3 ms·

Why not OpenCL? With AVX2, it's quite a performer even on Xeons.

• hobofan 10y ago
  It's still lacking as highly optimized libraries like cuDNN last time I looked :/