and the arrival of a
valid clock edge. This window may be different for each type of bus agent in the
system.
timing parameters and degradation in signal quality caused by multiple signal
outputs simultaneously switching voltage levels (e.g., high-to-low) in the opposite
direction from a single signal (e.g., low-to-high) or in the same direction (e.g.,
high-to-low). These are respectively called odd-mode switching and even-mode
switching. This simultaneous switching of multiple outputs creates higher current
swings that may cause additional propagation delay (or “push-out”), or a decrease
in propagation delay (or “pull-in”). These SSO effects may impact the setup
and/or hold times and are not always taken into account by simulations. System
timing budgets should include margin for SSO effects.
to the other end agent pad.
these designs support 66/100/133 MHz system bus operation, VRM 8.5 DC-DC
converter guidelines, and Intel
Pentium® III processor (CPUID=068xh), and future Pentium III processors in
single-microprocessor based designs.
®
Celeron™ processors (CPUID=068xh), Intel
SU_MIN
®
Victim A network that receives a coupled crosstalk signal from another network is called
the victim network.
®
815 Chipset Platform Design Guide 15
Intel
Introduction
1.2 Reference Documents
R
Document Document Number
Intel® 815 Chipset Family: 82815 Graphics and Memory Controller Hub (GMCH) for
use with the Universal Socket 370 Datasheet
Pentium® III Processor Specification Update (latest revision from website) (http://developer.intel
AP 907 Pentium® III Processor Power Distribution Guidelines 245085
AP-585 Pentium® II Processor AGTL+ Guidelines 243330
AP-587 Pentium® II Processor Pow er Di stribution Guidelines 243332
Accelerated Graphics P ort Specification, Revision 2.0 (ftp://download.intel.c
PCI Local Bus Specification, Revision 2.2
Universal Serial Bus S pecification, Revision 1.0
/ Location
298351
.com/design/Pentium
III/specupdt/)
om/technology/agp/d
ownloads/agp20.pdf)
1.3 System Overview
The Intel 815 chipset for use with the Universal Socket 370 contains a Graphics Memory
Controller Hub (GMCH) component and I/O Controller Hub (ICH) component for desktop
platforms.
The GMCH provides the processor interface (optimized for the Pentium
068xh) and future 0.13 micron 370 socket processors), DRAM interface, hub interface, and an
accelerated Graphics Port (AGP) interface or internal graphics. This product provides flexibility
and scalability in graphics and memory subsystem performance. Competitive internal graphics
may be scaled via an AGP card interface, and PC100 SDRAM system memory may be scaled to
PC133 system memory.
The Accelerated Hub Architecture interface (i.e., the chipset component interconnect) is designed
into the chipset to provide an efficient, high-bandwidth communication channel between the
GMCH and the I/O controller hub. The chipset architecture also enables a security and
manageability infrastructure through the Firmware Hub component.
An ACPI-compliant Intel 815 chipset platform for use with the universal socket 370 can support
the Full-on (S0), Stop Grant (S1), Suspend to RAM (S3), Suspend to Disk (S4), and Soft-off (S5)
power management states. The chipset also supports Wake-on-LAN
troubleshooting. The chipset architecture removes the requirement of the ISA expansion bus that
was traditionally integrated into the I/O subsystem of PCIsets/AGPsets. This removes many of the
conflicts experienced when installing hardware and drivers into legacy ISA systems. The
III processor (CPUID =
*
for remote administration and
®
16 Intel
815 Chipset Platform Design Guide
Introduction
R
elimination of ISA provides true plug-and-play for the platform. Traditionally, the ISA interface
was used for audio and modem devices. The addition of AC’97 allows the OEM to use software-configurable AC’97 audio and modem coder/decoders (codecs), instead of the traditional ISA
devices.
1.3.1 System Features
The Intel 815 chipset for use with the Universal Socket 370 platform contains two components:
the Intel
®
82815 Graphics and Memory Controller Hub (GMCH) and the Intel® 82801AA I/O
Controller Hub (ICH). The GMCH integrates a 66/100/133 MHz, P6 family system bus controller,
integrated 2D/3D graphics accelerator or AGP (2X/4X) discrete graphics card, 100/133 MHz
SDRAM controller, and a high-speed accelerated hub architecture interface for communication
with the ICH. The ICH integrates an Ultra ATA/66 controller, USB host controller, LPC interface
controller, FWH interface controller, PCI interface controller, AC’97 digital controller, and a hub
interface for communication with the GMCH.
Figure 1. System Block Diagram
AGP G raphics Card
or
Display Cache
(AGP in-line m emory
module)
Analog display out
Digital video out
Audio codec
Modem codec
AGP 2X/4X
2x USB
2x ID E
AC97
Chipset
Processor
66/100/133 MHz system bus
GMCH
(544 BGA)
Hub interface
ICH
PCI bus
LPC I/F
100/133 MHz
SDRAM
PCI slots
KBC /SIO
FWH Flash
BIOS
sys_blk
®
815 Chipset Platform Design Guide 17
Intel
Introduction
1.3.2 Component Features
Figure 2. GMCH Block Diagram
R
GPA
or AGP
2X/4X
card
AGP I/F
Local memory I/F
System bus (66/100/133 MHz)
Processor I/F
Data
stream
control &
dispatch
Hub I/F
Hub
Primary display
Overlay
H/W cursor
3D pipelin e
2D (blit engine)
1.3.2.1 Graphics Memory Controller Hub (GMCH)
• Processor/System Bus Support
Optimized for Intel
frequency
Support for Intel
Supports 32-bit AGTL or AGTL+ bus addressing
Supports uniprocessor systems
Utilizes AGTL and AGTL+ bus driver technology (gated AGTL/AGTL+ receivers for
reduced power)
®
Pentium® III processors (CPUID = 068xh) at 133 MHz system bus
®
Celeron™ processors (CPUID = 068xh); 66 MHz system bus
System
memory I/F
RAMDAC
FP / TVout
Internal graphics
comp_blk_1
SDRAM
100/133
MHz, 64 bit
Monitor
Digital
video out
• Integrated DRAM controller
32 MB to 512 MB using 16-Mb/64-Mb/128-Mb technology
Supports up to three double-sided DIMMS (six rows)
100 MHz, 133 MHz SDRAM interface
64-bit data interface
Standard Synchronous DRAM (SDRAM) support (x-1-1-1 access)
Supports only 3.3V DIMM DRAM configurations
No registered DIMM support
Support for symmetrical and asymmetrical DRAM addressing
Support for x8, x16 DRAM device width
Refresh mechanism: CAS-before-RAS only
Support for DIMM serial PD (presence detect) scheme via SMbus interface
STR power management support via self-refresh mode using CKE
®
18 Intel
815 Chipset Platform Design Guide
Introduction
R
• Accelerated Graphics Port (AGP) Interface
Supports AGP 2.0, including 4X AGP data transfers, but not the 2X/4X Fast Write
protocol
AGP universal connector support via dual-mode buffers to allow AGP 2.0 3.3V or 1.5V
signaling
32-deep AGP request queue
AGP address translation mechanism with integrated fully associative 20-entry TLB
High-priority access support
Delayed transaction support for AGP reads that can not be serviced immediately
AGP semantic traffic to the DRAM is not snooped on the system bus and is therefore not
coherent with the processor caches
• Integrated Graphics Controller
Full 2D/3D/DirectX acceleration
Texture-mapped 3D with point sampled, bilinear, trilinear, and anisotropic filtering
Hardware setup with support for strips and fans
Hardware motion compensation assist for software MPEG/DVD decode
Digital Video Out interface adds support for digital displays and TV-Out
PC99A/PC2001 compliant
Integrated 230 MHz DAC
• Integrated Local Graphics Memory Controller (Display Cache)
0 MB to 4 MB (via Graphics Performance Accelerator) using zero, one, or two parts
32-bit data interface
133 MHz memory clock
Supports ONLY 3.3V SDRAMs
• Packaging/Power
544 BGA with local memory port
1.85V (± 3% within margins of 1.795V to 1.9V) core and mixed 3.3V, 1.5V, and
AGTL/AGTL+ I/O
®
815 Chipset Platform Design Guide 19
Intel
Introduction
1.3.2.2 Intel® 82801AA I/O Controller Hub (ICH)
The I/O Controller Hub provides the I/O subsystem with access to the rest of the system, as
follows:
• Upstream accelerated hub architecture interface for access to the GMCH
• PCI 2.2 interface (6 PCI Request/Grant pairs)
• Bus master IDE controller; supports Ultra ATA/66
• USB controller
• I/O APIC
• SMBus controller
• FWH interface
• LPC interface
• AC’97 2.1 interface
• Integrated system management controller
• Alert on LAN*
• IRQ controller
R
• Packaging/Power
241 BGA
3.3V core and 1.8V and 3.3V standby
1.3.2.3 Firmware Hub (FWH)
The hardware features of the firmware hub include:
• An integrated hardware Random Number Generator (RNG)
• Register-based locking
• Hardware-based locking
• 5 GPIs
• Packaging/Power
40-L TSOP and 32-L PLCC
3.3V core and 3.3V / 12V for fast programming
®
20 Intel
815 Chipset Platform Design Guide
Introduction
R
1.3.3 Platform Initiatives
1.3.3.1 Universal Socket 370 Design
The Intel 815 chipset platform for use with the Universal Socket 370 allows systems designers to
build one system that is compatible with the Pentium III processor (CPUID=068xh), Celeron
processor (CPUID=068xh), and future 0.13 micron socket 370 processors. When implemented,
the Intel 815 chipset platform for use with the Universal Socket 370 can detect which processor is
present in the socket and function accordingly.
1.3.3.2 PC 133
The Intel PC133 initiative provides the memory bandwidth necessary to obtain high performance
from the processor and AGP graphics controllers. The platform’s SDRAM interface supports
100 MHz and 133 MHz operations. The latter delivers 1.066 GB/s of theoretical memory
bandwidth compared with the 800-MB/s theoretical memory bandwidth of 100 MHz SDRAM
systems.
1.3.3.3 Accelerated Hub Architecture Interface
As I/O speeds increase, the demand placed on the PCI bus by the I/O bridge becomes significant.
With the addition of AC’97 and Ultra ATA/66, coupled with the existing USB, I/O requirements
could affect PCI bus performance. The chipset platform’s acceleratedhub architecture ensures
that the I/O subsystem, both PCI and integrated I/O features (IDE, AC’97, USB), receives
adequate bandwidth. By placing the I/O bridge on the accelerated hub architecture interface
instead of PCI, I/O functions integrated into the ICH and the PCI peripherals are ensured the
bandwidth necessary for peak performance.
1.3.3.4 Internet Streaming SIMD Extensions
The Pentium III processor (CPUID = 068xh) provides 70 new SIMD (single-instruction, multipledata) instructions. The new extensions are floating-point SIMD extensions. Intel
technology provides integer SIMD instructions. The Internet Streaming SIMD extensions
complement the MMX technology SIMD instructions and provide a performance boost to floatingpoint-intensive 3D applications.
1.3.3.5 AGP 2.0
The AGP 2.0 interface allows graphics controllers to access main memory at more than 1 GB/s,
which is twice the bandwidth of previous AGP platforms. AGP 2.0 provides the infrastructure
necessary for photorealistic 3D. In conjunction with the Internet Streaming SIMD Extensions,
AGP 2.0 delivers the next level of 3D graphics performance.
®
MMX™
®
815 Chipset Platform Design Guide 21
Intel
Introduction
1.3.3.6 Manageability
The Intel 815 chipset platform integrates several functions designed to manage the system and
lower the system’s total cost of ownership (TCO) of the system. These system management
functions are designed to report errors, diagnose the system, and recover from system lock-ups,
without the aid of an external microcontroller.
TCO Timer
The ICH integrates a programmable TCO Timer. This timer is used to detect system locks. The
first expiration of the timer generates an SMI# that the system can use to recover from a software
lock. The second expiration of the timer causes a system reset to recover from a hardware lock.
Processor Present Indicator
The ICH looks for the processor to fetch the first instruction after reset. If the processor does not
fetch the first instruction, the ICH will reboot the system.
R
Function Disable
The ICH provides the ability to disable the following functions: AC’97 Modem, AC’97 Audio,
IDE, USB, and SMBus. Once disabled, these functions no longer decode I/O, memory or PCI
configuration space. Also, no interrupts or power management events are generated by the
disabled funct ions.
Intruder Detect
The ICH provides an input signal (INTRUDER#) that can be attached to a switch that is activated
when the system case is opened. The ICH can be programmed to generate an SMI# or TCO event
as the result of an active INTRUDER# signal.
Alert on LAN*
The ICH supports Alert on LAN. In response to a TCO event (intruder detect, thermal event,
processor boot failure), the ICH sends a hard-coded message over the SMBus. A LAN controller
supporting the Alert on LAN protocol can decode this SMBus message and send a message over
the network to alert the network manager.
®
22 Intel
815 Chipset Platform Design Guide
Introduction
R
1.3.3.7 AC’97
The Audio Codec ’97 (AC’97) specification defines a digital interface that can be used to attach an
audio codec (AC), a modem codec (MC), an audio/modem codec (AMC) or both an AC and an
MC. The AC’97 specification defines the interface between the system logic and the audio or
modem codec, known as the AC’97 Digital Link.
The chipset platform’s AC’97 (with the appropriate codecs) not only replaces ISA audio and
modem functionality, but also improves overall platform integration by incorporating the AC’97
digital link. Using the chipset’s integrated AC’97 digital link reduces cost and eases migration
from ISA.
The ICH is an AC’97-compliant controller that supports up to two codecs, with independent PCI
functions for audio and modem. The ICH communicates with the codec(s) via a digital serial link
called the AC-link. All digital audio/modem streams and command/status information are
communicated over the AC-link. Microphone input and left and right audio channels are supported
for a high-quality, two-speaker audio solution. Wake-on-ring-from-suspend also is supported with
an appropriate modem codec.
By using an audio codec, the AC’97 digital link allows for cost-effective, high-quality, integrated
audio. In addition, an AC’97 soft modem can be implemented with the use of a modem codec.
Several system options exist when implementing AC’97. The chipset platform’s integrated digital
link allows two external codecs to be connected to the ICH. The system designer can provide
audio with an audio codec or a modem with a modem codec. For systems requiring both audio and
a modem, there are two solutions: the audio codec and the modem codec can be integrated into an
AMC, or separate audio and modem codecs can be connected to the ICH.
Modem implementation for different countries must be taken into consideration, as telephone
systems may vary. By implementing a split design, the audio codec can be on board and the
modem codec can be placed on a riser. Intel is developing an AC’97 digital link connector. With a
single integrated codec, or AMC, both audio and modem can be routed to a connector near the rear
panel where the external ports can be located.
1.3.3.8 Low-Pin-Count (LPC) Interface
In the Intel 815 chipset platform, the Super I/O (SIO) component has migrated to the Low-PinCount (LPC) interface. Migration to the LPC interface allows for lower-cost Super I/O designs.
The LPC Super I/O component requires the same feature set as traditional Super I/O components.
It should include a keyboard and mouse controller, floppy disk controller, and serial and parallel
ports. In addition to the Super I/O features, an integrated game port is recommended, because the
AC’97 interface does not provide support for a game port. In systems with ISA audio, the game
port typically existed on the audio card. The fifteen-pin game port connector provides for two
joysticks and a two-wire MPU-401 MIDI interface. Consult your preferred Super I/O vendor for a
comprehensive list of the devices offered and the features supported.
In addition, depending on system requirements, specific system I/O requirements may be
integrated into the LPC Super I/O. For example, a USB hub may be integrated to connect to the
ICH USB output and extend it to multiple USB connectors. Other SIO integration targets include a
device bay controller or an ISA-IRQ-to-serial-IRQ converter to support a PCI-to-ISA bridge.
Contact your Super I/O vendor to ensure the availability of the desired LPC Super I/O features.
®
815 Chipset Platform Design Guide 23
Intel
Introduction
R
This page is intentionally left blank.
