AltiVec Parallel
Vector Technology
Scalable from Two
to Hundreds of
Processors
1K CFFT in 20 µs
on Each Processor
125 MHz Memory
System with
Prefetch and ECC
Advanced DMA
Engine for Chained
Submatrix Moves
267 MB/s RACE++
Switch Fabric
Interconnect
Fast L2 Cache
(250 MHz)
Embedded computing
reaches a new level of
performance with
RACE++
®
Series
PowerPC
®
7410 daugh-
tercards from Mercury
Computer Systems.
Each PowerPC 7410
daughtercard contains two
500 MHz MPC7410 microprocessors
with AltiVec™ technology. These unique
microprocessors combine a modern superscalar RISC architecture with an AltiVec
parallel vector execution unit.
The AltiVec vector processing unit
revolutionizes the performance of
computationally intensive applications such
as image and signal processing. Each
vector unit can operate in parallel on up to
four floating-point numbers or up to
sixteen 8-bit integers. This dramatically
accelerates vector arithmetic and provides
greater application performance on smaller,
less power-hungry processors.
AltiVec technology also represents a leap
in simplifying the programming required
to achieve high performance. Whereas
previous DSP-based systems required
handcrafted assembly language code for
optimal performance, easy-to-use extensions to the C language provide a direct
mapping to AltiVec instructions. This
permits developers to program more
productively in a higher-level language,
even for critical sections of code.
Optimized Performance
With the huge increase in processing
performance brought by AltiVec, most
applications are no longer CPU-limited.
To keep the processor fed with ample data,
increased emphasis is placed on the
memory system and communications fabric
that delivers data to the processor. Each
compute node on the 500 MHz PowerPC
7410 daughtercard has a dedicated fabric
interface at 267 MB/s and maximum memory speed of 125 MHz. By maximizing the
performance of the memory and the fabric
interface to the processor, Mercury has
optimized RACE++ compute nodes for
processing continuous streams of data.
AltiVec in RACE++ Computers
The computational power of RACE++
Series systems is built from compute nodes
comprised of processors, memory, and
interfaces to the RACE++ interconnect.
PowerPC 7410 daughtercards each contain
two compute nodes.
Each compute node (CN) consists of an
MPC7410 microprocessor with AltiVec
technology, level 2 (L2) cache, synchronous
DRAM (SDRAM), and a Mercury-designed
ASIC. This CN ASIC contains architectural
advancements that enhance concurrency
between arithmetic and I/O operations.
500 MHz PowerPC
7410 Daughtercard
PowerPC 7410 Daughtercard Architecture
Mercury can configure systems with hundreds of compute
nodes, communicating over the second-generation RACE++
switch fabric interconnect. Merging RACE++ and AltiVec
technology provides embedded computers with unprecedented computational power.
AltiVec Vector Processing Unit
The AltiVec vector processing unit operates on 128 bits of
data concurrently with the other PowerPC execution units.
AltiVec instructions may be interleaved with other PowerPC
instructions without any penalty such as a context switch. The
128-bit wide execution unit can be used to operate on four
floating-point numbers, four 32-bit integers, eight 16-bit
integers, or sixteen 8-bit integers simultaneously.
AltiVec instructions are carried out by one of two AltiVec
sub-units. The Vector arithmetic logic unit handles the
vector fixed-point and vector floating-point operations. Two
floating-point operations are possible in a single cycle with the
vector multiply-add instruction and the vector negative
multiply-subtract instruction.
The Permute sub-unit incorporates a crossbar network to
perform 16 individual byte moves in a single cycle. This
capability can be used for simple tasks such as converting the
"endian-ness" of data or for more complicated tasks such as
byte interleaving, dynamic address alignment, or accelerating
small look-up tables.
PowerPC RISC Architecture
In addition to the AltiVec execution unit, the MPC7410
contains a floating-point unit and two integer units that can
operate concurrently with the AltiVec unit. Data and instructions are fed through two on-chip, 32-Kbyte, eight-way
set-associative caches that enhance performance of both
vector and scalar code.
Each PowerPC 7410 CN also includes a fully pipelined
backside L2 cache operating at 250 MHz. This high-
performance cache system provides quick access to data
previously loaded from memory but too large to fit into the
on-chip cache.
