Intel NCSM2485.DK User Manual

Product Brief

Intel® Movidius™ Myriad™ X VPU

Enhanced Visual Intelligence at the Network Edge

Intel® Movidius™ Myriad™ X VPU with Neural Compute Engine

Take your imaging, computer vision and machine intelligence applications into network edge devices with the newest Movidius family of vision processing units (VPUs) by Intel.

Industry Leading Performance at Ultra-Low Power

Intel’s Myriad X third generation VPU delivers class-leading performance in computer vision and deep neural network inferencing applications. As the newest member of the Movidius VPU family known for ultra-low power consumption, the Myriad X VPU is capable of delivering a total performance of over 4 trillion operations per second (TOPS).

X VPU is a power ecient solution that brings advanced vision and articial

intelligence applications to devices such as drones, smart cameras, smart home, security, VR/AR headsets, and 360 cameras.

With new performance enhancements, the Myriad

New Generation of Deep Neural Network Performance

Intel has introduced an entirely new deep neural network processing unit into

the Myriad X VPU architecture: the Neural Compute Engine. Specically designed

to run deep neural networks at high speed and low power, the Neural Compute Engine enables the Myriad X VPU to reach over 1 TOPS of compute performance on deep neural network inferences.

part of the power ecient Movidius VPU architecture which minimizes power by

reducing data movement on-chip. While the Myriad 2 VPU has provided superior deep neural network support at low power, the Myriad X VPU can now reach 10X higher performance for applications requiring multiple neural networks running simultaneously.²

The Neural Compute Engine is integrated as

Customizable Imaging & Vision Pipelines

The Movidius family of VPUs have always provided a unique, exible

architecture for image processing, computer vision, and deep neural networks.

The architecture provides a modular approach to conguring imaging and

vision workloads because it combines a set of imaging and vision hardware accelerators, such as stereo depth or the Neural Compute Engine, with an array of C-programmable VLIW vector processors, all accessing a common on-chip memory. This approach enables world-class image signal processing (ISP)

without the need to make trips to memory for best power eciency, in addition

to interleaved computer vision and deep neural network inference application

pipelines, all with a dataow methodology that reduces power by minimizing data

movement. Movidius VPUs deliver an optimal balance between programmability and performance at low power.

Support for 8 HD Sensors and 4K Encoding

The Myriad X VPU features 16 MIPI lanes, which supports up to 8 HD resolution RGB sensors to be connected directly. The high-throughput inline ISP ensures streams are processed at high speeds, while new hardware encoders provide support

for 4K resolutions at both 30 Hz (H.264/H.265) and 60 Hz (M/

JPEG) frame rates. Other featured interfaces include USB 3.1 and PCI-E Gen 3.

deep learning applications on Myriad X VPU. The SDK

also includes a specialized FLIC framework with a plug-in

approach to developing application pipelines including image processing, computer vision, and deep learning. This framework helps developers focus on the processing, leaving

dataow optimization to the tools. For deep neural network

development, the SDK includes a neural network compiler that enables developers to rapidly port neural networks from

common frameworks such as Cae* and Tensorow* with an automated conversion and optimization tool that maximizes

Software Development Kit (SDK) and Tools

performance while retaining network model accuracy.

The Myriad X VPU ships with a rich SDK that contains all of the software development frameworks, tools, drivers and libraries to implement custom imaging, vision and

Movidius® Myriad™ X VPU at a Glance

FE ATURES BENEFITS

Neural Compute Engine With this dedicated on-chip accelerator for deep neural networks, the Myriad X VPU delivers

over 1 trillion operations per second of DNN inferencing performance. networks in real time at the edge without compromising on power consumption or accuracy.

16 Programmable 128-bit VLIW Vector Processors Run multiple concurrent imaging and vision application pipelines with the exibility of 16

vector processors optimized for computer vision workloads.

16 Congurable MIPI Lanes Connect up to 8 HD resolution RGB cameras direc tly to the Myriad X VPU with suppor t for up

to 700 million pixels per second of image signal processing throughput.

Enhanced Vision Accelerators Utilize over 20 hardware accelerators to perform tasks such as optical ow and stereo

depth without introducing additional compute overhead. For example, the new stereo depth

accelerator can simultaneously process 6 camera inputs (3 stereo pairs) each running 720p

resolution at 60 Hz frame rate.

2.5 MB of Homogenous On-Chip Memor y The centralized on-chip memory architecture allows for up to 400 GB/sec of internal

bandwidth, minimizing latency and reducing power consumption by minimizing o-chip data

transfer.

2 chip packages oered MA2085: No memory in-package; interfaces to external memory

MA2485: 4 Gbit LPDDR4 memory in-package

Where to Get More Information

For more information, visit www.movidius.com/MyriadX

Run deep neural

Over all per formance is the ar chitectur al calculati on base d on maxi mum per formance of ope rations- per-se cond over al l availa ble compu te units. Application per formance var ies bas ed

on the application.

Maximum per formance ba sed on peak oatin g-point comp utati onal thr oughp ut of Neur al Compute Engine . Actual result s on deep neu ral net works may achi eve less than peak thr oughp ut.

Intel technolo gies’ feature s and bene ts depend on sy stem congur ation and may requ ire enabled hardware, soft ware or serv ice activation. Per formance va ries depend ing on sys tem congura tion. Intel, Movidiu s, and Myriad ar e trade mark s of Intel Corporat ion or its subsidi aries in the U.S . and/or oth er countr ies. * Other nam es and brands may be claime d as the prop erty of other s.

Please Recycle

333867-0 01-001US