search

allegroai / clearml-fractional-gpu

ClearML Fractional GPU - Run multiple containers on the same GPU with driver level memory limitation ✨ and compute time-slicing
https://clear.ml
Other
57 stars 3 forks source link

【Feature request】cuDNN support #1

Open incomingflyingbrick opened 5 months ago

incomingflyingbrick commented 5 months ago

Dear clearml team, Is there any plan for cuDNN support? For example I want to use tensorflow, but it requires a certain version of cuDNN to work, link here -> https://www.tensorflow.org/install/source#gpu According to this table, tensorflow 2.15.0 requires CUDA 12.2 and cuDNN 8.9, is it possible to support something like that? like a build-in or pre-installed cuDNN to the docker image. Thz!

jkhenning commented 5 months ago

Hi @incomingflyingbrick,

We tried to keep it as lean as possible, but to add cudnn you could build your own docker with:

FROM clearml/fractional-gpu:u22-cu12.3-8gb

# upgrade torch to the latest version
RUN pip3 install -U clearml torch torchvision torchaudio torchdata torchmetrics torchrec torchtext nvidia-cudnn-cu12

Notice that the new torch versions actually install cudnn via pip install nvidia-cudnn-cu12 (see https://pypi.org/project/nvidia-cudnn-cu12/)