Compute Node ASIC
The CN ASIC, included in each compute node, acts as both
a memory controller and as a network interface to the
RACE++ switch fabric interconnect. The CN ASIC includes
an enhanced DMA controller, a high-performance memory
system with error checking and correcting, metering logic,
and a RACE++ interface. By combining memory control
and network interface into a single chip, Mercury's compute
node provides the highest performance with the lowest power
consumption and highest reliability.
High-Performance Memory System
Mercury's high-performance memory subsystem allows the
memory to reach the intrinsic limits of its performance
capability with:
125-MHz Synchronous DRAM
Prefetch Buffers: bring sequential data to the ASIC ahead
of their explicit requests by the processor. These prefetch
buffers greatly improve the performance of the CN in vector operations such as those used in DSP applications.
FIFO Buffers: efficiently overlap accesses to SDRAM from
the local processor and the RACEway interconnect.
The PowerPC CN contains error-correcting circuitry for
improved data integrity. One-bit errors are corrected on the
fly, and multi-bit errors generate an interrupt error condition.
Enhanced DMA Controller
Each CN has an advanced DMA controller to support
RACEway transfers at 267 MB/s with chaining and striding.
MPC7410 Data
and Instruction Flow
Compute Node ASIC Architecture
With chained DMA, the DMA controller works from a
linked list in memory so that a complex chain of DMA
requests requires no overhead from remote
processors. DMA requests also support non-sequential
access to and from local memory. This strided DMA
capability enables high-performance submatrix transfers as
required for distributed 2-D applications in image
processing and synthetic aperture radar (SAR).
Performance Metering Logic
The CN ASIC also provides non-intrusive hardware
performance metering facilities that enable application
performance tuning and real-time data-flow analysis,
essential for developing real-time multicomputer
applications. A RACE++ node can be directed to collect
performance metrics on memory bandwidth utilization,
RACEway crossbar port usage and contention, or memory
contention. The information can be used to locate suboptimal code sections during execution of the application.
RACE++ Switch Fabric Interconnect
The PowerPC 7410 daughtercard is fully compatible with
the RACE++, the second generation of the RACE architecture. The PowerPC CN ASIC connects the processor and
memory directly to the 66.66-MHz RACEway interconnect.
As the latest version of the RACE architecture, the
RACE++ interconnect represents an architectural evolution
that includes increased communication speed, more richly
connected topologies, and augmented adaptive routing.
Together these enhancements represent a performance
revolution yielding significantly higher bisection bandwidth
and lower latency suitable for the most challenging
real-time problems.
Crossbar-Connected Communication
All communication paths among compute nodes, memory
nodes, and I/O nodes are connected using eight-port crossbars, both on-board and between boards. By providing these
point-to-point connections, the RACE architecture eliminates
the bottlenecks found in systems that use buses to connect
processors together on a board or between boards.
Each node on a RACE++ daughtercard is connected to the
RACEway switch fabric through an interface ASIC on the
daughtercard. The RACE++ VME motherboard contains
the first set of crossbars in the RACEway switch fabric. The
second set of crossbars is contained on RACEway Interlink
modules that connect multiple motherboards. This architecture assures each compute node direct and unblocked access
to the RACEway network.
High Bandwidth, Low Latency
Each connection through a RACE++ crossbar can run at
66.66 MHz for a peak bandwidth of 267 MB/s. Each crossbar can connect four simultaneous communication paths for a
total peak bandwidth over 1 GB/s.
Low latency is often as important, if not more important,
than high bandwidth. When making a connection through a
RACE++ system at 66.66 MHz, each crossbar along the data
transfer path adds only 75 ns to the latency. Once the connection is established, each crossbar adds only 15 ns of latency.
Priority-Based Communication
Real-time applications demand more than speed alone. The
priority-based communication of RACEway is designed to
minimize the worst case latency for selected communications.
Adaptive Routing
The RACE++ crossbars adaptively route data using
alternate path selections to actively avoid any congestion in
the network without user intervention.
PowerPC 7410 Daughtercard
RACE++ is a registered trademark, and PAS, RACE Series and the RACE logo, and TATL are trademarks of Mercury Computer
Systems, Inc. AltiVec is a trademark of Motorola, Inc. PowerPC is a registered trademark of IBM Corp. Other products mentioned
may be trademarks or registered trademarks of their respective holders. Mercury Computer Systems, Inc. believes this information
is accurate as of its publication date and is not responsible for any inadvertent errors. The information contained herein is subject to
change without notice.