®
24 Intel
815 Chipset Platform Design Guide
General Design Considerations
R
2 General Design Considerations
This document provides motherboard layout and routing guidelines for systems based on the Intel
815 chipset platform for use with the Universal Socket 370. The document does not discuss the
functional aspects of any bus or the layout guidelines for an add-in device.
If the guidelines listed in this document are not followed, it is very important that thorough signal
integrity and timing simulations be completed for each design. Even when the guidelines are
followed, it is recommended that critical signals be simulated to ensure proper signal integrity and
flight time. Any deviation from these guidelines should be simulated.
The trace impedance typically noted (i.e., 60 Ω ± 15%) is the “nominal” trace impedance for a
5-mil-wide trace. That is, it is the impedance of the trace when not subjected to the fields created
by changing current in neighboring traces. When calculating flight times, it is important to
consider the minimum and maximum impedance of a trace, based on the switching of neighboring
traces. Using wider spaces between the traces can minimize this trace-to-trace coupling. In
addition, these wider spaces reduce the settling time.
Coupling between two traces is a function of the coupled length, the distance separating the traces,
the signal edge rate, and the degree of mutual capacitance and inductance. To minimize the effects
of trace-to-trace coupling, the routing guidelines documented in this section should be followed.
Additionally, the routing guidelines in this document are created using a PCB stack-up similar to
that described in the following section.
2.1 Nominal Board Stackup
The Intel 815 chipset platform requires a board stack-up yielding a target impedance of
60 Ω ± 15%, with a 5-mil nominal trace width. Figure 3 shows an example stack-up that achieves
this. It is a 4-layer printed circuit board (PCB) construction using 53%-resin, FR4 material.
Figure 3. Board Construction Example for 60 ΩΩΩΩ Nominal Stackup
Component-side layer 1: ½ oz. Cu
4.5-mil prepreg
Power plane layer 2: 1 oz. Cu
~48-mil Core
Ground layer 3: 1 oz. Cu
4.5-mil prepreg
Solder-side layer 4: ½ oz. Cu
Total thickness:
62 mils
board_4.5mil_stackup
®
815 Chipset Platform Design Guide 25
Intel
General Design Considerations
This page is intentionally left blank.
R
®
26 Intel
815 Chipset Platform Design Guide
Component Quadrant Layouts
R
3 Component Quadrant Layouts
Figure 4 illustrates the relative signal quadrant locations on the GMCH ballout. It does not
represent the actual ballout. Refer to the Intel
®
815 Chipset Family: 82815 Graphics and Memory
Controller Hub (GMCH) for use with the Universal Socket 370 Datasheet for the actual ballout.
The universal socket 370 platform supports Pentium III processor (CPUID=068xh) and Celeron
processor (CPUID=068xh) as well as future 0.13 micron socket 370 processors. The Pentium III
processor (CPUID=068xh) and Celeron processor (CPUID=068xh) have different requirements
for functioning properly in a platform than the future 0.13 micron socket 370 processors. It is
necessary to understand these differences and how they affect the design of the platform. Refer to
Table 1 through Table 4 for a high-level description of the differenc es that require additional
circuitry on the motherboard. Specific details on implementing this circuitry are discussed further
in this chapter. For a detailed description of the differences between the Pentium III processor
(CPUID=068xh) / Celeron processor (CPUID=068xh), and future 0.13 micron socket 370
processor pins, refer to Section 5.4.
Table 1. Processor Considerations for Universal Socket 370 Design
Signal
Name or
Pin
Number
AF36 VSS No connect Addition of circuitry that generates a
AG1 VSS VTT Addition of FET switch to ground or VTT,
AJ3 VSS RESET Addition of stuffing option for pull-down to
AK22 GTL_REF VCMOS_REF Addition of resistor-divider network to provide
PICCLK Requires 2.5V Requires 2.0V Addition of FET switch to provide proper
Function In
®
Pentium®
Intel
III Processor
(CPUID=068xh)
and Intel
Celeron™
Processor
(CPUID=068xh)
®
Function In
Future
0.13 Micron
Socket 370
Processors
Universal Socket 370 Design
processor identification signal used to
configure board-level operation.
controlled by processor identification signal.
Note: FET must have no more than 100
milliohms resistance between source
and drain.
ground, which lets designer prevent future
0.13 micron socket 370 processors from
being used with incompatible stepping of
®
Intel
82815 GMCH.
1.0V, which will satisfy voltage tolerance
requirements of the Intel
processor (CPUID=068xh) and Intel
Celeron™ processor (CPUID=068xh) as well
as future 0.13 micron socket 370 processors.
voltage, controlled by processor identification
signal.
Implementation for
®
Pentium® III
®
®
815 Chipset Platform Design Guide 29
Intel
Universal Socket 370 Design
Signal
Name or
Pin
Number
Function In
®
Pentium®
Intel
III Processor
(CPUID=068xh)
and Intel
Celeron™
Processor
(CPUID=068xh)
®
Function In
Future
0.13 Micron
Socket 370
Processors
Implementation for
Universal Socket 370 Design
R
PWRGOOD Requires 2.5V Requires 1.8V Addition of resistor-divider network to provide
VTT Requires 1.5V Requires 1.25V Modification to VTT generation circuit to
VTTPWRGD Not used Input signal to
future
0.13 micron
socket 370
processors to
indicate that VID
signals are stable
2.1V, which will satisfy voltage tolerance
requirements of the Pentium III processor
(CPUID=068xh) and Celeron processor
(CPUID=068xh) as well as future 0.13 micron
socket 370 processors.
switch between 1.5V or 1.25V, controlled by
processor identification signal.
Addition of VTTPWRGD generation circuit.
Table 2. GMCH Considerations for Universal Socket 370 Design
Pin
Name/Number
SMAA12 New strap required for determining
Pentium
(CPUID=068xh) and Intel
Processor (CPUID=068xh)
or
Future 0.13 micron socket 370
processors
Issue Implementation For
Addition of FET switch controlled by
®
III Processor
®
Celeron™
processor identification signal.
Universal Socket 370 Design
Table 3. ICH Considerations for Universal Socket 370 Design
Signal Issue Implementation For
PWROK GMCH and Intel® CK-815 must not
sample BSEL[1:0] until VTTPWRGD
asserted. ICH must not initialize
before Intel CK-815 clocks stabilize
30 Intel
Universal Socket 370 Design
Addition of circuitry to have VTTPWRGD
gate PWROK from power supply to ICH. The
ICH will hold the GMCH in reset until
VTTPWRGD asserted plus 20 ms time delay
to allow Intel CK-815 clocks to stabilize.
®
815 Chipset Platform Design Guide
Universal Socket 370 Design
R
Table 4. Clock Synthesizer Considerations for Universal Socket 370 Design
Signal Issue Implementation For
VDD Intel® CK-815 does not support
VTTPWRGD
Universal Socket 370 Design
Addition of FET switch which supplies power
to VDD only when VTTPWRGD is asserted.
Note: FET must have no more than
100 milliohms resistance between
source and drain.
4.2 Processor Design Requirements
4.2.1 Use of Universal Socket 370 Design with Incompatible
GMCH
The universal socket 370 design is intended for use with the Intel 815 chipset platform for use with
the universal socket 370. A universal socket 370 design populated with an earlier stepping of the
GMCH is not compatible with future 0.13 micron socket 370 processors and, if used, will cause
eventual failure of these processors. To prevent a future 0.13 micron socket 370 processor from
being used with an incompatible stepping of the GMCH, the recommendation is to lay out the site
for a 0 Ω pull-down to ground on processor pin AJ3. This pin is a RESET# signal on future
0.13 micron socket 370 processors and, by populating the resistor, these future processors will be
prevented from functioning when placed in a board with an incompatible stepping of the GMCH.
All Pentium III (CPUID=068xh) and Celeron (CPUID=068xh) processors will continue to boot
normally. Not populating the resistor will allow future 0.13 micron socket 370 processors to boot.
Refer to Figure 7 for an example implementation.
For the platform to configure for the requirements of the processor in the socket, it must first
identify whether the processor is a Pentium III processor (CPUID=068xh) / Celeron processor
(CPUID=068xh), or a future 0.13 micron socket 370 processors. Pin AF36 is a ground pin on a
Pentium III processor (CPUID=068xh) and Celeron processor (CPUID=068xh); pin AF36 is an
unconnected pin on future 0.13 micron Socket 370 processors. Referring to Figure 8, the platform
uses a detect circuit connected to this processor pin. If a future 0.13 micron Socket 370 processor
is present in the socket, the TUAL5 reference schematic signal will be pulled to the 5V rail and the
TUAL5# reference schematic signal will be pulled to ground. Otherwise, for a Pentium III
processor (CPUID=068xh) or Celeron processor (CPUID=068xh), the TUAL5 reference
schematic signal will be pulled to ground and the TUAL5# will be pulled to the 5V rail.
Figure 8. Processor Detect Mechanism at Socket/TUAL5 Generation Circuit
VCC5
R
Processor Pin
AF36
1 KΩ
VTT
VCC5
2.2 KΩ
NPN
2.2 KΩ
MOSFET N
TUAL5#
Proc_Detect
TUAL5
®
32 Intel
815 Chipset Platform Design Guide
Universal Socket 370 Design
R
4.2.3 Setting the Appropriate Processor VTT Level
Because the Pentium III processor (CPUID=068xh) / Celeron processor (CPUID=068xh), and
future 0.13 micron socket 370 processors require different VTT levels, the platform must be able
to provide the appropriate voltage level after determining which processor is in the socket.
Referring to Figure 9, the TUAL5 reference schematic signal serves to control the FET, and by
doing so determines whether the voltage regulator supplies 1.25V or 1.5V to VTT for AGTL or
AGTL+, respectively.
Figure 9. VTT
Selection Switch
VCC3_3
10
µF
LT1587-ADJ
Vin
TUAL5
Vout
ADJ
µF
0.1
MOSFET N
49.9
1%
10
1%
VTT
Ω
22
µF
Tantalum
Ω
Vtt_Sel_Sw
®
815 Chipset Platform Design Guide 33
Intel
Universal Socket 370 Design
4.2.4 VTT Processor Pin AG1
Processor pin AG1 requires additional attention since it is a ground pin on a Pentium III processor
(CPUID=068xh) / Celeron processor (CPUID=068xh) and a VTT pin on a future 0.13 micron
socket 370 processor. A separate switch controlled by the TUAL5 reference schematic signal
determines whether pin AG1 is pulled to ground or VTT. Refer to Figure 10 for an example
implementation.
Figure 10. Switching Pin AG1
TUAL5
VTT
R
1 KΩ
Note: The FET must have no m ore than
100 milliohms resistance between the
source and the drain.
4.2.5 Identifying the Processor at the GMCH
The GMCH determines whether the socket contains a future 0.13 micron socket 370 processor or
Pentium III processor (CPUID=068xh) / Celeron processor (CPUID=068xh) based on the input to
pin SMAA12 on the GMCH. In a system using future 0.13 micron socket 370 processors,
SMAA12 will be pulled down during reset to indicate to the GMCH that a future 0.13 micron
socket 370 processor is in the socket. Refer to Figure 11. for an example implementation.
Processor Pin
AG1
AG1_Switch
®
34 Intel
815 Chipset Platform Design Guide
Universal Socket 370 Design
R
Figure 11. Processor Identification Strap on GMCH
SMAA[12]
10 KΩ
TUAL5
Proc_ID_Strap
Table 5 provides the logic decoding to determine which processor is installed in a PGA370 design.
Table 5. Determining the Installed Processor via Hardware Mechanisms
Low 0 Intel® Pentium® III processor (CPUID=068xh) or Intel® Celeron™
X 1 No processor installed.
CPUPRES# Notes
processor (CPUID=068xh) installed.
®
815 Chipset Platform Design Guide 35
Intel
Universal Socket 370 Design
4.2.6 Configuring Non-VTT Processor Pins
R
When asserted, the VTTPWGRD signal must be level-shifted to 12V to properly drive the gating
circuitry of the Intel
®
CK-815. Furthermore, while the VTTPWRGD signal is connected to the
VTTPWRGD pin on a future 0.13 micron socket 370 processor, on a Pentium III processor
(CPUID=068xh) or Celeron processor (CPUID=068xh) that same pin is a ground. To provide
proper functionality, a 1.0 kΩ resistor must be placed in series between the circuitry that generates
the signal VTTPWRGD and the processor pin VTTPWRGD. Refer to Figure 12 for an example
implementation. Voltage regulators that generate the standard VTTPWRGD signal are available.
Figure 12. VTTPWRGD Configuration Circuit
VCC5
BAT54C
231
V1_8SB
732
1%
1
1%
ΚΩ
Ω
VTT
VCC5
5
3
2
4
Vcc
IN+ 1
Out 1
IN- 1
Gnd
LM393 Ch 1
20 K
1
VCC5
Ω
2.0 ms de lay nominal
V1_8SB
µF
0.1
ASSERTED
8
7
LOW
IN+ 2
Out 2
IN- 2
LM393 Ch2
1 K
6
VCC5
Ω
2.2 K
MOSFET N
VCC12
Ω
MOSFET N
VTTPW RGD12
VTT
1 KΩ
VTTPWRGD5#
VTTPWRGD
1 K
Ω
VTTPWRGD_Config
NOTE: The diode is included so that repeated pressing of the reset or power button does not cause the
capacitor to build up enough charge to circumvent the 20 ms delay.
®
36 Intel
815 Chipset Platform Design Guide
Universal Socket 370 Design
R
4.2.7 VCMOS Reference
In previous platforms supporting the Pentium III processor (CPUID=068xh) and Celeron processor
(CPUID=068xh), VCMOS was generated by the processor itself. The future 0.13 micron socket
370 processors do not generate VCMOS, and the universal platform is required to generate this
separately on the motherboard. Processor pin AK22, which is a GTL_REF pin on a Pentium III
processor (CPUID=068xh) and Celeron processor (CPUID=068xh), has been changed to a
VCMOS_REF pin on future 0.13 micron socket 370 processors. Referring to Figure 13, a network
of resistors and a capacitor must be added so that this pin operates appropriately for whichever
processor is in the socket.
Figure 13. GTL_REF/VCMOS_REF Voltage Divider Network
VCMOS
75 Ω
1%
150
1%
Processor Pin
AK22
Ω
0.1
µF
GTL_CMOS_Ref
®
815 Chipset Platform Design Guide 37
Intel
Universal Socket 370 Design
4.2.8 Processor Signal PWRGOOD
The processor signal PWRGOOD is specified at different voltage levels depending on whether it
is a Pentium III processor (CPUID=068xh) / Celeron processor (CPUID=068xh), or whether it is a
future 0.13 micron socket 370 processor. As there is an overlap between the ranges of accepted
voltage levels for these two processor groups, a resistor divider network that provides 2.1V will
satisfy the requirements of all supported processors. See Figure 14 for an example implementation.
Figure 14. Resistor Divider Network for Processor PWRGOOD
VCC2_5
330
R
Ω
PW RGOO D from ICH 2
1.8 ΚΩ
PWRGOOD to Processor
PWRGOO D_Divider
®
38 Intel
815 Chipset Platform Design Guide
Universal Socket 370 Design
R
4.2.9 APIC Clock Voltage Switching Requirements
The processor’s APIC clock is also specified at different voltage levels depending on whether it is
for the Pentium III processor (CPUID=068xh) / Celeron processor (CPUID=068xh) or whether it
is for a future 0.13 micron socket 370 processor. There is no overlap in the range of accepted
voltage levels for the two processor groups, so a voltage switch is required to ensure proper
operation. Figure 15 shows an example implementation.
Figure 15. Voltage Switch For APIC Clock from Clock Synthesizer to Processor
IOAPIC
30 Ω
APICCLK_CPU
130 Ω
MOSFET N
NOTE: The 30 Ω resistor represents the series resistor typically used in connecting the APIC clock to the
processor.
