[visarga]

> the first open-source HPC/Hyperscale-class platform for GPU computing that’s also programming-language independent

Weird website, took me a while to classify its purpose. So, is this like an AMD version of NVIDIA's CuDNN? Can we run TensorFlow on AMD GPUs?

[redtuesday]

Looks like the title was changed. Yesterday it was called "AMD deep learning library ROCm 1.6 released (Caffee support and more)"

It seems at the moment only Caffe is public and it's a port of the CUDA version to OpenCL [0][1] to allow it to run on AMD GPUs like the recently released AMD Vega Frontier Edition. From the wording on the page it looks like it could be an equivalent to CuDNN.

[0] through AMDs tool called HIP which can convert CUDA

[1] http://gpgpu10.athoura.com/ROCM_GPGPU_Keynote.pdf