[SOLVED] How to modify this command to install specific cuDNN and torch version "apt install libcudnn8"

Issue

this is my system :

No LSB modules are available.
Distributor ID: Ubuntu
Description:    Ubuntu 22.04.2 LTS
Release:        22.04
Codename:       jammy

I want to install cudnn 8.7.0 - i think exact version is 8.7.0.84_1

And also cuda11.7

this command : apt install libcudnn8

Installs : Unpacking libcudnn8-dev (8.9.2.26-1+cuda12.1) over (8.5.0.96-1+cuda11.7) ...

So how to modify apt install libcudnn8 so that it installs cuDNN 8.7.0 and Cuda 11.7?

Thank you

Solution

According to the Nvidia docs (https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html#package-manager-ubuntu-install):

sudo apt-get install libcudnn8=${cudnn_version}-1+${cuda_version}
sudo apt-get install libcudnn8-dev=${cudnn_version}-1+${cuda_version}
sudo apt-get install libcudnn8-samples=${cudnn_version}-1+${cuda_version}

Where:

• ${cudnn_version} is for example 8.9.2.*
• ${cuda_version} is for example cuda12.1 or cuda11.8

You can find a list of installable versions for example by running:

sudo apt update
sudo apt list -a libcudnn8-dev

And then substitute the variables in the commands above with the commands that you require.

For RunPod you are logged into a docker container, and the sudo command is not installed so you can obviously drop the sudo prefix since you are logged in as the root user anyway, so for your specific use case on RunPod the command would be:

apt install libcudnn8=8.7.0.84-1+cuda11.8 libcudnn8-dev=8.7.0.84-1+cuda11.8

Note: Although you specifically specify CUDA 11.7, there is no available package to satisfy that particular requirement, which is why CUDA 11.8 was selected intead.

Answered By - Ashley Kleynhans

Answer Checked By - Mary Flores (WPSolving Volunteer)