TUAL5
API_CLK_SW
®
815 Chipset Platform Design Guide 39
Intel
Universal Socket 370 Design
4.2.10 GTLREF Topology and Layout
In a platform supporting the future 0.13 micron socket 370 processors, the voltage requirements
for GTLREF are different for the processor and the chipset. The GTLREF on the processor is
specified to be 2/3 * VTT, while the GTLREF on the chipset is 0.7 * VTT. This difference
requires that separate resistor sites be added to the layout to split the GTLREF sources. In a
universal motherb oard design, a Pentium III processor (CPUID=068xh) and Celeron processor
(CPUID=068xh) will be unaffected by the difference in GTLREF. The recommended GTLREF
circuit topology is shown in Figure 16.
Note: If an A-2 stepping of the GMCH is used with the universal motherboard design, the GTLREF for
the GMCH should be set at 2/3 * VTT. This requires changing the 63.4 Ω, 1% resistor on the
GMCH side to 75 Ω, 1%.
Figure 16. GTLREF Circuit Topology
VTT
R
63.4 Ω75 Ω
ProcessorGMCH
150 Ω150 Ω
gtlref_circuit
GTLREF Layout and Routing Guidelines
• Place all resistor sites for GTLREF generation close to the GMCH.
• Route GTLREF with as wide a trace as possible.
• Use one 0.1 µF decoupling capacitor for every two GTLREF pins at the processor (four
capacitors total). Place as close as possible (within 500 mils) to the Socket 370 GTLREF pins.
• Use one 0.1 µF decoupling capacitor for each of the two GTLREF pins at the GMCH
(two capacitors total). Place as close as possible to the GMCH GTLREF balls.
Given the higher GTLREF level fo r the GMCH, a debug test hook should be added for validation
purposes. The debug test hook should be placed on the processor signal ADS# and consists of
laying down the site for a 56 Ω pull-up to VTT. The resistor site should be located within 150 mils
of the GMCH, and placed as close to the ADS# signal trace as possible.
®
40 Intel
815 Chipset Platform Design Guide
Universal Socket 370 Design
R
4.3 Power Sequencing on Wake Events
In addition to the mechanism for identifying the processor in the socket, special handling of wake
events is required for the Intel 815 chipset platform that support functionality of the future
0.13 micron socket 370 processors. When a wake event is triggered, the GMCH and the Intel
CK-815 must not sample BSEL[1:0] until the signal VTTPWRGD is asserted. This is handled by
setting up the following sequence of events:
1. Power is not connected to the Intel CK-815-compliant clock driver until VTTPWRGD12 is
asserted.
2. Clocks to the ICH stabilize before the power supply asserts PWROK to the ICH. There is no
guarantee this will occur as the implementation for the previous step relies on the 12V supply.
Thus, it is necessary to gate PWROK to the ICH from the power supply while the Intel
CK-815 is given sufficient time for the clocks to become stable. The amount of time required
is a minimum 20 ms.
3. ICH takes the GMCH out of reset.
4. GMCH samples BSEL[1:0]. Intel CK-815 will have sampled BSEL[1:0] much earlier.
4.3.1 Gating of Intel® CK-815 to VTTPWRGD
System designers must ensure that the VTTPWRGD signal is asserted before the Intel CK-815compliant clock driver receives power. This is handled by having the 3.3V rail of the clock driver
gated by the VTTPWRGD12 reference schematic signal. Unlike previous Intel 815 chipset
designs, the 3.3V standby rail is not used to power the clock as the VTTPWRGD12 reference
schematic signal will cut power to the clock when going into any sleep state. Refer to Figure 17 for
an example implementation.
Figure 17. Gating Power to Intel
VTTPW RGD12
Note: The FET must have no more than
100 milliohm s resistan ce between the
source and the drain.
®
CK-815
VCC3_3
MOSFET N
VDD on CK-815
Gating_Pwr
®
815 Chipset Platform Design Guide 41
Intel
Universal Socket 370 Design
4.3.2 Gating of PWROK to ICH
With power being gated to the Intel CK-815 by the signal VTTPWRGD12, it is important that the
clocks to the ICH are stable before the power supply asserts PWROK to the ICH. As the clocking
power gating circuitry relies on the 12V supply, there is no guarantee that these conditions will be
met. This is why an estimated minimum time delay of 20 ms must be added after power is
connected to the Intel CK-815 to give the clock driver sufficient time to stabilize. This time delay
will gate the power supply’s assertion of PWROK to the ICH. After the time delay, the power
supply can safely assert PWROK to the ICH, with the ICH subsequently taking the GMCH out of
reset. Refer to Figure 18 for an example implementation.
Figure 18. PWROK Gating Circuit For ICH
R
VDD on CK-815
43 kΩ
1.0 µF
NOTE: The diode is included so that repeated pressing of the reset or power button does not cause the
capacitor to build up enough charge to circumvent the 20 ms delay.
VCC3_3
PWROK
Note:
Delay 20 ms after VDD
on CK-815 is powered
ICH_PWROK
8.2 kΩ
ICH_PWROK_GATING
®
42 Intel
815 Chipset Platform Design Guide
System Bus Design Guidelines
R
5 System Bus Design Guidelines
The Pentium III processor delivers higher performance by integrating the Level-2 cache into the
processor a nd running it at the pr ocessor’s core speed. T he Pentium III processor runs at higher
core and system bus speeds than previous-generation Intel
hardware and software compatibility with earlier Pentium III processors. The new Flip Chip-Pin
Grid Array 2 (FC-PGA2) package technology enables compatibility with previous Flip Chip-Pin
Grid Array (FC-PGA) packages using the PGA370 socket.
This section presents the considerations for designs capable of using the Intel 815 chipset platform
with the full range of Pentium III processors using the PGA370 socket.
5.1 System Bus Routing Guidelines
®
IA-32 processors while maintaining
The following layout guide supports designs using Pentium III processor (CPUID=068xh) /
Celeron processor (CPUID=068xh), and future 0.13 micron socket 370 processors with the Intel
815 chipset platform for use with the universal socket 370. The solution covers system bus speeds
of 66/100/133 MHz for the Pentium III processor (CPUID=068xh) / Celeron processor
(CPUID=068xh), and future 0.13 micron socket 370 processors. All processors must also be
configured to 56 Ω on-die termination.
5.1.1 Initial Timing Analysis
Table 6 lists the AGTL/AGTL+ component timings of the processors and GMCH defined at the
pins.
Note: These timings are for reference only. Obtain each processor’s specifications from the respective
processor datasheet and the chipset values from the appropriate Intel 815 chipset datasheet.
®
815 Chipset Platform Design Guide 43
Intel
System Bus Design Guidelines
R
Table 6. Intel
NOTES:
Table 7 contains an example AGTL+ initial maximum flight time, and Table 8 contains an
example minimum flight time calculation for a 133 MHz, uniprocessor system using the Pentium
III processor and the Intel 815 chipset platform’s system bus. Note that assumed values were used
for the clock skew and clock jitter.
®
Pentium® III Processor AGTL/AGTL+ Parameters for Example Calculations
IC Parameters Intel® Pentium® III Processor at 133 MHz System
Clock to Output maximum
(T
Clock to Output minimum
(T
Setup time (T
Hold time (T
)
CO_MAX
)
CO_MIN
) • 1.20 ns (for BREQ Lines)
SU_MIN
) • 1.0 ns (for 66/100/133 MHz system bus speeds) 0.10 ns 1
HOLD
1. All times in nanoseconds.
2. Numbers in table are for reference only. These timing parameters are subject to change. Check the
appropriate component datasheet for the valid timing parameter values.
3. T
= 2.65 ns assumes that the GMCH sees a minimum edge rate equal to 0.3 V/ns.
SU_MIN
• 3.25 ns (for 66/100/133 MHz system bus speeds) 4.1 ns 1, 2
Note: The clock skew and clock jitter values depend on the clock components and the distribution
method chosen for a particular design and must be budgeted into the initial timing equations, as
appropriate for each design.
Table 7and Table 8 were derived assuming the following:
• CLK
= 0.20 ns (Note: This assumes that the clock driver pin-to-pin skew is reduced to
SKEW
50 ps by tying the two host clock outputs t ogether (i.e., “ganging”) at t he clock driver output
pins, and that the PCB clock routing skew is 150 ps. The system timing budget must assume
0.175 ns of clock driver skew if outputs are not tied together as well as the use of a clock
driver that meets the Intel CK-815 Clock Synthesizer/Driver Specification.)
• CLK
See the respective processor’s datasheet, the appropriate Intel 815 chipset platform documentation,
and the Intel
= 0.250 ns
JITTER
®
CK-815 Clock Synthesizer/Driver Specification for details on clock skew and jitter
specifications. Exact details regarding the host clock routing topology are provided with the
platform design gui deline.
®
44 Intel
815 Chipset Platform Design Guide
System Bus Design Guidelines
R
Table 7. Example T
Driver Receiver Clk Period2T
Processor GMCH 7.50 3.25 2.65 0.20 0.25 0.40 1.1
GMCH Processor 7.50 4.1 1.20 0.20 0.25 0.40 1.35
NOTES:
1. All times in nanoseconds
2. BCLK period = 7.50 ns at 133.33 MHz
Table 8. Example T
Driver Receiver T
Processor GMCH 0.10 0.20 0.40 0.10
GMCH Processor 1.00 0.20 1.05 0.15
NOTES: All times in nanoseconds
The flight times in Table 7 include margin to account for the following phenomena that Intel
observed when multiple bits are switching simultaneously. These multi-bit effects can adversely
affect the flight time and signal quality and sometimes are not accounted for during simulation.
Accordingly, the maximum flight times depend on the baseboard design, and additional adjustment
factors or margins are recommended.
Calculations for 133 MHz Bus 1
FLT_MAX
CO_MAXTSU_MIN
Calculations (Frequency Independent)
FLT_MIN
Clk
HOLD
SKEW TCO_MIN
Clk
Clk
SKEW
Recommended
T
FLT_MIN
JITTER MADJ
Recommended
T
FLT_MAX
• SSO push-out or pull-in
• Rising or falling edge rate degradation at the receiver caused by inductance in the current
return path, requiring extrapolation that causes additional delay
• Cross-talk on the PCB and inside the package which can cause variation in the signals
Additional effects exist that may notnecessarily be covered by the multi-bit adjustment factor
and should be budgeted as appropriate to the baseboard design. These effects are included as M
in the example calculations in Table 7. Examples include:
• The effective board propagation constant (S
), which is a function of:
EFF
Dielectric constant (εr) of the PCB material
Type of trace connecting the components (stripline or microstrip)
Length of the trace and the load of the components on the trace. Note that the board
propagation constant multiplied by the trace length is a component of the flight time, but
not necessarily equal to the flight time.
ADJ
®
815 Chipset Platform Design Guide 45
Intel
System Bus Design Guidelines
5.2 General Topology and Layout Guidelines
Figure 19. Topology for 370-Pin Socket Designs with Single-Ended Termination (SET)
R
GMCH
L(1): Z
= 60 Ω ± 15%
0
Table 9. Trace Guidelines for Figure 19
Description Min. Length (inches) Max. Length (inches)
GMCH to PGA370 socket trace 1.90 4.50
NOTES:
1. All AGTL/AGTL+ bus signals should be referenced to the ground plane for the entire route.
2. Use an intragroup AGTL/AGTL+ spacing : line width : dielectric thickness ratio of at least 2:1:1 for
microstrip geometry. If
routing could use 10-mil spacing, 5-mil traces, and a 5-mil prepreg between the signal layer and the
plane it references (assuming a 4-layer motherboard design).
3. The recommended trace width is 5 mils, but not greater than 6 mils.
ε
= 4.5, this should limit coupling to 3.4%. For example, intragroup AGTL+
r
Table 10 contains the trace width space ratios assumed for this topology. Three types of cross-talk
are considered in this guideline: Intragroup AGTL/AGTL+, Intergroup AGTL/AGTL+, and
AGTL/AGTL+ to non-AGTL/AGTL+. Intragroup AGTL/AGTL+ cross-talk involves interference
between AGTL/AGTL+ signals within the same group. Intergroup AGTL/AGTL+ cross-talk
involves interference from AGTL/AGTL+ signals in a particular group to AGTL/AGTL+ signals
in a different group. An examp le o f AGTL/AGTL+ to non-AGTL/AGTL+ cross-talk is when
CMOS and AGTL/AGTL+ signals interfere with each other. The AGTL/AGTL+ signals consist of
the following groups: data signals, control signals, c lock signals, and address signals.
1, 2, 3
PGA370 socket
sys_bus_topo_PG A370
Table 10. Trace Width:Space Guidelines
Cross-Talk Type Trace Width:Space Ratios
Intragroup AGTL/AGTL+ signals (same group AGTL/AGTL+) 5:10 or 6:12
Intergroup AGTL/AGTL+ signals (different group AGTL/AGTL+) 5:15 or 6:18
AGTL/AGTL+ to System Memory Signals 5:30 or 6:36
AGTL/AGTL+ to non-AGTL/AGTL+ 5:25 or 6:24
NOTES:
1. Edge-to-edge spacing.
2. Units are in mils.
1, 2
®
46 Intel
815 Chipset Platform Design Guide
System Bus Design Guidelines
R
5.2.1 Motherboard Layout Rules for AGTL/AGTL+ Signal s
Ground Reference
It is strongly recommended that AGTL/AGTL+ signals be routed on the signal layer next to the
ground layer (referenced to ground). It is important to provide an effective signal return path with
low inductance. The best signal routing is directly adjacent to a solid GND plane with no splits or
cuts. Eliminate parallel traces between layers not separated by a power or ground plane. If a signal
has to go through routing layers, the recommendations a re in the following list.
Note: Following these layout rules is critical for AGTL/AGTL+ signal integrity, particularly for
0.18-micron and smaller process technology.
• For signals going from a ground reference to a power reference, add capacitors between
ground and power near the vias to provide an AC return path. One capacitor should be used
for every three signal lines that change reference layers. Capacitor requirements are as
follows: C=100 nF, ESR=80 mΩ, ESL=0.6 nH. Refer to Figure 20 for an example of
switching reference layers.
• For signals going fr om one ground re ference to another, separate ground reference, add vias
between the two ground planes to provide a better return path.
Figure 20. AGTL/AGTL+ Trace Routing
GMCHProcessor
Layer 2
Layer 3
Reference Plane Splits
Splits in reference planes disrupt signal return paths and increase overshoot/undershoot due to
significantly increased inductance.
Processor Connector Breakout
It is strongly recommended that AGTL/AGTL+ signals do not traverse multiple signal layers. Intel
recommends breaking out all signals from the connector on the same layer. If ro uting is tight,
break out from the connector on the opposite routing layer over a ground reference and cross over
to main signal layer near the processor connector.
0-500 mils
1.2V Power Plane
Ground Plane
1.5-3.5 inches
Socket Pin
AGTL_trace_route
®
815 Chipset Platform Design Guide 47
Intel
System Bus Design Guidelines
Minimizing Cross-Talk
The following general rules minimize the impact of cross-talk in a high-speed AGTL/AGTL+ bus
design:
• Maximize the space between traces. Where possible, maintain a minimum of 10 mils
(assuming a 5-mil trace) between trace edges. It may be necessary to use tighter spacing when
routing between component pins. When traces must be close and parallel to each other,
minimize the distance that they are close together and maximize the distance between the
sections when the spacing restrictions are relaxed.
• Avoid parallelism between signals on adjacent layers, if there is no AC reference plane
between them. As a rule of thumb, route adjacent layers orthogonally.
• Since AGTL/AGTL+ is a low-signal-swing technology, it is important to isolate
AGTL/AGTL+ signals from other signals by at least 25 mils. This will avoid coupling from
signals that have lar ger voltage swings (e.g. , 5V PCI).
• AGTL/AGTL+ signals must be well isolated from system memory signals. AGTL/AGTL+
signal trace edges must be at least 30 mils from system memory trace edges within 100 mils of
the ball of the GMCH.
