4 ms·
Right now the only thing I can see holding AMD back is their lack of vision to drop a bit of cash on a CuDNN equivalent (and the ML framework support for it), e
by dharma1 10y ago
Right now the only thing I can see holding AMD back is their lack of vision to drop a bit of cash on a CuDNN equivalent (and the ML framework support for it), especially fast convolution kernels optimised for their hardware. They are waiting for the open source community to do the work for them, but no one is picking it up because it's a lot of work.
A couple of good AMD engineers should be able to knock out fast Winograd kernels pretty quick.
- ansible 10y agoTheir best bet it to increase support for OpenCL rather than invent something new. And then improve / create OpenCL support for TensorFlow and other popular toolkits.
- dharma1 10y agoYes, that would be OpenCL/SPIR-V. But you still need to put in the work to optimise convolution kernels for specific hardware. TF has openCL support in the works, AFAIK outsourced to Codeplay - https://www.codeplay.com/portal/tensorflow%E2%84%A2-for-opencl%E2%84%A2-using-sycl%E2%84%A2 https://www.codeplay.com/portal/tensorflow%E2%84%A2-for-open... Caffe has an OpenCL branch, works on AMD but about quarter of the speed of equivalent NVidia hardware using CuDNN 4/5, mostly due to unoptimised kernels - https://github.com/BVLC/caffe/tree/opencl https://github.com/BVLC/caffe/tree/opencl