Setting up CUDA for RM

pblack476pblack476 Member Posts: 83 Maven
So I am on PopOS 19.10 and I have CUDA 10.1, which is not compatible with RM ATM. Can I have CUDA 9 installed in parallel? I have actually ran conda install cudatoolkit=9.0 but it made no difference for RM and I still cannot configure GPU use on the settings.

How does it work for others around here? Do I have to be on 18.04? Do I have to roll back drivers? Is there any way to get a 19.10 system running RM on GPU?

I really appreciate the help.

Best Answer

  • Options
    pblack476pblack476 Member Posts: 83 Maven
    edited November 2019 Solution Accepted
    @pschlunder I have tried your reccomendation but RM still throws an error when selecting GPU in 'Deep Learning Backend' setting.

    I have installed cuda 9 by:
    conda install cudatoolkit=9.0
    This installed CUDA 9.0 on my /home/[username]/.anaconda3/lib directory.

    I then proceeded to
    /home/[username]/.anaconda3/lib' >> ~/.bashrc</code>export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/[username]/.anaconda3/lib<br><code>echo 'export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:
    And running 'echo $LD_LIBRARY_PATH' I get the correct location as output.

    However RM still throws the error that I must have a free liscence or CUDA 9.0 might not be installed. (I have an Educational Liscence).

    EDIT: This actually broke my linux installation and I had to Timeshift from Live USB to get it working again. I could run the system but console stopped launching and I could not reboot. After a cold restart, system would not start. I would not attempt it anywhere else.


  • Options
    sgenzersgenzer Administrator, Moderator, Employee, RapidMiner Certified Analyst, Community Manager, Member, University Professor, PM Moderator Posts: 2,959 Community Manager
  • Options
    pschlunderpschlunder Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, RMResearcher, Member Posts: 96 RM Research
    without having it tested, yet. I'd should be possible to have multiple installations along-side each other. For the Deep Learning extension it's important that the CUDA installation is on the LD_LIBRARY_PATH environment variable. You should even be able to have both on the path, since our extension checks specifically for CUDA 9.0.

    This is referring to the Deep Learning extension, not the Keras extension.

    Hope this helps,
Sign In or Register to comment.