search

chochain / tensorForth

Forth does tensors, in CUDA.
MIT License
27 stars 2 forks source link

How hard would it be to port this to OpenCL? #4

Open znmeb opened 2 months ago

znmeb commented 2 months ago

The reason I ask is twofold:

  1. There are a lot more ways one can use OpenCL than CUDA. NVIDIA GPUs will run OpenCL. Intel CPUs and FPGAs will run OpenCL. Using pocl just about every CPU will run OpenCL.
  2. One of the accelerators on the BeagleY-AI single-board computer runs OpenCL. I have one, and this little $80 card gets 50 32-bit GFLOPS in clpeak!
chochain commented 2 months ago

Ed: apologize getting back to you late. Was traveling/cooling in Canada.

  1. I haven't done any OpenCL but it looks promising by Intel's SYCL converter. I'll do some analysis before giving you an answer.
  2. Beagle Y-AI's MMA + Arm looks very interesting!