Skip to content

Install latest CUDA version

This short guide will demonstrate how to install the latest CUDA version on your CentOS VM with access to GPU resources (e.g. P100 GPU).

First of all make sure that your VM is up to date and reboot the VM if a kernel update was installed.

sudo yum update -y && sudo reboot

We will install the EPEL repository because it contains the dkms package which we will need to install the NVIDIA drivers.

sudo yum install epel-release -y

Now we install the following packages to allow the installation of CUDA and the GPU drivers.

sudo yum install kernel-devel-`uname -r` kernel-headers-`uname -r` pciutils dkms wget -y

As the CentOS repositories do not provide any CUDA versions we have to add the official NVIDIA CUDA repository. Therefore visit the NVIDIA CUDA repository and choose the (rhel7) CUDA version you want to install. We will download the rpm with wget and install it via yum.

sudo yum install cuda-repo-rhel7-10.2.89-1.x86_64.rpm -y

In the last step we will install the latest CUDA package and reboot the VM.

sudo yum install cuda -y && sudo reboot

Now You can use nvidia-smi to validate if the installation was successfully.

$ sudo nvidia-smi
Mon Apr 27 14:18:12 2020       
| NVIDIA-SMI 440.64.00    Driver Version: 440.64.00    CUDA Version: 10.2     |
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|   0  Tesla P100-PCIE...  Off  | 00000000:00:05.0 Off |                    0 |
| N/A   29C    P0    27W / 250W |      0MiB / 16280MiB |      4%      Default |

| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|  No running processes found                                                 |