Copyright © 2003 Mercury Computer Systems, Inc. DS-6C-11
RACEway Broadcast Support
The PowerPC compute node supports RACEway crossbar
broadcast capabilities for enhanced bandwidth utilization
when sending the same data to multiple nodes. This
hardware broadcast also reduces latency compared to
software-based schemes that emulate a hardware broadcast by
repetitively sending the same data in separate transfers.
As the initiator of a broadcast, either the PowerPC
processor or the DMA controller may use any of the broadcast or multicast modes provided by the RACEway crossbar.
As a receiver of broadcast data, the PowerPC CN has the
programmable option of accepting or ignoring the data.
Software
Mercury also provides programming tools that simplify
development and testing of real-time applications. The
RACE++ Series MULTI
®
Integrated Development
Environment (IDE) enables programmers to use familiar,
mainstream tools to develop real-time processing routines.
The MULTI IDE's integrated debug, program build, profiling,
and version control facilities streamline development of
complex multiprocessor software.
Mercury's Scientific Algorithm Library (SAL) provides the
industry's easiest method of accessing the full power of the
AltiVec vector unit. The SAL contains hundreds of image and
signal processing functions that are optimized for AltiVec execution. These high-performance algorithms can also be
accessed through the industry-standard VSIPL interface using
Mercury's VSIPL-Lite signal processing library. For applications that require custom algorithms with demanding vector
processing requirements, Mercury's C compiler includes the
AltiVec C extensions that allow direct access to the AltiVec
vector unit.
For large-scale programming, the Parallel Acceleration System
(PAS™) supports data reorganization techniques for a wide
variety of efficient and scalable programming models, by
means of a high-level set of library calls. Scaling is achieved
by changing runtime variables rather than recompiling
the core code.
The TATL™ Trace Analysis Tool and Library from Mercury
is also available to assist during software development and
debugging. TATL is a minimally intrusive software logic
analyzer that allows developers to visualize multiprocessor
interactions. Using TATL, developers can spot bottlenecks,
such as contention for shared resources or load imbalances.
Electrical/Mechanical Specifications
P2J256J-ROC
RACEway ports 2
Processor frequency 500 MHz
Compute nodes 2
Memory frequency 125 MHz
SDRAM per CN 256 Mbytes
SDRAM per daughtercard 512 Mbytes
L2 cache frequency 250 MHz
L2 cache per CN 2 Mbytes
Weight 0.43 pounds*
Dimensions 5.0" x 4.435"
Power consumption** 20.8 W
Daughtercards per MCJ6*** 2
Daughtercards per MCJ9 8
* Rugged version weighs an additional 0.01 pounds
** Maximum typical power consumption measured with concurrent FFTs and I/O
*** Requires 5-row connectors on MCJ6 and VME backplane
Commercial Environmental Specifications
Operating
temperature 0°C to 40°C up to an altitude of 10,000 feet
(inlet air temperature at motherboard's recommended minimum airflow)
Storage temperature -40°C to 85°C
Relative humidity 10 to 90% (non-condensing)
As altitude increases, air density decreases, hence the cooling effect of a particular
CFM rating decreases. Many manufacturers specify altitude and temperature ranges
that are not simultaneous. Notice that the above operating temperature is specified
simultaneously with an altitude. Different limits can be achieved by trading among
altitude, temperature, performance, and airflow.
Contact Mercury for more information.
Nihon Mercury Computer Systems K.K.
No. 2 Gotanda Fujikoshi Bldg. 4F
5-23-1 Higashi Gotanda
Shinagawa-ku, Tokyo 141-0022
JAPAN
+81 3 3473 0140
Fax +81 3 3473 0141
Mercury Computer Systems, SARL
Immeuble le Montreal
19 bis, avenue du Quebec
Villebon
91951 COURTABOEUF CEDEX
FRANCE
+ 33 (0) 1 69 59 94 00
Fax + 33 (0) 69 86 10 875
Mercury Computer Systems, Ltd.
Campbell Court, Unit 19
Bramley, Tadley
HANTS RG26 5EG
UNITED KINGDOM
+ 44 1 256 880090
Fax + 44 1 25688 4004
For more information, go to www.mc.com
Loading...