High-performance accelerator engine focused around Inference at the Edge applications

07-02-2020 | Concurrent | Subs & Systems

Concurrent Technologies now offer their first, high-performance 3U VPX accelerator engine based on an Intel Altera FPGA. TR AEx/3sd-RCx is focused around Inference at the Edge applications such as real-time object recognition and behaviour monitoring. The device has been created to work in parallel with the company processor boards like TR H4x/3sd-RCx and TR J4x/6sd-RCx that are in alignment with a proposed SOSA Technical Standard.

The application of AI techniques to produce Inference at the Edge has generated a great deal of interest in consumer-orientated applications such as autonomous driving. This new AI Accelerator Engine, permits a user to receive and process real-time actionable intelligence from RF, vision or other sensors, giving a faster solution to numerous problems faced at the edge. The inferencing hardware used on the device is supported by Intel’s OpenVINO Toolkit, a broad toolkit for deploying inferencing solutions on Intel-based products. As well as Deep Learning, OpenVINO supports OpenCV and OpenVX: an open-source library for computer vision applications and an API for heterogeneous compute use across Intel devices respectively.

The device supports popular AI frameworks including Caffe, TensorFlow and MXNet along with neural network models like AlexNet and Resnet for ease of use and compatibility. The distribution is provided with approximately 25 pre-trained models but also has a model downloader function for those clients who have their own data model.

Jane Annear, managing director of Concurrent Technologies, commented: “This new AI Accelerator Engine marks an exciting step for Concurrent Technologies, demonstrating our ability to create more proactive solutions for problems and demands for challenging markets. Our objective is to provide solutions based around our existing processor board products and this is a significant step forward.”