problem installing via conda

[robomotic]

problem installing via conda

[priamai]

Holy moly guacamole, that worked pretty well.
Please add this in the tutorial or documentation, it will be helpful for other people.
Cheers!

[BartleyR]

It looks like you're trying to use CUDA 11.1, which we don't currently support due to our dependency on PyTorch. There is an internal team at Facebook working on addressing this. We don't have a timeline for this currently though.

For the time being, we'd advise using a CUDA version that's compatible with PyTorch (CUDA 10.2 is a great choice) if possible. You can have multiple CUDA versions installed on your system, if that's an option for you. Another option is to use our Docker container, which runs CUDA 10.2 inside but will work fine with the CUDA 11.0 toolkit on your bare metal (assuming you have a supported driver, which is very likely).

[BartleyR]

@robomotic did you have an opportunity to test this at all?

[robomotic]

Ops sorry my GPU server got torched, I have to reconfigure it. Please feel free to close the issue and I will re-open if it doesn't work from a fresh install.

[robomotic]

Testing now!

[robomotic]

Okay this is what I did:

docker pull rapidsai/rapidsai:cuda10.2-runtime-ubuntu18.04-py3.8
docker run --gpus all --rm -it -p 8888:8888 -p 8787:8787 -p 8786:8786  rapidsai/rapidsai:cuda10.2-runtime-ubuntu18.04-py3.8

Host Environment:

Ubuntu 20.04 x64
Cuda 11.0
Nvidia-Toolkit

Kernel driver:

alias: char-major-195-*
version: 455.38
supported: external
license: NVIDIA
srcversion: 3069C056DC628FE163D2CF4
name: nvidia
vermagic: 5.4.0-54-generic SMP mod_unload

I then try to import clx and I get the module not found error.
Let me know if there is something else I can try.

[efajardo-nv]

@robomotic Looks like the issue is that you're installing the nightly conda package (0.17) on the stable rapidsai docker image (0.16). Could you try this to install on rapidsai/rapidsai:cuda10.2-runtime-ubuntu18.04-py3.8:

conda install -c rapidsai -c nvidia -c pytorch -c conda-forge -c defaults clx

If you want to use the very latest, you would use the nightly rapidsai image and nightly clx conda package:

docker pull rapidsai/rapidsai-nightly:cuda10.2-runtime-ubuntu18.04-py3.8
conda install -c rapidsai-nightly -c nvidia -c pytorch -c conda-forge -c defaults clx

[BartleyR]

@robomotic Is this working for you now?

[robomotic]

Hi!
Thanks for the update I am going to test it shortly.
Cheers.

[efajardo-nv]

Closing this. Please create new issue if problem still persists.