3 ms·
I haven't seen CuDNN equivalents (in terms of perf) for common machine learning frameworks from AMD. I don't think they exist, if they did, people would shift t
by dharma1 10y ago
I haven't seen CuDNN equivalents (in terms of perf) for common machine learning frameworks from AMD. I don't think they exist, if they did, people would shift to using AMD.
For NVidia I have seen some faster kernels than the ones supplied by NVidia - https://github.com/NervanaSystems/neon https://github.com/NervanaSystems/neon - though CuDNN introduced Winograd kernels too in their last update
- dragandj 10y agoI do not know about the quality of this since I do not use NNs, but there is https://github.com/hughperkins/DeepCL https://github.com/hughperkins/DeepCL and I think I have seen others.