About CUDA, Cudnn

1

Hi,
I want to setup a DeepLearing environment for my project.

After installing ubuntu 20.04 on my MSI GT83VR 6RF which has two Nvidia GTX 1080 gpus, I then installed Nvidia Driver 530 on it! after that i then installed CUDA and Cudnn on it! from the website!

Now i can see
$ nvidia-smi
Tue Jun 27 11:39:23 2023
±--------------------------------------------------------------------------------------+
| NVIDIA-SMI 530.41.03 Driver Version: 530.41.03 CUDA Version: 12.1 |
|-----------------------------------------±---------------------±---------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+======================+======================|
| 0 NVIDIA GeForce GTX 1080 Off| 00000000:01:00.0 Off | N/A |
| N/A 34C P8 4W / N/A| 131MiB / 8192MiB | 0% Default |
| | | N/A |
±----------------------------------------±---------------------±---------------------+
| 1 NVIDIA GeForce GTX 1080 Off| 00000000:02:00.0 Off | N/A |
| N/A 38C P8 7W / N/A| 6MiB / 8192MiB | 0% Default |
| | | N/A |
±----------------------------------------±---------------------±---------------------+

±--------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=======================================================================================|
| 0 N/A N/A 1148 G /usr/lib/xorg/Xorg 100MiB |
| 0 N/A N/A 1485 G /usr/bin/gnome-shell 28MiB |
| 1 N/A N/A 1148 G /usr/lib/xorg/Xorg 4MiB |
±--------------------------------------------------------------------------------------+

It has Cuda version 12.1 as can be seen from the output!
and then
$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Sun_Jul_28_19:07:16_PDT_2019
Cuda compilation tools, release 10.1, V10.1.243

shows cuda V10 !!

but for my deeplearning env, i have used conda to activate an ENV!
So after activating, i installed pytorch with the cuda 11.8 version!

Now inside my env using vscode in python :
Pytorch Version: 2.0.1
Cuda Availablity: True
GPU Count: 2
CUDA version: 11.8
cuDNN version: 8700
GPU Count: 2
GPU Name: NVIDIA GeForce GTX 1080
_CudaDeviceProperties(name=‘NVIDIA GeForce GTX 1080’, major=6, minor=1, total_memory=8105MB, multi_processor_count=20)

I’m not sure if my system has conflicts as the cuda and cudnn version?
Ho to fix this conflict? ( do i need to remove all the cuda and cudnn and create a new ENV with desired cuda and cudnn inside each one?

For now i want to run a program that needs libcudnn8=8.1.0.77-1+cuda11.2
What should i do?

Are CONDA ENVs isolated from each other and also to the original system?
can i have multiple cuda and cudnn in the same ENV or in the same system with different ENVs?

Thanks
Best regards

2

Your locally installed CUDA toolkit(s) won’t be used unless you build PyTorch from source or a custom CUDA extension. The PyTorch binaries (pip wheels and conda binaries) ship with their own CUDA dependencies and just require a properly installed NVIDIA driver.

3

@ptrblck
Thanks for the prompt reply!
So you mean that each conda ENV are isolated from the others?
I mean can i install cuda 12.1 on the base system and create a new ENV with diffrenet Cuda and Cudnns?

How can i install old version of cudnn and cuda ( libcudnn8=8.1.0.77-1+cuda11.2) along with the pytorch 2.1?
Does my GPU supports cuda 12.1 or just A100, V100, T4 supports it?

Best regards

4

Yes, because as I mentioned before the locally installed CUDA toolkit won’t even be used by the PyTorch binaries.

You can select the desired PyTorch and CUDA runtime on our website and copy/paste the install command.

Your Pascal GPU is supported in CUDA 12.1.