R
• Select a board stack-up that minimizes the coupling between adjacent signals. Minimize the
nominal characteristic impedance within the AGTL/AGTL+ specification. This can be done
by minimizing the height of the trace from its reference plane, which minimizes cross-talk.
• Route AGTL/AGTL+ address, data, and control signals in separate groups to minimize cross-
talk between groups. Keep at least 15 mils between each group of signals.
• Minimize the dielectric used in the system. This makes the traces closer to their reference
plane and thus reduces the cross-talk magnitude.
• Minimize the dielectric process variation used in the PCB fabrication.
• Minimize the cross-sectional area of the traces. This can be done by means of narrower traces
and/or by using thinner copper, but the trade-off for this smaller cross-sectional area is higher
trace resistivity, which can reduce the falling-edge noise margin because of the I*R loss along
the trace.
®
48 Intel
815 Chipset Platform Design Guide
System Bus Design Guidelines
R
5.2.1.1 Motherboard Layout Rules for Non-AGTL/AGTL+ (CMOS) Signals
Table 11. Routing Guidelines for Non-AGTL/Non-AGTL+ Signals
Signal Trace Width Spacing to Other Traces Trace Length
A20M# 5 mils 10 mils 1” to 9”
FERR# 5 mils 10 mils 1” to 9”
FLUSH# 5 mils 10 mils 1” to 9”
IERR# 5 mils 10 mils 1” to 9”
IGNNE# 5 mils 10 mils 1” to 9”
INIT# 5 mils 10 mils 1” to 9”
LINT[0] (INTR) 5 mils 10 mils 1” to 9”
LINT[1] (NMI) 5 mils 10 mils 1” to 9”
PICD[1:0] 5 mils 10 mils 1” to 9”
PREQ# 5 mils 10 mils 1” to 9”
PWRGOOD 5 mils 10 mils 1” to 9”
SLP# 5 mils 10 mils 1” to 9”
SMI# 5 mils 10 mils 1” to 9”
STPCLK 5 mils 10 mils 1” to 9”
THERMTRIP# 5 mils 10 mils 1” to 9”
NOTE: Route these signals on any layer or combination of layers.
®
815 Chipset Platform Design Guide 49
Intel
System Bus Design Guidelines
5.2.1.2 THRMDP and THRMDN
These traces (THRMDP and THRMDN) route the processor’s thermal diode connections. The
thermal diode operates at very low currents and may be susceptible to cross-talk. T he traces should
be routed close together to reduce loop area and inductance.
Figure 21. Routing for THRMDP and THRMDN
Signal Y
THRMDP
1 — Maximize (min. – 20 mils)
R
2 — Minimize
THRMDN
NOTES:
1. Route these traces parallel and equalize lengths within
2. Route THRMDP and THRMDN on the same layer.
1 — Maximize (min. – 20 mils)
Signal Z
bus_routing_thrmdp-thrmdn
±0.5 inch.
5.2.1.3 Additional Routing and Placement Considerations
• Distribute VTT with a wide trace. A 0.050 inch minimum trace is recommended to minimize
DC losses. Route the VTT trace to all components on the host bus. Be sure to include
decoupling capacitors.
• The VTT voltage should be 1.5V ± 3% for static conditions, and 1.5V ± 9% for worst-case
transient conditions when the Pentium III processor (CPUID=068xh) or Celeron processor
(CPUID=068xh) is present in the socket. If a future 0.13 micron so cket 370 processor is being
used, the VTT voltage should then be 1.25V ± 3% for static conditions, and 1.25V ± 9% for
worst-case transient conditions.
• Place resistor divider pairs for VREF generation at the GMCH component. VREF also is
delivered to the processor.
®
50 Intel
815 Chipset Platform Design Guide
System Bus Design Guidelines
R
5.3 Electrical Differences for Universal PGA370
Designs
There are several electrical changes between previous PGA370 designs and the universal PGA370
design, as follows:
• Changes to the PGA370 socket pin definitions.
• Addition of VTTPWRGD signal to ensure stable VID selection for future 0.13 micron socket
370 processors.
• Addition of THERMTRIP circuit to allow processor to detect catastrophic overheat.
• Addition of VID[25 mV] signal to support future 0.13 micron socket 370 processors.
• Processor VTT level is switchable to 1.25V or 1.5V, depending on which processor is present
in the socket.
• In designs using future 0.13 micron socket 370 processors, the processor does not generate
_REF.
V
CMOS
5.3.1 THERMTRIP Circuit
Figure 22. Example Implementation of THERMTRIP Circuit
VCC1_8SB
2
R10
Ω
1 K
1
Thermtrip#
R8
1.6 K
5.3.1.1 THERMTRIP Timing
When the THERMTRIP signal is asserted, both the VCC and VTT supplies to the processor must
be turned off to prevent thermal runaway of the processor. The time required from THERMTRIP
asserted to VCC rail at ½ nominal is 5 sec and THERMTRIP asserted to VTT rail at ½ nominal is
5 sec. System designers must ensure that the decoupling scheme used on these rails does not
violate the THERMTRIP timing specifications.
VCC1.8
2
R11
Ω
1 K
1
Q2
21
Ω
Q2N3904
Can Use MBT3904
Dual XSTR Part
VCC3_3SB
2
R12
Ω
22 K
Q3
2
1
R9
Ω
1 K
1
Q2N3904
Connect to ICH
SW_ ON#
Thermstrip
®
815 Chipset Platform Design Guide 51
Intel
System Bus Design Guidelines
5.4 PGA370 Socket Definition Details
The following table compares the pin names and functions of the Intel processors supported in the
Intel 815 chipset platform for use with the universal socket 370.
Table 12. Processor Pin Definition Comparison
R
Pin # Pin Name
AA33 Reserved VTT VTT • AGTL/AGTL+ termination
AA35 Reserved VTT VTT • AGTL/AGTL+ termination
AB36 VCC
®
Intel
Celeron™
Processor
(CPUID=068xh)
VCC
CMOS
Pin Name
Intel® Pentium®
III Processor
(CPUID=068xh)
VTT • CMOS voltage level for Intel
CMOS
Pin Name
Future 0.13
Micron
Socket 370
Processors
Function
voltage
voltage
Pentium® III processor
(CPUID=068xh) and Intel
Celeron™ processor
(CPUID=068xh).
• AGTL termination voltage for
future 0.13 micron socket 370
processors.
AD36 VCC1.5 VCC1.5 VTT • VCC1.5 for Pentium III
processor (CPUID=068xh)
and Celeron processor
(CPUID=068xh).
• VTTfor future 0.13 micron
socket 370 processors.
AF36 VSS VSS NC • Ground for Pentium III
processor (CPUID=068xh)
and Celeron processor
(CPUID=068xh).
• No connect for future 0.13
micron socket 370 processors.
1
AG1
VSS VSS VTT • Ground for Pentium III
processor (CPUID=068xh)
and Celeron processor
(CPUID=068xh).
• VTT for future 0.13 micron
socket 370 processors
AH4 Reserved RESET# RESET# • Processor reset for the
AH20 Reserved VTT VTT • AGTL/AGTL+ termination
Pentium III processor (068xh)
and Future 0.13 micron socket
370 processors
voltage
®
®
®
52 Intel
815 Chipset Platform Design Guide
System Bus Design Guidelines
R
Pin # Pin Name
1
AJ3
VSS VSS RESET • Ground for Pentium III
AK4 VSS VSS VTTPWRGD • Ground for Pentium III
AK16 Reserved VTT VTT • AGTL/AGTL+ termination
AK22 GTL_REF GTL_REF VCMOS_REF • GTL reference voltage for
AK36 VSS VSS VID[25mV] • Ground for Pentium III
AL13 Reserved VTT VTT • AGTL/AGTL+ termination
AL21 Reserved VTT VTT • AGTL/AGTL+ termination
AN3 GND GND DYN_OE • Ground for Pentium III
AN11 Reserved VTT VTT • AGTL/AGTL+ termination
AN15 Reserved VTT VTT• AGTL/AGTL+ termination
AN21 Reserved VTT VTT• AGTL/AGTL+ termination
®
Intel
Celeron™
Processor
(CPUID=068xh)
Pin Name
Intel® Pentium®
III Processor
(CPUID=068xh)
Pin Name
Future 0.13
Micron
Socket 370
Processors
Function
processor (CPUID=068xh)
and Celeron processor
(CPUID=068xh).
• RESET for future 0.13 micron
socket 370 processors
processor (CPUID=068xh)
and Celeron processor
(CPUID=068xh).
• VID control signal on future
0.13 micron socket 370
processors.
voltage
Pentium III processor
(CPUID=068xh) and Celeron
processor (CPUID=068xh).
• CMOS reference voltage for
future 0.13 micron socket 370
processors
processor (CPUID=068xh)
and Celeron processor
(CPUID=068xh).
• 25mV step VID select bit for
future 0.13 micron socket 370
processors
voltage
voltage
processor (CPUID=068xh)
and Celeron processor
(CPUID=068xh).
• Dynamic output enable for
future 0.13 micron socket 370
processors
voltage
voltage
voltage
®
815 Chipset Platform Design Guide 53
Intel
System Bus Design Guidelines
Pin # Pin Name
Intel® Celeron™
Processor
(CPUID=068xh)
Pin Name
Intel® Pentium®
III Processor
(CPUID=068xh)
Pin Name
Future 0.13
Micron
Socket 370
Processors
Function
R
E23 Reserved VTT VTT• AGTL/AGTL+ termination
G35 Reserved VTT VTT• AGTL/AGTL+ termination
G37 Reserved Reserved VTT• Reserved for Pentium III
voltage
voltage
processor (CPUID=068xh)
and Celeron processor
(CPUID=068xh).
• AGTL termination voltage for
future 0.13 micron socket 370
processors
2
N37
NC NC NCHCTRL • No connect for Pentium III
processor (CPUID=068xh)
and Celeron processor
(CPUID=068xh).
• NCHCTRL for future 0.13
micron socket 370 processors
S33 Reserved VTT VTT• AGTL/AGTL+ termination
S37 Reserved VTT VTT• AGTL/AGTL+ termination
U35 Reserved VTT VTT• AGTL/AGTL+ termination
U37 Reserved VTT VTT• AGTL/AGTL+ termination
W3 Reserved A34# A34# • Additional AGTL/AGTL+
1
X4
RESET# RESET2# VSS • Processor reset for Pentium III
voltage
voltage
voltage
voltage
address
processor (CPUID=068xh)
and Celeron processor
(CPUID=068xh).
• Ground for future 0.13 micron
socket 370 processors
X6 Reserved A32# A32# • Additional AGTL/AGTL+
X342 VCC
VCC
CORE
VTT• Reserved for Pentium III
CORE
address
processor (CPUID=068xh)
and Celeron processor
(CPUID=068xh).
• AGTL termination voltage for
future 0.13 micron socket 370
processors
®
54 Intel
815 Chipset Platform Design Guide
System Bus Design Guidelines
R
Pin # Pin Name
®
Intel
Celeron™
Processor
(CPUID=068xh)
Pin Name
Intel® Pentium®
III Processor
(CPUID=068xh)
Pin Name
Future 0.13
Micron
Socket 370
Processors
Function
Y1 Reserved Reserved NC • Reserved for Pentium III
processor (CPUID=068xh)
and Celeron processor
(CPUID=068xh).
• No connect for future 0.13
micron socket 370 processors
Y33 Reserved CLKREF CLKREF • 1.25V PLL reference
2
Z36
VCC2.5 VCC2.5 NC • VCC2.5 for Pentium III
processor (CPUID=068xh)
and Celeron processor
(CPUID=068xh).
• No connect for future 0.13
micron socket 370 processors
NOTES:
1. Refer to Chapter 4.
2. Refer to Section 13.2
®
815 Chipset Platform Design Guide 55
Intel
System Bus Design Guidelines
5.5 BSEL[1:0] Implementation Differences
A future 0.13 micron socket 370 processor will select the 133 MHz system bus frequency setting
from the clock synthesizer. A Pentium III processor (CPUID=068xh) utilizes the BSEL1 pin to
select either the 100 MHz or 133 MHz system bus frequency setting from the clock synthesizer.
An Celeron processor (CPUID=068xh) will use both BSEL pins to select 66 MHz system bus
frequency from the clock synthesizer. Processors in an FC-PGA or an
FC-PGA2 are 3.3V tolerant for these signals, as are the clock and chipset.
Intel CK-815 has been designed to support selections of 66 MHz, 100 MHz, and 133 MHz. The
REF input pin has been redefined to be a frequency selection strap (BSEL1) during power-on and
then becomes a 14 MHz reference clock output. The following figure details the new BSEL[1:0]
circuit design for universal PGA370 designs. Note that BSEL[1:0] now are pulled up using 1 kΩ
resistors. Also refer to Figure 24 for more details.
Note: In a design supporting future 0.13 micron socket 370 processors, the BSEL[1:0] lines are not valid
until VTTPWRGD is asserted. Refer to Section 4.2.10 for details.
R
Figure 23. BSEL[1:0] Circuit Implementation for PGA370 Designs
3.3V
3.3V
Processor
1 kΩ1 kΩ
BSEL0BSEL1
Chipset
Clock Driver
sys_ bus_BSEL_PGA370
®
56 Intel
815 Chipset Platform Design Guide
System Bus Design Guidelines
R
5.6 CLKREF Circuit Implementation
The CLKREF input (used by the Pentium III processor (CPUID=068xh), Celeron processor
(CPUID=068xh), and future 0.13 micron socket 370 processors) requires a 1.25V source. It can be
generated from a voltage divider on the VCC2.5 or VCC3.3 sources using 1% to l erant resistors. A
4.7 µF decoupling capacitor should be included on this input. See Figure 24 and Table 13 for
example CLKREF circuits. Do not use VTT as the source for this reference!
Figure 24. Examples for CLKREF Divider Circuit
Vcc2.5
150
150
PGA370
CLKREF
Y33
Ω
Ω
4.7 µF
Vcc3.3
R1
R2
PGA370
CLKREF
Y33
4.7 µF
sys_bus_CLKREF_divider
Table 13. Resistor Values for CLKREF Divider (3.3V Source)
R1 (Ω)
Ω), 1% R2 (Ω)
Ω)Ω)
182 110 1.243
301 182 1.243
374 221 1.226
499 301 1.242
Ω), 1% CLKREF Voltage (V)
Ω)Ω)
5.7 Undershoot/Overshoot Requirements
Undershoot and overshoot specifications become more critical as the process technology for
microprocessors shrinks due to thinner gate oxide. Violating these undershoot and overshoot limits
will degrade the life expectancy of the processor.
The Pentium III processor (CPUID=068xh), Celeron processor (CPUID=068xh), and future
0.13 micron socket 370 processo rs have more restrictive overshoot and undershoot requirements
for system bus signals than previous processors. These requirements stipulate that a signal at the
output of the driver buffer and at the input of the receiver buffer must not exceed the maximum
absolute overshoot voltage limit or the minimum absolute undershoot voltage limit. Exceeding
either of these limits will dama ge the pro cessor. There is also a time-dependent, non-linear
overshoot and undershoot requirement that depends on the amplitude and duration of the
overshoot/undershoot. See the appropriate processor datasheet for more details on the processor
overshoot/undershoot specifications.
®
815 Chipset Platform Design Guide 57
Intel
System Bus Design Guidelines
5.8 Processor Reset Requirements
Universal PGA370 designs must route the AGTL/AGTL+ reset signal from the chipset to two pins
on the processor as well as to the debug port connector. This reset signal is connected to the
following pins at the PGA370 socket:
• AH4 (RESET#). The reset signal is connected to this pin for the Pentium III processor
• X4 (Reset2# or GND, depending on processor). The X4 pin is RESET2# for Pentium III
processor (CPUID=068xh) and Celeron processor (CPUID=068 xh). X4 is GND for future
0.13 micron socket 370 processors. An additional 1kΩ resistor is connected in series with pin
X4 to the reset circuitry since pin X4 is a ground pin in future 0.13 micron socket 370
processors.
