Is possible make ANN, especially deep learning in FPGA to parallelize net? I want get data from PC, send to FPGA (for exmaple MAXimator), compute it and resend results to PC.
Andrzej B. wrote: > Is possible make ANN, especially deep learning in FPGA to parallelize net? For sure. But you must be able to transfer your algorithms into the FPGA in a way that they can be computed in parallel. And theres the crux: usually you have so many data that you cannot do that computing in parallel. So you must break it up it in serial calculations. And there every cheap graphic controller beats a softcore implemented inside a FPGA due to its broad and fast memory interface.
I would use Cuda. Much cheaper than anything else with FPGAs. The problem is the parallel programming. Your algorithm needs to be parallelized. If your problem can not be parallelized you won't ahieve any improvements ...
Xilinx has some deep learning examples for their new fpga series?