The PyTorch binaries include the CUDA and cuDNN libraries.05 Oct 2020
In the days of yore, one had to go through this agonizing process of installing the NVIDIA (GPU) drivers, cuda, cuDNN libraries, and PyTorch.
Ok, those days are somewhat over. If you are using the PyTorch binaries, they come with cuda and cuDNN built in. I have not found great documentation for this, only a thread in the PyTorch discussion section from two years ago.
If you build from source, yes, you will need to install those libraries yourself.
Ok, how to verfiy? Here goes, assuming you also have FastAI installed (lazy at the moment):
import torch from fastai.vision import * from fastai.metrics import error_rate print("Is cuda available?", torch.cuda.is_available()) print("Is cuDNN version:", torch.backends.cudnn.version()) print("cuDNN enabled? ", torch.backends.cudnn.enabled) x = torch.rand(5, 3) print(x)
Here’s the result –
[email protected]:~/testing$ python3 test2.py Is cuda available? True Is cuDNN version: 7603 cuDNN enabled? True tensor([[0.7559, 0.9504, 0.9759], [0.7765, 0.6080, 0.1925], [0.7885, 0.9641, 0.9562], [0.4040, 0.7394, 0.5701], [0.4912, 0.2765, 0.4441]])