[CUDA] How to manually install binaries

1

Heyho,
I would like to use my (not anymore supported) GPUs ‘Tesla K10.G1.8GB’ to train my model with Pytorch 0.4.0. The GPUs have a computing capability of 3.0. Does it work with Pytorch 0.4.0 when I compile the binaries found at https://pytorch.org/previous-versions/? Or do I need to downgrade Pytorch?

Thanks!

Raph

2

I’m not sure if it with the pytorch 0.4 binary out of the box. If it doesn’t, then it might be easier to install from source: https://github.com/pytorch/pytorch.

1 Like
3

Gonna try this. When I need help (which I probably, because I never installed anything from source :smiley:) I will come back at you. Thanks!

4

@richard so like promised: I am trying to install the thing from scratch in a conda enviroment: And i get following error:

cmake: /usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.20' not found (required by cmake)
cmake: /usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `CXXABI_1.3.9' not found (required by cmake)
cmake: /usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.21' not found (required by cmake)
cmake: /usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.22' not found (required by cmake)
Failed to run 'bash tools/build_pytorch_libs.sh --with-nnpack --with-mkldnn caffe2 nanopb libshm gloo THD'

cmake and gcc are both installed :confused:

5

What is your gcc version?

6

What GCC version is recommended?

7

I have GCC 4.9 in Ubuntu 14.04 with cuda 9.0.

GCC version is dependent on the Cuda toolkit version as well. You can go with the above combination mostly. Atleast I know it works.

8

I have GCC 7.3 with no CUDA.

I get GCC exit status 4