Intel Vision Accelerator Design with Arria 10 FPGA
The MUSTANG-F100-A10 is an Intel Vision Accelerator Design with Intel Arria 10 FPGA, offers exceptional performance, flexibility, and scalability for deep-learning and computer-vision solutions—from NVRs (network video recorders) to edge deep-learning inference appliances to on-premises servers—at a fraction of the cost and with significantly lower power requirements than most of the existing FPGA PCIe cards currently on the market.
The programmable, software-defined Intel Arria 10 FPGA ensures continual performance optimization—taking advantage of periodic FPGA bitstream updates provided by Intel—without necessitating hardware upgrades. The MUSTANG-F100-A10 is designed for long product life (15 years of longevity) and can adapt to a wide range of work conditions, including harsh industrial and/or outdoor environments.
Unlike fixed-function devices, functionality in the Intel Arria 10 FPGA can always be changed or modified to increase or deepen intelligence, which allows the MUSTANG-F100-A10 to be architected to solve very specific problems. And when used for deep learning inference, the MUSTANG-F100-A10 is able to achieve high-performance images per second at reduced power while providing dynamic flexibility, consistent power consumption, and future-proofing for custom or new workloads as well as low latency across a wide spectrum of vision use cases and applications.
The Intel Vision Accelerator Design with Intel Arria 10 FPGA offers a simplified path for developers to run customized topologies in an optimal way and works seamlessly with the OpenVINO toolkit. Fine-grained parallelism enables high throughput on low-batch workloads. The extremely high, fine-grained, on-chip memory bandwidth can more efficiently solve memory challenges. With the OpenVINO toolkit, you don’t need to be an FPGA expert to code applications integrating computer vision.
The MUSTANG-F100-A10 can support more than 20 channels of video inputs, along with vision use cases such as facial detection and recognition. Optimized network topologies include GoogLeNet v1, ResNet-18/50/101, SqeezeNet v1.1, VGG-16, and MobileNet v1. More hardware accelerated topologies are coming with new OpenVINO toolkit releases. Customizable data paths and precision create energy-efficient dataflow for system level optimization. The MUSTANG-F100-A10 is also ideal for complex or large workloads as well as customized applications and use cases where you may want to add your own primitives or sub-layer configurations.
Because of their innate adaptability, FPGAs are always on the cutting edge of new solutions and changing network topologies. This flexibility, and the ability to accelerate algorithm processing, make this Intel Vision Accelerator Design invaluable for complex, massive deep-learning analysis and visual intelligence.
- Intel Arria 10 GX1150 FPGA and 8GB of DDR4
- Low-profile card with PCI Express Gen3 x8 interface
- Supports Ubuntu 16.04.3 LTS 64-bit, CentOS 7.4 64-bit, and Windows 10 64-bit
- Supported natively by the Intel OpenVINO toolkit
- Intel FPGAs supports multiple float-points and inference workloads
- Supports popular deep learning frameworks such as Caffe, TensorFlow, Apache MXNet, and Open Neural Network Exchange (ONNX)
- Easily deploy open source deep learning frameworks via Intel Deep Learning Deployment Toolkit.
- Achieve optimized solutions through reprogramming flexibility
- Deliver high-performance, on-device deep learning inferences at low power and low latency
- Rapidly process and analyze vast streams of high-quality video data near the edge and respond in real time
- Deploy compact form factor systems that are suitable at the edge for demanding imaging and video applications
PCIe FPGA Highest Performance Accelerator Card with Arria 10 1150GX support DDR4 2400Hz 8GB, PCIe Gen3 x8 interface
USB FPGA download cable