The Omnitek Deep Learning Processing Unit (DPU) is optimised for Intel’s Arria 10 GX architecture and delivers 135 GOPS/W at full 32-bit floating point accuracy when running the VGG-16 CNN in an Arria 10 GX 1150. The DPU design uses a new mathematical framework combining low-precision fixed point maths with floating point maths to achieve very high compute density with no accuracy loss.
Scalable across a wide range of Arria 10 GX and Stratix 10 GX devices, the DPU can be tuned for low cost or high performance. The DPU is fully 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 as well as custom models. No FPGA design expertise is required to do this.
FPGAs are ideally suited to Machine Learning applications due to their massively parallel DSP architecture, distributed memory and ability to reconfigure the logic and connectivity for different algorithms. Omnitek’s DPU can be configured to provide optimal compute performance for CNNs, RNNs, MLPs and other neural network topologies existing today, or ones that still have to be developed.