Is there any plan of making an OpenCL accelerated version for non-CUDA users?

[CoryKornowicz]

The CUDA implementation is amazing, but for systems that don't have Nvidia GPU's are there any plans of making an (expectably much slower) more general implementation?

[dkoes]

To do that well would be a massive undertaking. The shortcut approach is to replace the CPU's matrix operations with OpenCL accelerated versions which in theory might require minimal effort. I plan at some point to do this to take advantage of Apple's Neural Engine and can look into doing it for OpenCL as well, but this isn't a high priority. Contributions are always welcome.