Intel® Vision Accelerator Design
Overview
Deploy power-efficient deep neural network inference for fast, accurate video analytics and computer vision applications.
- High throughput and efficiency
- Scalable, optimized performance for cost- and power-efficient AI at the network edge
- Works with Intel® Distribution of OpenVINO™ toolkit
- Run analytics on edge servers, appliances, and network video recorders (NVR).
- Full use of Intel® architecture across CPUs, CPUs with integrated graphics, FPGAs, and VPUs
Hardware
Accelerate with Intel® Movidius™ Vision Processing Unit (VPU)
Get high-performance machine vision at ultra-low power from specialized processors.
Accelerate with Intel® Arria® 10 FPGA
Adapt a flexible, customizable processor to advanced display, video, and image processing workloads.
Who Needs This Product
System integrators and independent software vendors (ISV) working on the following types of solutions:
- Media processing from camera to cloud
- Broad inferencing acceleration portfolios
- Development and deployment of high-performance vision applications
Use Cases
- Industrial: Detect defects quickly and accurately
- Smart cities: Identify congestion and redirect traffic
- Healthcare: Diagnose illnesses with real-time feedback
- Digital security and surveillance: Identify behavior anomalies and property demographics
- Retail: Interpret shopper behavior and improve inventory tracking
Acceleration Design Options
| With Intel® Movidius™ VPU | With Intel® Arria® 10 FPGA | |
|---|---|---|
Differentiation |
Specialized processors High-performance machine vision Ultra-low power |
Flexible, customizable processors Advanced video and image processing workloads Larger compute-intensive networks |
Form Factor |
M.2, M-PCIe* PCIe* half height, half length, single slot (passive cooling) |
PCIe half height, half length, two slot (active cooling) PCIe half height, half length, single slot (passive cooling) |
Host Interface |
M.2 (USB) M-PCIe (USB) 4 - PCIe Gen2 |
8 - PCIe Gen3 |
| Power | Up to 25 W |
60 W maximum, 40 W typical |
| Latency | NA | Lowest (most networks less than 10 ms) |
| Precision | FP16 | FP16, FP11 |
| Thermal | 0°C - 55°C (passive and active cooling options) |
0°C - 65°C (active cooling) 0°C - 55°C (passive cooling, built-in cooling system) 0°C - 45°C (passive cooling, no cooling system) |
Operating Profile |
NA | Continuous operation 8 years active cooling, 10 years passive cooling |
Customization |
Hardware optimized for generic use cases Scale with multiple VPUs |
Custom network support Custom FPGA functions |