Intel® Vision Accelerator Design with Intel® Arria® 10 FPGA
Overview
A flexible, customizable processor that adapts to advanced display, video, and image processing workloads.
- Low-latency for real-time inference
- Ideal for compute-intensive networks in network video recorder (NVR), gateway, and edge servers
- Long-life continual operation in industrial temperatures
Flexible Hardware
FPGA compute performance and features evolve over time and can be optimized (or tailored) to meet unique system requirements.
Distributed, Fine-Grain Digital Signal Processor (DSP) Blocks, Memory, and Logic
Deliver higher on-chip bandwidth compared to a general purpose GPU with deterministic low latency and power efficiency.
Core and I/O Programmability Enables Flexibility
Versatile I/O configurations and multifunction OpenCL™ kernels accelerate deep learning applications.
Virtual Testing with Intel® DevCloud for the Edge
Quickly prototype and develop AI applications in the cloud using the latest Intel® hardware and software tools.
Who Needs This Product
Information and operational technologists who:
- Are new to IoT commercial platforms and need a simple path without a steep learning curve
- Create solutions that offload deep learning and AI workloads from the CPU or GPU to dedicated accelerator products
- Need a quicker path to deployment
Use Cases
- Smart cities
- Automotive and transportation
- Healthcare
- Retail
- Digital Security
Specifications
Interfaces to Host
PCIe*
Configuration
For edge servers including NVRs, gateways, and edge video analytics servers
Supported Streams
Aggregation of 3 to 32 video streams per device
Batch Size
1-144
Power Consumption
Board capable of 60 W maximum, 40 W typical operation
Efficiency
Good efficiency versus general purpose GPU
Memory
Larger memory footprint networks (more than 250 MBs) and compute-intensive networks (more than 3 GMACs)
30-40 generic size networks can be preloaded
Precision
Supports lower precision networks (such as FP16 and FP11)
Customization
Customizable hardware architecture for optimization and to support unique topologies
Lifespan
Eight years with active fan and more than 10 years with passive thermal solution
Operating System Support
Ubuntu* 16.04.3, kernel 4.14.0
Ubuntu 16.04.3, kernel 4.15.0
Host System List
IEI Tank* 870 (Intel® Core™ i7 or i5 processors)
Host CPU Configurations
Intel® Core™ i7 processor 6700TE & Intel® Q170 chipset
Intel Core i5 processor 6500TE & Intel Q170 chipset
Intel® Xeon® E3-1275 processor version 5
Intel Xeon E5-1650 processor version 4
Hardware
IoT Developer Kits Supporting Intel Arria 10 FPGAs
Software
Intel® Distribution of OpenVINO™ Toolkit
- Enable deep learning inference on the edge based on convolutional neural networks
- Support for heterogeneous execution across various accelerators—CPU, GPU, Intel® Movidius™ Myriad™ X, and FPGA—using a common API
- Speed up time to market via a library of functions and preoptimized kernels
- Preinstalled models included with release 5