Inconsistency when viewing the cuda version with nvcc -V and nvidia-smi:
nvcc -V version is 10.1
The version of nvidia-smi is 12.2
If the version can be displayed above, there is already a driver, and the previous driver must be deleted first:
1. Execute the following command to delete the old version of the driver
sudo apt-get purge nvidia*
At this time, execute nvidia-smi, it will prompt Command 'nvidia-smi' not found, follow the prompts to install:
sudo apt install nvidia-utils-535-server
After the installation is successful, execute nvidia-smi again, it is successful, and the cuda version is displayed as 12.2
2. Download the corresponding version of cuda
CUDA Toolkit 12.2 Downloads | NVIDIA Developer
Just select the corresponding options:
Get the download link:
wget https://developer.download.nvidia.com/compute/cuda/12.2.0/local_installers/cuda_12.2.0_535.54.03_linux.run
sudo sh cuda_12.2.0_535.54.03_linux.run
The cuda successful download interface is as follows:
3. Installation
Executing the original command installation will report an error, but the following command succeeds:
sudo sh ./cuda_12.2.0_535.54.03_linux.run --toolkit --silent --override
The above command installed successfully
Execute it again at this time, nvcc -V prompts, Command 'nvcc' not found,..., do not follow the prompts to install nvidia-cuda-toolkit, otherwise, the previous steps will be in vain. At this time, you only need to configure the environment variables:
4. Add environment variables
Execute: vim ~/.bashrc, open the file:
Add (here you need to modify it according to your own directory):
export LD_LIBRARY_PATH=/usr/local/cuda-12.2/lib64
export PATH=$PATH:/usr/local/cuda-12.2/bin
Execute: source ~/.bashrc, effective immediately
Execute again: nvcc -V, successful
This is the end.
References to this article:
CUDA has been installed but nvcc -V shows command not found (Ubuntu20.04LTS system)