radix-ai/scipy-notebook-gpu

code can be updated to 10.1

jan-perl opened this issue · 4 comments

I have made some updates, and have this running for the current version.
Shall I push my changes ?

Hi Jan, sure, feel free to launch a PR and we'll review it!

PR doesn't seem to work in my browser.
Data is in my fork.
What I did

  1. make it cuda version independent
  2. re-base compatibility to 10.1
  3. added the pin file, as I ended up having incompatible nvidia drivers otherwise.

This version should say in sync with current updates.
The 10.1 compatibility version could still be changed to a symbolic name.

Hi Jan, thanks for the proposal! Are the TensorFlow binaries compatible with 10.1 yet? According to the docs [1], it seems that they still require 10.0.

[1] https://www.tensorflow.org/install/gpu#ubuntu_1804_cuda_10

Well, at least libcudnn7 exists for 10.1, and it installs
http://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64/
I have not tested tensorflow code.

I also added cudatoolkit to the dockerfile, as it is required for numba