FPGA claims world-leading CNN performance
Optimised for the Intel Arria 10 GX architecture, the Omnitek deep learning processing unit achieves 135Goperations per second per Watt at full 32-bit floating point accuracy when running the VGG-16 convolutional neural network (CNN) in an Arria 10 GX 1150. This, says Omnitek, is a world-leading performance for a mid-range SoC FPGA.
Omnitek has designed the deep learning processing unit around a mathematical framework combining low-precision fixed point maths with floating point maths to achieve the high compute density with zero loss of accuracy.
Scalable across a wide range of Arria 10 GX and Stratix 10 GX devices, the processing unit can be tuned for low cost or high performance in either embedded or data centre applications.
It is software programmable in C/C++ or Python using standard frameworks such as TensorFlow, enabling it to be configured for a range of standard CNN models including GoogLeNet, ResNet-50 and VGG-16 and also for custom models. No FPGA design expertise is required to do this, adds Omnitek.
Omnitek’s deep learning processing unit can be configured to provide optimal compute performance for CNNs, RNNs, MLPs and other neural network topologies and for as-yet unknown algorithms and optimisation techniques.
Omnitek was formed in 1998 to design intelligent video and vision systems based on programmable FPGAs and SoCs. Its technology enables customsed vision and artificial intelligence (AI) inferencing capabilities on FPGAs for customers across a range of end markets. Omnitek’s IP addresses demanding application requirements in areas such as video conferencing, projection and display and medical vision systems.
Today, it was announced that Intel had acquired Omnitek. “Omnitek’s technology is a great complement to our FPGA business,” said Dan McNamara, Intel senior vice president and general manager of the Programmable Solutions Group. “Their deep, system-level FPGA expertise and high performance video and vision related technology have made them a trusted partner for many of our most important customers. Together, we will deliver leading FPGA solutions for video, vision and AI inferencing applications on Intel FPGAs and speed time-to-market for our existing customers while winning new ones.”