Note: The AGTL/AGTL+ reset signal must always terminate to VTT on the motherboard.
Designs that do not support the debug port will not utilize the 240 Ω series resistor or the
connection of RESET# to the debug port connector. RESET2# is not required for platforms that
do not support the Celeron processor (CPUID=068xh). Pin X4 should then be connected to
ground.
R
The routing rules for the AGTL/AGTL+ reset signal are shown in Figure 25.
Figure 25. RESET#/RESET2# Routing Guidelines
lenITP
Chipset
VTT
Ω
91
cs_rtt_stub
240
Ω
lenCPUlenCS
VTT
10 pF
86
cpu_rtt_stub
22
Table 14. RESET#/RESET2# Routing Guidelines (see Figure 25)
Parameter Minimum (in) Maximum (in)
LenCS 0.5 1.5
LenITP 1 3
LenCPU 0.5 1.5
cs_rtt_stub 0.5 1.5
cpu_rtt_stub 0.5 1.5
ITP
Ω
Daisy c hain
Ω
1 k
Pin X4
Ω
Processor
Pin AH4
sys_bus_reset_routin
®
58 Intel
815 Chipset Platform Design Guide
System Bus Design Guidelines
R
5.9 Processor PLL Filter Recommendations
Intel PGA370 processors have internal phase lock loop (PLL) clock generators that are analog and
require quiet power supplies to minimize jitter.
5.9.1 Topology
The general desired topology for these PLLs is shown in Figure 27. Not shown are the parasitic
routing and local decoupling capacitors. Excluded from the external circuitry are parasitics
associated with each component.
5.9.2 Filter Specification
The function of the filter is to protect the PLL from external noise through low-pass attenuation.
The low-pass specification, with input at VCC
follows:
and output measured across the capacitor, is as
CORE
• < 0.2 dB gain in pass band
• < 0.5 dB attenuation in pass band (see DC drop in next set of requirements)
• > 34 dB attenuation from 1 MHz to 66 MHz
• > 28 dB attenuation from 66 MHz to core frequency
The filter specification is graphically shown in Figure 26.
®
815 Chipset Platform Design Guide 59
Intel
System Bus Design Guidelines
Figure 26. Filter Specification
0.2dB
0dB
-0.5 dB
-28dB
-34dB
Forbidden
Zone
Forbidden
Zone
R
1 MHz66 MHzfcorefpeak1HzDC
passband
high frequency
band
filter_spec
NOTES:
1. Diagram not to scale.
2. No specification for frequencies beyond fcore.
3. fpeak should be less than 0.05 MHz.
Other requirements:
• Use shielded-type inductor to minimize magnetic pickup.
• Filter should support DC current > 30 mA.
• DC voltage drop from VCC to PLL1 should be < 60 mV, which in practice implies series
R < 2 Ω. This also means pass-band (from DC to 1 Hz) attenuation < 0.5 dB for
VCC = 1.1V, and < 0.35 dB for VCC = 1.5V.
®
60 Intel
815 Chipset Platform Design Guide
System Bus Design Guidelines
R
5.9.3 Recommendation for Intel Platforms
The following tables contain examples of components that meet Intel’s recommendations, when
configured in the topology of Figure 27.
1 Ω10% 1/16 W Resistor may be implemented with trace resistance,
in which case a discrete R is not needed. See
Figure 28.
To satisfy damping requirements, total series resistance in the filter (from VCC
of the capacitor) must be at least 0.35 Ω. This resistor can be in the form of a discrete component
or routing or both. For example, if the chosen inductor has a minimum DCR of 0.25 Ω, then a
routing resistance of at least 0.10 Ω is required. Be careful not to exceed the maximum resistance
rule (2 Ω). For example, if using discrete R1 (1 Ω ± 1%), the maximum DCR of the L (trace plus
inductor) should be less than 2.0 - 1.1 = 0.9 Ω; this preclud es the use of some inductors and sets a
maximum trace length.
Current
DCR (Typical)
to the top plate
CORE
Other routing requirements:
• The capacitor (C) should be close to the PLL1 and PLL2 pins, < 0.1 Ω per route. These routes
do not count towards the minimum damping R requirement.
• The PLL2 route should be parallel and next to the PLL1 route (i.e., minimize loop area).
• The inductor (L) should be close to C. Any routing resistance should be inserted between
CORE
and L.
CORE
and L.
VCC
• Any discrete resistor (R) should be inse rted between VCC
®
815 Chipset Platform Design Guide 61
Intel
System Bus Design Guidelines
Figure 27. Example PLL Filter Using a Discrete Resistor
VCC
CORE
R
LR
<0.1 Ω route
Discrete resistor
C
<0.1
Figure 28. Example PLL Filter Using a Buried Resistor
VCC
CORE
LR
Trace resistance
<0.1 Ω route
C
<0.1
PLL1
Processor
PLL2
Ω route
PLL_filter_1
PLL1
Processor
PLL2
Ω route
PLL_filter_2
®
62 Intel
815 Chipset Platform Design Guide
System Bus Design Guidelines
R
5.9.4 Custom Solutions
As long as designers satisfy filter performance and requirements as specified and outlined in
Section 5.9.2, other solutions are acceptable. Custom solutions should be simulated against a
standard reference core model (see Figure 29).
Ω resistor represents small signal PLL resistance.
4. Be sure to include all component and routing parasitics.
5. Sweep across component/parasitic tolerances.
6. To observe IR drop, use DC current of 30 mA and minimum VCC
7. For other modules (interposer, DMM, etc.), adjust routing resistor if desired, but use minimum numbers.
5.10 Voltage Regulation Guidelines
A universal PGA370 design will need the voltage regulation module (VRM) or on-board voltage
regulator (VR) to be compliant with Intel VRM 8.5 DC-DC Converter Design Guidelines.
Ω
sys_bus_core_ref_ mo del
level.
CORE
Processor
5.11 Decoupling Guidelines for Universal PGA370
Designs
These preliminary decoupling guidelines for universal PGA370 designs are estimated to meet the
specifications of VRM 8.5 DC-DC Converter Design Guidelines.
5.11.1 VCC
• Sixteen or more 4.7 µF capacitors in 1206 packages.
All capacitors should be placed within the PGA370 socket cavity and mounted on the primary side
of the motherboard. The capacitors are arranged to minimize the overall inductance between the
VCC
CORE
®
815 Chipset Platform Design Guide 63
Intel
Decoupling Design
CORE
/VSS power pins, as shown in Figure 30.
System Bus Design Guidelines
Figure 30. Capacitor Placement on the Motherboard
R
5.11.2 VTT Decoupling Design
For Itt = 2.3 A (maximum)
• Twenty 0.1 µF capacitors in 0603 packages placed as closed as possible to the processor VTT
pins. The capacitors are shown on the exterior of the previous figure.
5.11.3 VREF Decoupling Design
• Four 0.1 µF capacitors in 0603 package placed near VREF pins (within 500 mils).
®
64 Intel
815 Chipset Platform Design Guide
System Bus Design Guidelines
R
5.12 Thermal Considerations
5.12.1 Heatsink Volumetric Keepout Regions
Current heatsink recommendations are only valid for supported Celeron and Pentium III processor
frequencies.
Figure 31 shows the system component keepout volume above the socket connector required for
the reference design thermal sol ut ion for high frequency processo rs. This keepout envelop e
provides adequate room for the heatsink, fan and attach hardware under static conditions as well as
room for installation of these components on the socket. The heatsink must be compatible with the
Integrated Heat Spreader (IHS) used by higher frequency Pentium III processors.
Figure 32 shows component keepouts on the motherboard required to prevent interference with the
reference design thermal solution. Note portions of the heatsink and attach hardware hang over the
motherboard.
Adhering to these keepout areas will ensure compatibility with Intel boxed processor products and
Intel enabled third party vendor thermal solutions for high frequency processors. While the
keepout requirements should provide adequate space for the reference design thermal solution,
systems integrators should check with their vendors to ensure their specific thermal solutions fit
within their specific system designs. Ensure that the thermal solutions under analysis comprehend
the specific thermal design requirements for higher frequency Pentium III processors.
While thermal solutions for lower frequency processors may not require the full keepout area,
larger thermal solutions will be required for higher frequency processors, and failure to adhere to
the guidelines will result in mechanical interference.
®
815 Chipset Platform Design Guide 65
Intel
System Bus Design Guidelines
Figure 31. Heatsink Volumetric Keepout Regions
R
Figure 32. Motherboard Component Keepout Regions
®
66 Intel
815 Chipset Platform Design Guide
System Bus Design Guidelines
R
5.13 Debug Port Changes
Due to the lower voltage technology employed with newer processors, changes are required to
support the debug port. Previously, test access port (TAP) signals used 2.5V logic, as is the case
with the Intel
Celeron processor (CPUID=068xh), and future 0.13 micron socket 370 processors utilize 1.5V
logic levels on the TAP. As a result, the type of debug port connecter used in universal PGA370
designs is dependent on the processor that is currently in the socket. The 1.5V connector is a
mirror image of the older 2.5V connector. Either connector will fit into the same printed circuit
board layout. Only the pin numbers change (see Figure 33). Also requi red, along with the ne w
connector, is an In-Target Probe* (ITP) that is capable of communicating with the TAP at the
appropriate logic levels.
Figure 33. TAP Connector Compa rison
Celeron processor in the PPGA package. Pentium III processor (CPUID=068xh),
2.5 V connector, AMP 104068-3 vertical plug, top view
24681012141618202224262830
RESET#
RESET#
1357911131517192123252729
1.5 V connector, AMP 104078-4 vertical receptacle, top view
1357911131517192123252729
24681012141618202224262830
sys_bus_TAP_conn
Caution: Pentium III processor (CPUID=068xh) and Celeron processor (CPUID=068xh) require an in-
target probe (ITP) compatible with 1.5V signal levels on the TAP. Previous ITPs were designed to
work with higher voltages and may damage the processor if connected to any of these specified
processors.
See the processor datasheet for more information regarding the debug port.
®
815 Chipset Platform Design Guide 67
Intel
System Bus Design Guidelines
R
This page is intentionally left blank.
®
68 Intel
815 Chipset Platform Design Guide
System Memory Design Guidelines
R
6 System Memory Design
Guidelines
6.1 System Memory Routing Guidelines
Ground plane reference all system memory signals. To provide a good current return path and
limit noise on the system memory signals, the signals should be ground referenced from the
GMCH to the DIMM connectors and from DIMM connector-to-DIMM connector. If ground
referencing is not possible, system memory signals sho uld be, at a minimum, refer enced to a single
plane. If single plane referencing is not possible, stitching capacitors should be added no more
than 200 mils from the signal via field. System memory signals may via to the backside of the PCB
under the GMCH without a stitching capacitor as long as the trace on the topside of the PCB is
less than 200 mils.
Note: Intel recommends that a parallel p late capacitor between VCC3.3SUS and GND be added to
account for the current return path discontinuity (See decoupling section). Use one 0.01 µF X7R
capacitor per every five system memory signals that switch plane references. No more than two
vias are allowed on any system memory signal.
If a group of system memory signals must to change layers, a via field should be created and a
decoupling capacitor should be added at the end of the via field. Do not route signals in the middle
of a via field; this causes noise to be generated on the current return path of these signals and can
lead to issues on these signals (see Figure 34). The traces shown are on layer 1 only. The figure
shows signals that are changing layer and two signals that are not changing layer.
Note: The two signals around the via field create a keepout zone where no signals that do not change
Figure 36. System Memory 2-DIMM Routing Topologies
Topology 1
Topology 2
Topology 3
Topology 4
Topology 5
82815
A
C
D
10
10
Ω
F
Ω
F
E
E
DIMM 0DIMM 1
B
sys_mem_2DIMM_routing_topo
Table 18. System Memory 2-DIMM Solution Space
Signal Top.
A B C D E F
Width Spacing Min. Max. Min. Max. Min. Max. Min. Max. Min. Max. Min. Max.
SCS[3:2]# 3 5 10 1 4.5
SCS[1:0]# 2 5 10 1 4.5
SMAA[7:4] 4 10 10 0.4 0.5 2 4
SMAB[7:4]# 5 10 10 0.4 0.5 2 4
SCKE[3:2] 3 10 10 3 4
SCKE[1:0] 2 10 10 3 4
SMD[63:0] 1 5 10 1.75 4 0.4 0.5
SDQM[7:0] 1 10 10 1.5 3.5 0.4 0.5
SCAS#, SRAS#,
SWE#
SBS[1:0],
SMAA[12:8,3:0]
1 5 10 1 4.0 0.4 0.5
1 5 10 1 4.0 0.4 0.5
Trace Lengths (inches) Trace (mils)
In addition to meeting the spacing requirements outlined in Table 18, system memory signal trace
edges must be at least 30 mils from any other non-system memo ry signal trace edge.
®
815 Chipset Platform Design Guide 71
Intel
System Memory Design Guidelines
Figure 37. System Memory Routing Example
R
sys_mem_routing_ex
NOTE: Routing in this figure is for example purposes only. It does not necessarily represent complete and
Figure 39. System Memory 3-DIMM Routing Topologies
R
82815
Topology 1
Topology 2
Topology 3
Topology 4E
Topology 5
Topology 6
Topology 7
Topology 8E
10
G
10
G
10
G
A
C
D
F
Ω
Ω
Ω
In addition to meeting the spacing requirements outlined in Table 19, system memory signal trace
edges must be at least 30 mils from any other non-system memory signal trace edge.
Table 19. System Memory 3-DIMM Solution Space
Signal
A B C D E F G
Top. Width Spacing Min. Max. Min. Max. Min. Max. Min. Max. Min. Max. Min. Max. Min. Max.
DIMM 0DIMM 1
B
BB
C
D
sys_mem_3DIMM_routing_topo
Trace Lengths (inches) Trace (mils)
DIMM 2
B
SCS[5:4]# 4 5 10 1 4.5
SCS[3:2]# 3 5 10 1 4.5
SCS[1:0]# 2 5 10 1 4.5
SMAA[7:4] 6 10 10 2 4 0.4 0.5
SMAB[7:4]# 7 10 10 2 4 0.4 0.5
SMAC[7:4} 8 10 10 2 4 0.4 0.5
SCKE[5:4] 4 10 10 3 4
SCKE[3:2] 3 10 10 3 4
SCKE[1:0] 2 10 10 3 4
SMD[63:0] 1 5 10 1.75 4 0.4 0.5
SDQM[7:0] 1 10 10 1.5 3.5 0.4 0.5
SCAS#,
SRAS#, SWE#
SBS[1:0],
SMAA[12:8,3:0]
74 Intel
5 5 10 0.4 0.5 1 4
5 5 10 0.4 0.5 1 4
®
815 Chipset Platform Design Guide
System Memory Design Guidelines
R
6.4 System Memory Decoupling Guidelines
A minimu m of eight 0.1 µF low-ESL ceramic capacitors (e.g., 0603 body type, X7R dielectric) are
required and must be as close as possible to the GMCH. They should be placed within at most
70 mils to the edge of the GMCH package edge for VSUS_3.3 decoupling, and they should be
evenly distributed around the system memory interface signal field including the side of the
GMCH where the system memory interface meets the host interface. There are power and GND
balls throughout the system memory ball field of the GMCH that need good local decoupling.
Make sure to use at least 14 mil drilled vias and wide traces from the pads of the capacitor to the
power or ground plane to create a low-inductance path. If possible, multiple vias per capacitor pad
are recommended to further reduce inductance. To add the decoupling capacitors within 70 mils of
the GMCH and/or close to the vias, the trace spacing may be reduced as the traces go around each
capacitor. The narrowing of space between traces should be minimal and for as short a distance as
possible (500 mils maximum).
To further decouple the GMCH and provide a solid current return path for the system memory
interface signals it is recommended that a parallel plate capacitor be added under the GMCH. Add
a topside or bottom side copper flood under center of the GMCH to create a parallel plate
capacitor between VCC3.3 and GND, see following figure. The dashed lines indicate power plane
splits on layer 2 or layer 3 depending on stack-up. The filled region in the mid dle of the GMCH
indicates a ground plate (on layer 1 if the power plane is on layer 2 or on layer 4 if the power layer
is on layer 3).
