Compatibility with CUDA 11

[alejandrojcastaneira]

Hello, I am trying to train over a cuda 11.0 gpu card, I noticed then that cuda 11 its only compatible with dlg>=0.6. Are there any workaround to train with dgl>=0.6.0 or I would love to hear some insights on how to adapt myself the code to train with the newest dlg version?

[lzh66]

I faced the same question. I want to know do you solve this problem? Thank U

[logan-markewich]

Cuda 11 now works with dlg (see image from their github)

There will be a few tracebacks in the code when you run, but you just have to use graph.to() to move the graph to the GPU

[alejandrojcastaneira]

Hello everyone, any update on this? I am able to install dlg 0.7 with cuda 11.2, however, when I run the training code, it throws some errors.

[LawsonAbs]

I have solved this problem. The machine environment is configured as follows： dgl 0.7.2 (notice that you should install on GPU version.) CUDA 11.3 torch 1.10.0+cu113 at the same time, you should modify the source code to make the graph and parameter to cuda.

[LawsonAbs]

The runtime status is as follows：

[lrongzheni]

@LawsonAbs Can you paste the difference of code for CUDA 11? Thanks.

[LawsonAbs]

maybe you could get answer from my repository https://github.com/LawsonAbs/GAIN, it can be executable in above-mentioned configuration.