1 comment

[ 3.2 ms ] story [ 14.5 ms ] thread
The TensorFlow version using DirectML was 4x slower than CUDA on the same machine. Is this comparable?