®
815 Chipset Platform Design Guide 75
Intel
System Memory Design Guidelines
Figure 40. Intel
®
815 Chipset Platform Decoupling Example
R
Yellow lines in Figure 40 show layer-two plane splits. (Printed versions of this document will
show the layer two plane splits in the left-side, bottom, right-side, and upper-right-side quadrants
enclosed in gray lines.) Note that the layer 1 shapes do NOT cross the plane splits. The bottom
shape is a VSS fill over VddSDRAM. The left-side shape is a VSS fill over VddAGP. The larger
upper-right-side shape is a VSS fill over VddCORE.
Additional decoupling capacitors shown in Figure 41 should be added between the DIMM
connectors to provide a current return path for the reference plane discontinuity created by the
DIMM connectors themselves. One 0.01 µF X7R capacitor should be added per every ten
SDRAM signals. Capacitors should be placed between the DIMM connectors and evenly spread
out across the SDRAM interface.
For debug purposes, four or more 0603 capacitor sites should be placed on the backside of the
board, evenly distributed under the Intel 815 chipset platform’s system memory interface signal
field.
®
76 Intel
815 Chipset Platform Design Guide
System Memory Design Guidelines
R
Figure 41. Intel
®
815 Chipset Platform Decoupling Example
6.5 Compensation
A system memory compensation resistor (SRCOMP) is used by the GMCH to adjust the buffer
characteristics to specific board and operating environment characteristics. Refer to the Intel
Chipset Family: 82815 Graphics and Memory Controller Hub (GMCH) for use with the Universal
Socket 370 Datasheet for details on compensation. Tie the SRCOMP pin of the GMCH to
40 Ω 1% or 2% pull-up resistor to 3.3 Vsus (3.3Volt standby) via a 10-mil-wide, 0.5 inch trace
(targeted for a nominal impedance of 40 Ω).
®
815
®
815 Chipset Platform Design Guide 77
Intel
System Memory Design Guidelines
This page is intentionally left blank.
R
®
78 Intel
815 Chipset Platform Design Guide
AGP/Display Cache Design Guidelines
R
7 AGP/Display Cache Design
Guidelines
For the detailed AGP interface functionality (e.g., protocols, rules, signaling mechanisms) refer to
the latest AGP Interface Specification, Revision 2.0, which can be obtained from
http://www.agpforum.org
recommendations and covers both standard add-in card AGP and down AGP solutions.
7.1 AGP Interface
A single AGP connector is supported by the GMCH’s AGP interface. LOCK# and SERR#/PERR#
are not supported. See the display cache discussion for a display cache/AGP muxing description
and a description of the Graphics Performance Accelerator (GPA).
. This design guide focuses only on specific Intel 815 chipset platform
The AGP buffers operate in one of two selectable modes to support the AGP universal connector:
• 3.3V drive, not 5V safe. This mode is compliant with the AGP 1.0 66 MHz specification.
• 1.5V drive, not 3.3V safe. This mode is compliant with the AGP 2.0 specification.
The AGP 4X must operate at 1.5V and only use differential clocking mode. The AGP 2X can
operate at 3.3V or 1.5V. The AGP interface supports up to 4X AGP signaling, though 4X fast
writes are not supported. AGP semantic cycles to DRAM are not snooped on the host bus.
The GMCH supports PIPE# or SBA[7:0] AGP address mechanisms, but not both simultaneously.
Either the PIPE# or the SBA[7:0] mechanism must be selected during system initialization. The
GMCH contains a 32-deep AGP request queue. High-priority accesses are supported. All AGP
semantic accesses hitting the graphics aperture pass through an address translation mechanism
with a fully-associative 20-entry TLB.
Accesses between AGP and the hub interface are limited to hub interface-originated memory
writes to AGP. Cacheable accesses from the IOQ queue flow through one path, while aperture
accesses follow another path. Cacheable AGP (SBA, PIPE#, and FRAME#) reads to DRAM all
snoop the cacheable global write buffer (GWB) for system data coherency. Aperture AGP (SBA,
PIPE#) reads to DRAM snoop the aperture queue (GCMCRWQ). Aperture AGP (FRAME#) reads
and writes to DRAM proceed through a FIFO and there is no RAW capability, so no snoop is
required.
The AGP interface is clocked from the 66 MHz clock (3V66). T he AGP-to-host/memory interface
is synchronous with a clock ratio of 1:1 (66 MHz : 66 MHz), 2:3 (66 MHz : 100 MHz) and
1:2 (66 MHz : 133 MHz).
®
815 Chipset Platform Design Guide 79
Intel
AGP/Display Cache Design Guidelines
7.1.1 Graphics Performance Accelerator (GPA)
The GMCH multiplexes the AGP signal interface with the integrated graphics’ display cache
interface. As a result, for a universal motherboard that supports both integrated graphics and addin AGP video cards, display cache (for integrated graphics) must be populated on a card in the
universal AGP slot. The card is called a Graphics Performance Accelerator (GPA) card. Intel
provides a specification for this card in a separate document (Graphics Performance Accelerator Specification).
AGP guidelines are presented in this section for motherboards that support the population of a
GPA card in their AGP slot, as well as for those that do not. Where there are distinct guidelines
dependent on whether or not a motherboard will support a GPA card, the section detailing routing
guidelines is divided into subsections, as follows:
• If the motherboard supports a GPA card populated in the AGP slot, then the guidelines in the
Flexible Motherboard subsections are to be followed.
• If the motherboard will NOT support a GPA card populated in the AGP slot, then the
guidelines in the AGP-Only Motherboard subsections are to be followed.
R
7.1.2 AGP Universal Retention Mechanism (RM)
Environmental testing and field reports indicate that AGP cards and Graphics Performance
Accelerator (GPA) cards may come unseated during system shipping and handling without proper
retention. To avoid disengaged AGP cards and GPA modules, Intel recommends that AGP-based
platforms use the AGP retention mechanism (RM).
The AGP RM is a mounting bracket that is used to properly locate the card with respect to the
chassis and to assist with card retention. The AGP RM is available in two different handle
orientations: left-handed (see Figure 42) and right-handed. Most system boards accommodate the
left-handed AGP RM. The manufacturing capacity of the left-handed RM currently exceeds the
right-handed capacity, and as a result Intel recommends that customers design their systems to
insure they can use the left-handed version of the AGP RM (see Figure 42). The right-handed
AGP RM is identical to the left-handed AGP RM, except for the position of the actuation handle.
This handle is located on the same end as the primary design, but extends from the opposite side
(mirrored a bout the center axis running parallel to the length of the part) . Figure 43 co ntains
keepout information for the left hand AGP retention mechanism. Use this information to ensure the
motherboard design leaves adequate space to install the retention mechanism.
The AGP interconnect design requires that the AGP card must be retained to the extent that the
card not back out more than 0.99 mm (0.039 in) within the AGP connector. To accomplish this it
is recommended that new cards implement an additional notch feature in the mechanical keying
tab to allow an anchor point on the AGP card for interfacing with an AGP RM. The retention
mechanism’s round peg engages with the AGP or GPA card’s retention tab and prevents the card
from disengaging during dynamic loading. The additional notch feature in the mechanical keying
tab is required for 1.5V AGP cards and is recommended for the new 3.3V AGP cards.
®
80 Intel
815 Chipset Platform Design Guide
AGP/Display Cache Design Guidelines
R
Figure 42. AGP Left-Handed Retention Mechanism
Figure 43. AGP Left-Handed Retention Mechanism Keepout Information
Engineering Change Request number 48 (ECR #48) of the AGP specification details the AGP RM,
which is recommended for all AGP cards. These are approved changes to the Accelerated Graphics Port (AGP) Interface Specification, Revision 2.0. Intel intends to incorporate the AGP
RM changes into later revisions of the AGP Interface Specification. In addition, Intel has defined a
reference design of a mechanical device to utilize the features defined in ECR #48.
®
815 Chipset Platform Design Guide 81
Intel
AGP/Display Cache Design Guidelines
ECR #48 can be viewed on the Intel Web site at:
R
http://developer.intel.com/technology/agp/ecr.htm
More information regarding this component (AGP RM) is available from the following vendors.
Resin Color Supplier Part
Number
AMP P/N 136427-1 136427-2 Black
Foxconn P/N 006-0002-939 006-0001-939
Green Foxconn P/N 009-0004-008 009-0003-008
7.2 AGP 2.0
The AGP Interface Specification, Revision 2.0 enhances the functionality of the original AGP
Interface Specification, Revision 1.0 by allowing 4X data transfers (4 data samples per clock) and
1.5V operation. The 4X operation of the AGP interface provides for “quad-pumping” of the AGP
AD (address/data) and SBA (side-band addressing) buses. That is, data is sampled four times
during each 66 MHz AGP clock, which means that each data cycle is ¼ of a 15 ns (66 MHz)
clock, or 3.75 ns. Note that 3.75 ns is the data cycle time, not the clock cycle time. During 2X
operation, data is sampled twice during a 66 MHz clock cycle, so the data cycle time is 7.5 ns. To
allow for such high-speed data transfers, the 2X mode of AGP operation uses source-synchronous
data strobing. During 4X operation, the AGP interface uses differential source-synchronous
strobing.
“Left Handed” Orientation
(Preferred)
“Right Handed” Orientation
(Alternate)
With data cycle times as small as 3.75 ns and setup/hold times of 1 ns, propagation delay
mismatch is critical. In addition to reducing propagation delay mismatch, it is important to
minimize noise. Noise on the data lines causes the settling time to be long. If the mismatch
between a data line and the associated strobe is too great or if there is noise on the interface,
incorrect data will be sampled. The low-voltage operation on the AGP (1.5V) requires even more
noise immunity. For example, during 1.5V operation, V
is 570 mV. Without proper isolation,
il
max
cross-talk could create signal integrity issues.
®
82 Intel
815 Chipset Platform Design Guide
AGP/Display Cache Design Guidelines
R
7.2.1 AG P I nter face Si gnal Groups
The signals on the AGP interface are broken into three groups: 1X timing domain signals, 2X/4X
timing domain signals, and miscellaneous signals. Each group ha s different ro uting requirements.
In addition, within the 2X/4X timing domain signals, there are three sets of signals. All signals in
the 2X/4X timing domain must meet minimum and maximum trace length requirements as well as
trace width and spacing requirements. However, trace length matching requirements only need to
be met within each set of 2X/4X timing domain signals. The signal groups are listed in Table 20.
• Set #2: AD[31:16], C/BE[3:2]#, AD_STB1, AD_STB1#
• Set #3: SBA[7:0], SB_STB, SB_STB#1
INTB#
Table 21. AGP 2.0 Data/Strobe Associations
Data Associated Strobe in 1X Associated
AD[15:0] and C/BE[1:0]# Strobes are not used in 1X mode.
AD[31:16] and C/BE[3:2]# Strobes are not used in 1X mode.
SBA[7:0] Strobes are not used in 1X mode.
All data is sampled on rising clock
edges.
All data is sampled on rising clock
edges.
All data is sampled on rising clock
edges.
1
1
Strobe in 2X
AD_STB0 AD_STB0,
AD_STB1 AD_STB1,
SB_STB SB_STB,
Associated
Strobes in 4X
AD_STB0#
AD_STB1#
SB_STB#
Throughout thi s section, the te rm data refers to AD[31:0], C/BE[3:0]#, and SBA[7:0]. The term
strobe refers to AD_STB[1:0], AD_STB[1:0]#, SB_STB, and SB_STB#. When the term data is
used, it refers to one of the three sets of data signals, as listed in Table 21. When the term strobe is
used, it refers to one of the strobes as it relates to the data in its associated group.
The routing guidelines for each group of signals (1X timing domain signals, 2X/4 X timing domain signals, and miscellaneous signals) will b e addressed separately.
®
815 Chipset Platform Design Guide 83
Intel
AGP/Display Cache Design Guidelines
7.3 Standard AGP Routing Guidelines
These routing guidelines cover a standard AGP solution. This utilizes an AGP compliant device on
an external add-in card that plugs into a connector on the motherboard.
7.3.1 1X Timing Domain Routing Guidelines
7.3.1.1 Flexible Motherboard Guidelines
• The AGP 1X timing domain signals (Table 20) have a maximum trace length of 4 inches for
motherboards that support a Graphics Performance Accelerator (GPA) card. This maximum
applies to ALL signals listed as 1X timing domain signals in Table 20.
• AGP 1X signals multiplexed with display cached signals (listed in the following table) should
be routed with a 1:3 trace width-to-spacing ratio. All other AGP 1X timing domain signals can
be routed with 5-mil minimum trace separation.
R
• There are no trace length matching requirements for 1X timing domain signals.
The following are multiplexed AGP1X Signals on Flexible Motherboards
• RBF# • FRAME#
• ST[2:0] • IRDY#
• PIPE# • TRDY#
• REQ# • STOP#
• GNT# • DEVSEL#
• PAR
7.3.1.2 AGP-Only Motherboard Guidelines
• The AGP 1X timing domain signals (refer to Table 20) have a maximum trace length of
7.5 inches for motherboards that will not support a Graphics Performance Accelerator (GPA)
card. This maximum applies to ALL signa ls listed as 1X timing domain signals in Table 20.
• All AGP 1X timing domain signals can be routed with 5-mil minimum trace separation.
• There are no trace length matching requirements for 1X timing domain signals.
7.3.2 2X/4X Timing Domain Routing Guidelines
These trace length guidelines apply to ALL signals listed in Table 20 as 2X/4X timing domain
signals. These signals should be routed using 5 mil (60 Ω) traces.
The maximum line length and lengt h mismatch requirements depend on the routing rules used on
the motherboard. These routing rules were created to provide design freedom by making tradeoffs
between signal coupling (trace spacing) and line lengths. The maximum length of the AGP
interface defines which set of routing guidelines must be used. Guidelines for short AGP interfaces
(e.g., < 6 inches) and long AGP interfaces (e.g., > 6 inches and < 7.25 inches) are documented
®
84 Intel
815 Chipset Platform Design Guide
AGP/Display Cache Design Guidelines
R
separately. The maximum length allowed for the AGP interface (on AGP-only motherboards) is
7.25 inches.
7.3.2.1 Flexible Motherboard Guidelines
• For motherboards that support either an AGP card or a GPA card in the AGP slot, the
maximum length of AGP 2X/4X ti ming d omain signals is 4 inche s .
• 1:3 trace width-to-spacing is required for AGP 2X/4X signal traces.
• AGP 2X/4X signals must be matched their associated strobe (as outlined in Table 20), within
±0.5 inch.
For example, if a set of strobe signals (e.g., AD_STB0 and AD_STB0#) is 3.7 inches long, the
data signals associated with those strobe signals (e.g., AD[15:0] and C/BE[2:0]#) can be
3.2 inches to 4 inches long (since there is a 4 inch maximum length). Another strobe set (e.g.,
SB_STB and SB_STB#) could be 3.1 inches long, so that the associated data signals (e.g.,
SBA[7:0]) can be 2.6 inches to 3.6 inches long.
The strobe signals (AD_STB0, AD_STB0#, AD_STB1, AD_STB1#, SB_STB, and SB_STB#)
act as clocks on the source-synchronous AGP interface. Therefore, special care must be taken
when routing these signals. Since each strobe pair is truly a differential pair, the pair should be
routed together (e.g., AD_STB0 and AD_STB0# should be routed next to each other). The two
strobes in a strobe pair should be routed using 5-mil traces with at least 15 mils of space (1:3)
between them. This pair should be separated from the rest of the AGP signals (and all other
signals) by at least 20 mils (1:4). The strobe pair must be length-matche d to less than ±0.1 inch
(i.e., a strobe and its complement must be the same length within 0.1 inch).
