Open SiDinhThoi opened 5 years ago
Good, question... I don't know. All I know is that doing Deep Learning on Windows is most of the times a very painful thing to do... but you can use the code also without NCCL. You just need to exchange the MultiProcessParallelUpdater
with a different Updater like the StandardUpdater
but you won't be able to harness the full Multi-GPU power then...
@Bartzi Thank you for your response!
Hi @Bartzi , could you tell me how to check CUDA and CUDNN version for windows 10? I search Google and find no tutorials for checking CUDA version for windows 10. For CUDNN version, I found following 2 ways:
And this is output from my computer:
But my computer does not have any folder "NVIDIA GPU Computing Toolkit" in "Program Files"
Did you install CUDA and cuDNN?
Hi @Bartzi , I have experiences with Pytorch Framework and the boolean value of torch.cuda.is_available() is True so I think my computer already installed CUDA and cuDNN
Well, then it should be somewhere... maybe you need to dig deeper :wink:, because I can not tell you anything else than what you already know. You should be able to find the location where you installed CUDA by just browsing your hard drive
I think my computer automatically installed CUDA and CUDNN when I put 1070 ti to it (the reason why I don't know where CUDA located) so do you have any idea where CUDA and CUDNN should be located ?
somewhere in your program_files
directory??
Hi everyone and @Bartzi , I install the project directly on my PC. Could I install NCCL (>2.0) on window 10? If it is, where I can find step by step NCCL installation on window 10?