Is Win10+CUDA10.2+Anaconda3.7 OK

1

There is a laptop with RTX2060 using Windows10.
CUDA Driver is 10.2
I want to config the PyTorch with GPU using Anaconda3.7
Must I install CUDA Toolkit 10.1 in order to match Pytorch? If so, is the new CUDA driver needed, downloaded to 10.1?

2

If you install the binaries, you would just need the CUDA driver.
I’m not sure, how to properly check it on Windows, but you don’t need any CUDA compilers.

3

Is there a schedule to support CUDA 10.2 recently?

4

%E5%B1%8F%E5%B9%95%E6%88%AA%E5%9B%BE(4)
%E5%B1%8F%E5%B9%95%E6%88%AA%E5%9B%BE(4)696×237 11.3 KB

OS: Win10
As showing in the screenshot, there is NVCUDA.DLL
Does this mean I can just install PyTorch using the following command line:
conda install pytorch torchvision cudatoolkit=10.1 -c pytorch ?

All the setup about CUDA, cuDNN, setc will be completedautomatically?

5

I’m not deeply familiar with Windows, but I would guess so.
Could you give it a try and let us know, if you see any issues?

6

That’s OK!
The environment of PyTorch with GPU on Windows10 has been configured successfully.
I will make a brief summarization of the process and post it here soon.

1 Like
7

is pytorch1.3 support cuda10.2 ? official biniary only support cuda10.1…

8

A few days ago, I installed PyTorch with GPU support on Win10 OS.
The following is the key steps:

  1. Select the Anaconda version, I select a Python3.7 for Windows x64
    Attention please: do not use the non-ANSI name for Windows!!!
  2. In order to avoid unexpected errors, please select the official support version.
    Use the conda to install pytorch with the official command line.
    All the required packages will be installed automatically including the CUDA toolkit.

Now, the environment of PyTorch with a RTX GPU has been installed on Win10 successfully.
Use the following command to check the setting:

import torch
torch.version
torch.cuda.is_available()