Figure 44. AGP 2X/4X Routing Example for Interfaces < 6 Inches and GPA/AGP Solutions
5-mil trace
15 mils
5-mil trace
20 mils
5-mil trace
15 mils
5-mil trace
20 mils
5-mil trace
15 mils
Associated AGP 2X/4X data signal length
2X/4X signal
2X/4X signal
2X/4X signal
2X/4X signal
AGP STB#
AGP STB#
AGP STB
AGP STB
2X/4X signal
2X/4X signal
2X/4X signal
2X/4X signal
STB/STB# length
0.5"0.5"
Min.Max.
AGP_2x-4x_routing
®
815 Chipset Platform Design Guide 85
Intel
AGP/Display Cache Design Guidelines
7.3.2.2 AGP-Only Motherboard Guidelines
For motherboards that will no t support a GPA card populated in the AGP slot, the maximum AGP
2X/4X signal trace length is 7.25 inches. However, there are different guidelines for AGP
interfaces shorter than 6 inches (e.g., all AGP 2X/4X signals are less than 6 inches long) and those
longer than 6 inches but shorter than the 7.25 inches maximum.
AGP Interfaces Shorter Than 6 Inches
These guidelines are for designs that require less than 6 inches between the AGP connector and the
GMCH:
• 1:3 trace width-to-spacing is required for AGP 2X/4X timing domain signal traces.
• AGP 2X/4X signals must be matched with their associated strobe (as outlined in Table 20),
within ±0.5 inch.
For example, if a set of strobe signals (e.g., AD_STB0 and AD_STB0#) is 5.3 inches long, the
data signals associated with those strobe signals (e.g., AD[15:0] and C/BE[2:0]#) can be
4.8 inches to 5.8 inches long. Another strobe set (e.g., SB_STB and SB_STB#) could be
4.2 inches long, and the data signals associated with those strobe signals (e.g., SBA[7:0]) could be
3.7 inches to 4.7 inches long.
R
The strobe signals (AD_STB0, AD_STB0#, AD_STB1, AD_STB1#, SB_STB, and SB_STB#)
act as clocks on the source-synchronous AGP interface. Therefore, special care must be taken
when routing these signals. Because each strobe pair is truly a differential pair, the pair should be
routed together (e.g., AD_STB0 and AD_STB0# should be routed next to each other). The two
strobes in a strobe pair should be routed on 5-mil traces with at least 15 mils of space (1:3)
between them. This pair should be separated from the rest of the AGP signals (and all other
signals) by at least 20 mils (1:4). The strobe pair must be length-matche d to less than ±0.1 inch
(i.e., a strobe and its complement must be the same length, within 0.1 inch).
AGP Interfaces Longer Than 6 Inches
Since longer lines have more cross-talk, they require wider spacing between traces to reduce the
skew. The following guidelines are for designs that require more than 6 inches (but less than the
7.25 inches maximum) between the AGP connector and the GMCH:
• 1:4 trace width-to-spacing is required for AGP 2X/4X timing domain signal traces.
• AGP 2X/4X signals must be matched with their associated strobe (as outlined in Table 20),
within ±0.125 inch.
For example, if a set of strobe signals (e.g., AD_STB0 and AD_STB0#) is 6.5 inches long, the
data signals associated with those strobe signals (e.g., AD[15:0] and C/BE[2:0]#) could be
6.475 inches to 6.625 inches long. Another strobe set (e.g., SB_STB and SB_STB#) could be
6.2 inches long, and the data signals associated with those strobe signals (e.g., SBA[7:0]) could be
6.075 inches to 6.325 inches long.
The strobe signals (AD_STB0, AD_STB0#, AD_STB1, AD_STB1#, SB_STB, and SB_STB#)
act as clocks on the source-synchronous AGP interface. Therefore, special care must be taken
when routing these signals. Because each strobe pair is truly a differential pair, the pair should be
routed together (e.g., AD_STB0 and AD_STB0# should be routed next to each other). The two
strobes in a strobe pair should be routed on 5-mil traces with at least 20 mils of space (1:4)
®
86 Intel
815 Chipset Platform Design Guide
AGP/Display Cache Design Guidelines
R
between them. This pair should be separated from the rest of the AGP signals (and all other
signals) by at least 20 mils (1:4). The strobe pair must be length-matche d to less than ±0.1 inch
(i.e., a strobe and its complement must be the same length, within 0.1 inch).
7.3.3 AGP Routing Guideline Considerations and Summary
This section applies to all AGP signals in any motherboard support configuration (e.g., “flexible”
or “AGP only”), as follows:
• The 2X/4X timing domain signals can be routed with 5-mil spacing, when breaking out of the
GMCH. The routing must widen to the documented requirements, within 0.3 inch of the
GMCH package.
• When matching trace length for the AGP 4X interface, all traces should be matched from the
ball of the GMCH to the pin on the AGP connector. It is not necessary to compensate for the
lengths of the AGP signals on the GMCH package.
• Reduce line length mismatch to ensure added margin. Trace length mismatch for all signals
within a signal group should be as close to zero as possible, to provide timing margin.
• To reduce trace-to-trace coupling (cross-talk), separate the traces as much as possible.
• All signals in a signal group should be routed on the same layer.
• The trace length and trace spacing requirements must not be violated b y any signal.
Table 22. AGP 2.0 Routing Summary
Signal
1X Timing
Domain
2X/4X Timing
Domain Set 1
2X/4X Timing
Domain Set 2
2X/4X Timing
Domain Set 3
2X/4X Timing
Domain Set 1
2X/4X Timing
Domain Set 2
2X/4X Timing
Domain Set 3
Max.
Length
7.5” 4 5 mils No
7.25” 4 20 mils ±0.125” AD_STB0
7.25” 4 20 mils ±0.125” AD_STB1
7.25”4 20 mils ±0.125” SB_STB and
6” 3 15 mils1 ±0.5” AD_STB0
6”3 15 mils1 ±0.5” AD_STB1
6”3 15 mils1 ±0.5” SB_STB and
Trace Spacing
(5-mil Traces)
Length
Mismatch
requirement
Relative to Notes
N/A None
and
AD_STB0#
and
AD_STB1#
SB_STB#
and
AD_STB0#
and
AD_STB1#
SB_STB#
AD_STB0,
AD_STB0# must be
the same length
AD_STB1,
AD_STB1# must be
the same length
SB_STB, SB_STB#
must be the same
length
AD_STB0,
AD_STB0# must be
the same length
AD_STB1,
AD_STB1# must be
the same length
SB_STB, SB_STB#
must be the same
length
NOTES:
1. Each strobe pair must be separated from other signals by at least 20 mils.
2. These guidelines apply to board stack-ups with 15% impedance tolerance.
3. 4 inches is the maximum length for a flexible motherboards.
4. Solution valid for AGP-only motherboards.
®
815 Chipset Platform Design Guide 87
Intel
AGP/Display Cache Design Guidelines
7.3.4 AGP Clock Routing
The maximum total AGP clock skew, between the GMCH and the graphics component, is 1 ns for
all data transfer modes. This 1 ns includes skew and jitter that originates on the motherboard, addin card, and clock synthesizer. Clock skew must be evaluated not only at a single threshold
voltage, but at all points on the clock edge that fall within the switching range. The 1 ns skew
budget is divided such that the motherboard is allotted 0.9 ns of clock skew. (The motherboard
designer shall determine how the 0.9 ns is allocated between the board and the synthesizer.)
7.3.5 AGP Signal Noise Decoupling Guidelines
The following routing guidelines are recommended for an optimal system design. The main focus
of these guidelines is to minimize signal integrity problems on the AGP interface of the GMCH.
The following guidelines are not intended to rep lace thorough system validation for products
based on the Intel 815 chipset platform.
• A minimum of six 0.01 µF capacitors are required and must be as close as possible to the
GMCH. These should be placed within 70 mils of the outer row of balls on the GMCH for
VDDQ decoupling. The closer the placement, the better.
R
• The designer should evenly distribute placement of decoupling capacitors in the AGP
interface signal field.
• It is recommended that the designer use a low-ESL ceramic capacitor (e.g., with a 0603 body-
type X7R dielectric).
• To add the decoupling capacitors within 70 mils of the GMCH and/or close to the vias, the
trace spacing may be reduced as the traces go around each capacitor. The narrowing of space
between traces should be minimal and for as short a distance as possible (1 inch maximum).
• In addition to the minimum decoupling capacitors, the designer should place bypass
capacitors at vias that transition the AGP signal from one reference signal plane to another.
On a typical four layer PCB design, the signals transition from one side of the board to the
other. One extra 0.01 µF capacitor is required per 10 vias. The capacitor should be placed as
close as possible to the center of the via field.
The designer should ensure that the AGP connector is well decoupled, as described in the AGP Design Guide, Revision 1.0, Section 1.5.3.3.
®
88 Intel
815 Chipset Platform Design Guide
AGP/Display Cache Design Guidelines
R
Figure 45. AGP Decoupling Capacitor Placement Example
NOTE: This figure is for example purposes only. It does not necessarily represent complete and correct
routing for this interface.
7.3.6 AGP Routing Ground Reference
It is strongly recommended that, at a minimum, the following critical signals be referenced to
ground from the GMCH to an AGP connector (or to an AGP video controller if implemented as a
“down” solution on an AGP-only motherboard), using a minimum number of vias on each net:
AD_STB0, AD_STB0#, AD_STB1, AD_STB1#, SB_STB, SB_STB#, G_GTRY#, G_IRDY#,
G_GNT#, and ST[2:0].
In addition to the minimum signal set listed previously, it is strongly recommended that half of all
AGP signals be reference to ground, depending on board layout. In a n ideal design, the entire AGP
interface signal field would be referenced to ground. This recommendation is not specific to any
particular PCB stack-up, but should be applied to all designs using the Intel 815 chipset platform
for use with the universal socket 370.
AGP_decoupling_ca p_placement
®
815 Chipset Platform Design Guide 89
Intel
AGP/Display Cache Design Guidelines
7.4 AGP Down Routing Guidelines
These routing guidelines cover an AGP down solution. This allows for an AGP compliant device
to be implemented directly on the motherboard without the need for a connector or add-in card.
7.4.1 1X AGP Down Option Timing Domain Routing Guidelines
Routing guidelines for an AGP device on the motherboard are very similar to those when the
device is implemented with an AGP connector.
• AGP 1X timing domain signals (Table 20) have a maximum trace length of 7.5 inches. This
maximum applies to ALL signals listed as 1 X timing domain signals in Table 20.
• All AGP 1X timing domain signals can be routed with 5-mil minimum trace separation
• There are no trace length matching requirements for 1X timing domain signals
R
7.4.2 2X/4X AGP Down Timing Domain Routing Guidelines
These trace length guidelines apply to ALL signals listed in Table 20 as 2X/4X timing domain
signals. These signals should be routed using 5-mil (60 Ω) traces.
• The maximum AGP 2X/4X signal trace length is 6 inches.
• 1:3 trace width-to-spacing is required for AGP 2X/4X timing domain signal traces.
• AGP 2X/4X signals must be matched with their associated strobe (as outlined in Table 20),
within ± 0.5 inch.
For example, if a set of strobe signals (e.g., AD_STB0 and AD_STB0#) is 5.3 inches long, the
data signals associated with those strobe signals (e.g., AD[15:0] and C/BE[2:0]#) could be
4.8 inches to 5.8 inches long. Another strobe set (e.g., SB_STB and SB_STB#) could be
4.2 inches long, and the data signals associated with those strobe signals (e.g., SBA[7:0]) could be
3.7 inches to 4.7 inches long.
The strobe signals (AD_STB0, AD_STB0#, AD_STB1, AD_STB1#, SB_STB, and SB_STB#)
act as clocks on the source-synchronous AGP interface. Therefore, special care must be taken
when routing these signals. Because each strobe pair is truly a differential pair, the pair should be
routed together (e.g., AD_STB0 and AD_STB0# should be routed next to each other). The two
strobes in a strobe pair should be routed on 5-mil traces with at least 15 mils of space (1:3)
between them. This pair should be separated from the rest of the AGP signals (and all other
signals) by at least 20 mils (1:4). The strobe pair must be length-matche d to less than ±0.2 inch
(i.e., a strobe and its complement must be the same length, within 0.2 inch).
®
90 Intel
815 Chipset Platform Design Guide
AGP/Display Cache Design Guidelines
R
Figure 1. AGP Down 2X/4X Routing Recommendations
AGP Compliant
Graphics Device
Length: 0.5" - 6.0" Width to
spacing: 1:3 Strobe-to-Data
Mismatch: ±0.5"
Signals
Strobes
Length: Dependent on Data
Width : 5 mil
Spacing: 15 mils
Strobe-to-Strobe Mi sm atch: ±0.2"
Intel®
82815
AGP_Down_1x-2x
7.4.3 AGP Routing Guideline Considerations and Summary
This section applies to all AGP signals, as follows:
• The 2X/4X timing domain signals can be routed with 5-mil spacing, when breaking out of the
GMCH. The routing must widen to the documented requirements, within 0.3 of the GMCH
package.
• When matching the trace length for the AGP 4X interface, all traces should be matched from
the ball of the GMCH to the ball on the AGP compliant device. It is not necessary to
compensate for the lengths of the AGP signals on the GMCH package.
• Reduce line length mismatch to ensure added margin. Trace length mismatch for all signals
within a signal group should be as close to zero as possible, to provide timing margin.
• To reduce trace-to-trace coupling (cross-talk), separate the traces as much as possible.
• All signals in a signal group should be routed on the same layer.
• The trace length and trace spacing requirements must not be violated by any signal.
Table 23. AGP 2.0 Routing Summary
Signal
1X Timing
Max.
Length
7.5” 5 mils No
Domain
2X/4X Timing
6” 15 mils
Domain Set 1
2X/4X Timing
6” 15 mils
Domain Set 2
2X/4X Timing
6” 15 mils
Domain Set 3
NOTES:
1. Each strobe pair must be separated from other signals by at least 20 mils.
®
815 Chipset Platform Design Guide 91
Intel
Trace Spacing
(5-mil Traces)
1
1
1
Length
Mismatch
Relative to Notes
N/A None
requirement
±0.5” AD_STB0
and
AD_STB0#
±0.5” AD_STB1
and
AD_STB1#
±0.5” SB_STB
and
SB_STB#
AD_STB0,
AD_STB0# must be
the same length
AD_STB1,
AD_STB1# must be
the same length
SB_STB, SB_STB#
must be the same
length
AGP/Display Cache Design Guidelines
7.4.4 AGP Clock Routing
The maximum total AGP clock skew, between the GMCH and the graphics component, is 1 ns for
all data transfer modes. This 1 ns includes skew and jitter that originates on the motherboard and
clock synthesizer. Clock skew must be evaluated not only at a single threshold voltage, but at all
points on the clock edge that fall within the switching range.
For AGP clock routing guidelines for the Intel 815 chipset platform, refer to Section 11.3.
7.4.5 AGP Signal Noise Decoupling Guidelines
The following routing guidelines are recommended for the optimal system design. The main focus
of these guidelines is to minimize signal integrity problems on the AGP interface of the GMCH.
The following guidelines are not intended to rep l ace thorough system validation for products
based on the Intel 815 chipset platform.
• A minimum of six 0.01 µF capacitors are required and must be as close as possible to the
GMCH. These should be placed within 70 mils of the outer row of balls on the GMCH for
VDDQ decoupling. The closer the placement, the better.
R
• The designer should evenly distribute placement of decoupling capacitors in the AGP
interface signal field.
• It is recommended that the designer use a low-ESL ceramic capacitor, such as with a 0603
body-type X7R dielectric.
• To add the decoupling capacitors within 70 mils of the GMCH and/or close to the vias, the
trace spacing may be reduced as the traces go around each capacitor. The narrowing of space
between traces should be minimal and for as short a distance as possible (1 inch maximum).
• In addition to the minimum decoupling capacitors, the designer should place bypass
capacitors at vias that transition the AGP signal from one reference signal plane to another.
On a typical four-layer PCB design, the signals transition from one side of the board to the
other. One extra 0.01 µF capacitor is required per ten vias. The capacitor should be placed as
close as possible to the center of the via field.
