3 ms·

One should hope that the non-CUDA GPU compute library ecosystem has already advanced beyond being able to calculate FFTs!

• singhrac 6y ago
  Sure, but if Nvidia/OpenAI/Google/Facebook have shown anything, it's that there's always more kernels to invent and train bigger nets with.
• andi999 6y ago
  Last time I checked there was no good fft for AMD.