GoogleCloudPlatform / GoogleCloudPlatform/compute-gpu-installation
Uninstall CUDA toolkit
- Dominant language
- Python
- Stars
- 89
- Forks
- 40
- PR merge metrics
- No merged PRs in 30d
Description
**Describe the bug**
I have previously installed this toolkit using the provided startup script.
Now `nvidia-smi` shows no connection to the GPU. When I try re-running the installation script I get:
```
CUDA toolkit already marked as installed
```
Would there be an easy to way uninstall all related files and do a 'fresh install'? Have tried rebooting the server numerous times as well as running the script manually, with no luck.
Here are the last lines from the logs:
```
[2024-06-13 21:00:19,435] INFO - Installing prerequisite packages and updating kernel...
[2024-06-14 09:12:35,515] INFO - Switching to working directory: /opt/google/cuda-installer
[2024-06-14 09:12:35,703] INFO - Downloading CUDA installation kit...
[2024-06-14 09:12:35,703] INFO - Executing: sha256sum cuda_12.4.1_550.54.15_linux.run
[2024-06-14 09:13:03,226] INFO - 367d2299b3a4588ab487a6d27276ca5d9ead6e394904f18bccb9e12433b9c4fb cuda_12.4.1_550.54.15_linux.run
[2024-06-14 09:13:03,226] INFO - Installing CUDA toolkit...
[2024-06-14 09:13:03,227] INFO - Executing: sh cuda_12.4.1_550.54.15_linux.run --silent --toolkit
[2024-06-14 09:15:20,525] INFO - CUDA toolkit installation completed!
[2024-06-14 09:15:20,525] INFO - Executing post-installation actions...
[2024-06-14 09:15:20,527] INFO - CUDA post-installation actions completed!
[2024-06-14 09:15:20,528] INFO - The system needs to be rebooted to complete the installation process. The process will be continued after the reboot.
[2024-06-14 09:15:20,528] INFO - Rebooting now.
[2024-06-14 09:15:20,528] INFO - Executing: reboot now
[2024-06-14 09:15:42,209] INFO - Switching to working directory: /opt/google/cuda-installer
[2024-06-14 09:15:42,237] INFO - CUDA toolkit already marked as installed.
[2024-06-17 14:55:05,142] INFO - Switching to working directory: /opt/google/cuda-installer
[2024-06-17 14:55:05,166] INFO - CUDA toolkit already marked as installed.
```
**Environment**
- Operating system: Debian 5.10.221-1 (2024-07-14) x86_64 GNU/Linux
- Graphics card: A100 40GB x1
**To Reproduce**
Steps to reproduce the behavior. A copy-paste of the commands you executed and the output you got. The Linux installation script by default saves its' output to `/opt/google/gpu-installer/` - please provide the files you find there.
Contributor guide
Assessment
This issue has not been assessed yet.