7.4.6 AGP Routing Ground Reference
It is strongly recommended that at least the following critical signals be referenced to ground from
the GMCH to an AGP video controller on an AGP-only motherboard, using a minimum number of
vias on each net: AD_STB0, AD_STB0#, AD_STB1, AD_STB1#, SB_STB, SB_STB#,
G_GTRY#, G_IRDY#, G_GNT#, and ST[2:0].
In addition to the minimum signal set listed previously, it is strongly recommended that half of all
AGP signals be referenced to ground, depending on the board layout . In an ideal design, the
complete AGP interface signal field would be referenced to gr ound. This recommendation is not
specific to any particular PCB stack-up, but should be applied to all designs using the Intel 815
chipset platform.
®
92 Intel
815 Chipset Platform Design Guide
AGP/Display Cache Design Guidelines
R
7.5 AGP 2.0 Power Delivery Guidelines
7.5.1 VDDQ Generation and TYPEDET#
AGP specifies two separate power planes: VCC and VDDQ. VCC is the core power for the
graphics controller. This voltage i s always 3.3V. VDDQ is the interface voltage. In AGP 1.0
implementations, VDDQ was also 3.3V. For the designer developing an AGP 1.0 motherboard,
there is no distinction between VCC and VDDQ, as both are tied to the 3.3V power plane on the
motherboard.
AGP 2.0 requires that these power planes be separate. In conjunction with the 4X data rate, the
AGP 2.0 Interface Specification provides for low-voltage (1.5V) operation. The AGP 2.0
specification implements a TYPEDET# (type detect) signal on the AGP connector that determines
the operating voltage of the AGP 2.0 interface (VDDQ). The motherboard must provide either
1.5V or 3.3V to the add-in card, depending on the state of the TYPEDET# signal. (see Table 24).
1.5V low-voltage operation applies only to the AGP interface (VDDQ). VCC is always 3.3V.
Note: The motherboard provides 3.3V to the VCC pins of the AGP connector. If the graphics controller
needs a lower voltage, then the add-in card must regulate the 3.3VCC voltage to the controller’s
requirements. The graphics controller may only power AGP I/O buffers with the VDDQ power
pins.
The TYPEDET# signal indicates whether the AGP 2.0 interface operates at 1.5V or 3.3V. If
TYPEDET# is floating (i.e., No Connect) on an AGP add-in card, the interface is 3.3V. If
TYPEDET# is shorted to ground, the interface is 1.5V.
Table 24. TYPDET#/VDDQ Relationship
TYPEDET#
(on add-in card)
GND 1.5V
N/C 3.3V
As a result of this requirement, the motherboard must provide a flexible voltage regulator or key
the slot to preclude add-in cards with voltage requirements incompatible with the motherboard.
This regulator must supply the appropriate voltage to the VDDQ pins on the AGP connector. For
specific design recommendations, refer to the schematics in Appendix A. VDDQ generation and
AGP VREF generation must be considered together. Before developing VDDQ generation
circuitry, refer to both the above requirements and the AGP 2.0 Interface Specification.
VDDQ
(supplied by MB)
®
815 Chipset Platform Design Guide 93
Intel
AGP/Display Cache Design Guidelines
Figure 46. AGP VDDQ Generation Example Circuit
R
TYPEDET#
+3.3V
+12V
U1
LT1575
1
SHDN IPOS
2
VIN INEG
3
2.2 k
Ω
R1
1 µF
C1
GND GATE
4
FB COMP
.001 µF
C5
7.5 k
8
7
6
5
10 pF
R5
Ω
- 1%
C2
47 µF
5
R2
C4
301 - 1%
Ω
- 1%
1.21 k
C3
R3
R4
VDDQ
220 µF
AGP_VDDQ_gen_ex_circ
The previous figure demonstrates one way to design the VDDQ voltage regulator. This regulator
is a linear regulator with an external, low-Rds
FET. The source of the FET is connected to 3.3V.
on
This regulator converts 3.3V to 1.5V or passes 3.3V, depending on the state of TYPEDET#. If a
linear regulator is used, it must draw power from 3.3V (not 5V) to control thermals
(i.e., 5V regulated down to 1.5V with a linear regulator will dissipate approximately 7 W at 2 A).
Because it must draw power from 3.3V and, in some situations, must simply pass that 3.3V to
VDDQ (when a 3.3V add-in card is placed in the system), the regulator MUST use a low-Rds
on
FET.
AGP 1.0 ECR #44 modified VDDQ 3.3
3.3 V
to a FET with an Rds
is 3.168V. Therefore, 68 mV of drop is allowed across the FET at 2 A. This corresponds
min
of 34 mΩ.
on
to 3.1V. When an ATX power supply is used, the
min
How does the regulator switch? The feedback resistor divider is set to 1.5V. When a 1.5V card
is placed in the system, the transistor is Off and the regulator regulates to 1.5V. When a 3.3V card
is placed in the system, the transistor is On, and the feedback will be pulled to ground. When this
happens, the regulator will drive the gate of the FET to nearly 12V. This will turn the FET on and
pass 3.3V – (2 A * Rds
94 Intel
) to VDDQ.
on
®
815 Chipset Platform Design Guide
AGP/Display Cache Design Guidelines
R
7.5.2 VREF Generation for AGP 2.0 (2X and 4X)
VREF generation for AGP 2.0 is different, depending on the AGP card type used. The 3.3V AGP
cards generate VREF locally. That is, they have a resistor divider on the card that divides VDDQ
down to VREF
the GMCH and graphics controller, 1.5V cards use source-generated VREF. That is, the VREF
signal is generated at the graphi cs controlle r and sent to the GMCH, and another VREF is
generated at the GMCH and sent to the graphics controller (see Figure 47).
Both the grap hics controller and the GMCH must generate VREF and distribute it through the
connector (1.5V add-in cards only). The following two pins defined on the AGP 2.0 universal
connector allow this VREF passing:
• VREFGC : VREF from the graphics controller to the chipset
• VREFCG : VREF from the chipset to the graphics controller
To preserve the common mode relationship between the VREF and data signals, the routing of the
two VREF signals must be matched in length to the strobe lines, within 0.5 inch on the
motherboard and within 0.25 inch on the add-in card.
(see Figure 47). To account for potential differences between VDDQ and GND at
The voltage divider networks consist of AC and DC ele ments, as shown in Figure 47.
The VREF divider network should be placed as close as practical to the AGP interface, to get the
benefit of the common-mode power supply effects. However, the trace spacing around the VREF
signals must be a minimum of 25 mils to reduce cross-talk and maintain signal integrity.
During 3.3V AGP 2.0 operation, VREF must be 0.4 VDDQ. However, during 1.5V AGP 2.0
operation, VREF must be 0.5 VDDQ. This requires a flexible voltage divider for VREF. Various
methods of accomplishing this exist, and one such example is shown in Figure 47.
®
815 Chipset Platform Design Guide 95
Intel
AGP/Display Cache Design Guidelines
d
Figure 47. AGP 2.0 VREF Generation and Distribution
a) 1.5V AGP Car
+12V
R7
(See note 2)
Ω
1.5V AGP
Card
1 K
TYPEDET#
VrefGC
300
200
R9
R2
R
1%
1%
VDDQ
C8
500 pF
Ω
Ω
U6
VDDQ
AGP
REF
Device
GND
VrefCG
Notes:
1. The resistor dividers should be placed near the GMCH. The AGPREF signal must be 5 mils wide and routed 10 mils from adjacent signals.
2. R7 is the same resistor seen in AGP VDDQ generation example circuit figure (R1)
mosfet
C9
0.1 uF
REF
VDDQ
GMCH
GND
1 K
1 K
R6
Ω
R2
Ω
b) 3.3V AGP Card
+12V
(See note 2)
3.3V AGP
Card
VDDQ
AGP
Device
GND
REF
1 K
VrefGC
R7
Ω
TYPEDET#
U6
mosfet
R9
300
1%
R2
200
1%
C10
0.1 uF
Ω
Ω
REF
VDDQ
GMCH
GND
VDDQ
1 K
R2
R6
1 K
Ω
500 pF
Ω
R4
82
Ω
C9
82
82
500 pF
500 pF
R5
Ω
82
R5
Ω
R4
Ω
C9
C8
VrefCG
The resistor dividers should be placed near the GMCH. The AGPREF signal must be 5 mils wide and routed 25 mils from adjacent signals.
agp_2.0ref_gen_dist
The flexible VREF divider shown in Figure 47 uses a FET switch to switch be tween the locally
generated VREF (for 3.3V add-in cards) and the source-generated VREF (for 1.5V add-in cards).
Use of the source-generated VREF at the receiver is optional and is a product implementation
issue beyond the scope of this document.
®
96 Intel
815 Chipset Platform Design Guide
AGP/Display Cache Design Guidelines
R
7.6 Additional AGP Design Guidelines
7.6.1 Compensation
The GMCH AGP interface supports resistive buffer compensation (RCOMP). Tie the GRCOMP
pin to a 40 Ω, 2% (or 39 Ω, 1%) pull-down resistor (to ground) via a 10-mil-wide, very short
(<0.5 inch) trace.
7.6.2 AGP Pull-Ups
AGP control signals require pull-up resistors to VDDQ on the motherboard, to ensure that they
contain stable values when no agent is actively driving the bus.
Note: It is critical that these signals be pulled up to VDDQ, not 3.3V.
The trace stub to the pull-up resistor on 1X timing domain signals should be kept at less than
0.5 inch, to avoid signal reflections from the stub.
Note: The strobe signals require pull-ups/pull-downs on the motherboard to ensure that they contain
stable values when no agent is driving the bus.
Note: INTA# and INTB# should be pulled to 3.3V, not VDDQ.
2X/4X Timing Domain Signals
• AD_STB[1:0] (pull-up to VDDQ)
• SB_STB (pull-up to VDDQ)
• AD_STB[1:0]# (pull-down to ground)
• SB_STB# (pull-down to ground)
The trace stub to the pull-up/pull-down resistor on 2X/4X timing domain signals should be kept to
less than 0.1 inch to avoid signal reflections from the stub.
®
815 Chipset Platform Design Guide 97
Intel
AGP/Display Cache Design Guidelines
The pull-up/pull-down resistor value requirements are Rmin = 4 kΩ and Rmax = 16 kΩ. The
recommended AGP pull-up/pull-down resistor value is 8.2 kΩ
7.6.2.1 AGP Signal Voltage Tolerance List
The following signals on the AGP interface are 3.3V tolerant during 1.5V operation:
• PME#
• INTA#
• INTB#
• GPERR#
• GSERR#
• CLK
• RST
The following signals on the AGP interface are 5V tolerant (refer to the USB specification):
• USB+
• USB-
• OVRCNT#
R
.
The following special AGP signal is either GROUNDED or NOT CONNECTED on an AGP card.
• TYPEDET#
Note: All other signals on the AGP interface are in the VDDQ group. They are not 3.3V tolerant during
1.5V operation.
7.7 Motherboard / Add-in Card Interoperability
There are three AGP connectors: 3.3V AGP connector, 1.5V AGP connector, and Universal AGP
connector. To maximize add-in flexibility, it is highly advisable to implement the universal
connector in a system based on the Intel 815 chipset platform. All add-in cards are either 3.3V or
1.5V cards. The 4X transfers at 3.3V are not allowed due to timings.
As described earlier, the AGP and display cache interfaces of the Intel 815 chipset platform are
multiplexed or shared. In other words, the same component pins (balls) are used for both
interfaces, although obviously only one interface can be supported at any given time. As a result,
almost all display cache interface signals are mapped onto the new AGP interface. The Intel 815
chipset platform can be configured in either AGP mode or Graphics mode. In the AGP mode, the
interface supports a full AGP 4X interface. In the Graphics mode, the interface becomes a display
cache interface similar to the Intel
display cache is optional. There do not have to be any SDRAM devices connected to the interface.
The only dedicated display cache signals are OCLK and RCLK, which need not connect directly
to the SDRAM devices. These are not mapped onto existing AGP signals.
7.8.1 GPA Card Considerations
To support the fullest flexibility, the display cache exists on an add-in card (Graphics Performance
Accelerator, or GPA) that complies with the AGP connector form factor. If the motherboard
designer follows the flexible routing guidelines for the AGP interface detailed in previous sections,
the customer can choose to populate the AGP slot in a system based on the Intel 815 chipset
platform with either an AGP graphics card, with a GPA card to enable the highest-possible internal
graphics performance, or with nothing to get the lowest-cost internal graphics solution. Some of
the GPA/ Intel 815 chipset platform for use with the universal socket 370 interfacing implications
are listed below. For a complete description of the GPA card design, refer to the Graphics Performance Accelerator Card Specification available from Intel.
810E chipset. Note, however, that in the Graphics mode, the
• A strap is required to determine which frequency to select for display cache operation. This is
the L_FSEL pin of the GMCH. The GPA card will pull this signal up or down, as appropriate
to communicate to the appropriate operating frequency to the Intel 815 chipset platform. The
platform will sample this pin on the deasserting edge of reset.
• Since current SDRAM technology is always 3.3V rather than the 1.5V option also supported
by AGP, the GPA card should set the TYPEDET# signal correctly to indicate that it requires a
3.3V power supply. Furthermore, the GPA card should have only the 3.3V key and not the
1.5V key, thereby preventing it from being inserted into a 1.5V-only connector.
• The pad buffers on the chip will be the normal AGP buffers and will wo rk for both interfaces.
• In internal graphics mode, the AGPREF signal, which is required for the AGP mo de, should
remain functional as a reference voltage for sampling 3.3V LMD inputs. The voltage level on
AGPREF should remain exactly the same as in the AGP mode, as opposed to VCC/2 used fo r
previous products.
7.8.1.1 AGP and GPA Mechanical Considerations
The GPA card will be designed with a notch on the PCB to go around the AGP universal retention
mechanism. To guarantee that the GPA card will meet all shock and vibration requirements of the
system, the AGP universal retention mechanism will be req uired on all AGP sockets that are to
support a GPA card.
®
815 Chipset Platform Design Guide 99
Intel
AGP/Display Cache Design Guidelines
7.8.2 Display Cache Clocking
The display cache is clocked source-synchronously from a clock generated by the GMCH. The
display cache clocking scheme uses three clock signals.
• LTCLK clocks the SDRAM devices, is muxed with an AGP signal, and should be routed
according to the flexible AGP guidelines.
• LOCLK and LRCLK clock the input buffers of the universal platform. LOCLK is an output
of the GMCH and is a buffered copy of LTCLK. LOCLK should be connected to LRCLK at
the GMCH, with a length of PCB trace to create the appropriate clock skew relationship
between the clock input (LRCLK) and the SDRAMcapacitorclock input(s).
The guidelines are illustrated in Figure 48.
Figure 48. Display Cache Input Clocking
GMCH
LOCL
0.5"
15
Ω 1%
R
LRCL
15 pF 5%
1.5"
disp_cache_in_clk
The capacitor should be placed as close as possible to the GMCH LRCLK pin. To minimize skew
variation, Intel recommends a 1% series termination resistor and a 5% NP0 (also known as C0G)
capacitor, to stabilize the value across temperatures.In addition to the 15 Ω, 1% resistor and the
15 pF, 5% NP0 capacitor. The following combination also can be used: 10 Ω, 1% and 22 pF, 5%
NP0.
7.9 Designs That Do Not Use The AGP Port
Universal platform designs that do not use the AGP port should terminate the AGP pins of the
GMCH. Except for the GPAR pin (that requires a 100 kΩ pull-down resistor to ground), the pullup or pull-down resistor value should be 8.2 kΩ. Any external graphics implementation not using
the AGP port should terminate the GMCH AGP control and strobe signals as recommend ed in
Section 13.3.2.
®
100 Intel
815 Chipset Platform Design Guide
Loading...
+ hidden pages
You need points to download manuals.
1 point = 1 manual.
You can buy points or you can get point for every manual you upload.