I am looking to develop FPGA-accelerated CNN implementations and am considering intel Arria 10 chips as candidates for development. However, I wanted to figure out if intel has CNN (convolutional neural network) IP within in their IP library that I can integrate in my HDL design. I already know that Xilinx has such core (Xilinx DPU) but have not yet confirmed whether intel has something equivalent.
I did come across documentation on Intel DLA but I am not quite sure if those are FPGA IP Cores (equivalent to DPU cores from Xilinx) or a complete software suite that is configurable via sotware (more like Xilinx Vitis). Can someone with knowledge of this please confirm?
OpenVINO has option for selecting different standard neural network architectures,you can also use model optimizer to choose custom designs.
The DLA on FPGA is configurable through the OpenVINO tool kit .
The following Video is a good introduction.
In case you want to try out your designs on acceleration FPGA Devcloud can be helpful.
Thanks and Regards