[SOLVED] Titan V on PyTorch 0.3.0, CUDA 9.0, CUDNN 7.0 is much slower than 1080 Ti

[smth]

@Royi

By the way, what if I use PyTorch as “GPU Accelerated NumPy”, should I use this as well?

Doesn’t make sense then because cudnn only has neural-net functions, and benchmark only applies to torch.nn.Conv*d