Install NVIDIA GPU drivers on N-series VMs running Linux

Applies to: ✔️ Linux VMs

To take advantage of the GPU capabilities of Azure N-series VMs backed by NVIDIA GPUs, you must install NVIDIA GPU drivers.

If you choose to install NVIDIA GPU drivers manually, this article provides supported distributions, drivers, and installation and verification steps. Manual driver setup information is also available for Windows VMs.

For N-series VM specs, storage capacities, and disk details, see GPU Linux VM sizes.

Supported distributions and drivers

NVIDIA CUDA drivers

For the latest CUDA drivers and supported operating systems, visit the NVIDIA website. Ensure that you install or upgrade to the latest supported CUDA drivers for your distribution.

NVIDIA GRID drivers

Microsoft redistributes NVIDIA GRID driver installers for NV and NVv3-series VMs used as virtual workstations or for virtual applications. Install only these GRID drivers on Azure NV VMs, only on the operating systems listed in the following table. These drivers include licensing for GRID Virtual GPU Software in Azure. You don't need to set up a NVIDIA vGPU software license server.

The GRID drivers redistributed by Azure don't work on most non-NV series VMs like NCv3 VMs but works on NCasT4v3 series.

For more information on the specific vGPU and driver branch versions, visit the NVIDIA website.

Distribution	Driver
Ubuntu 20.04 LTS, 22.04 LTS Red Hat Enterprise Linux 7.9, 8.6, 8.8, 8.9, 8.10 SUSE Linux Enterprise Server 15 SP2, 12 SP2,12 SP5 Rocky Linux 8.4	NVIDIA vGPU 17.3, driver branch R550 NVIDIA vGPU 16.5, driver branch R535

Visit GitHub for the complete list of all previous Nvidia GRID driver links.

Install CUDA drivers on N-series VMs

Here are steps to install CUDA drivers from the NVIDIA CUDA Toolkit on N-series VMs.

C and C++ developers can optionally install the full Toolkit to build GPU-accelerated applications. For more information, see the CUDA Installation Guide.

To install CUDA drivers, make an SSH connection to each VM. To verify that the system has a CUDA-capable GPU, run the following command:

lspci | grep -i NVIDIA

Output is similar to the following example (showing an NVIDIA Tesla K80 card):

lspci lists the PCIe devices on the VM, including the InfiniBand NIC and GPUs, if any. If lspci doesn't return successfully, you may need to install LIS on CentOS/RHEL.

Then run installation commands specific for your distribution.

Ubuntu

Ubuntu packages NVIDIA proprietary drivers. Those drivers come directly from NVIDIA and are simply packaged by Ubuntu so that they can be automatically managed by the system. Downloading and installing drivers from another source can lead to a broken system. Moreover, installing third-party drivers requires extra-steps on VMs with TrustedLaunch and Secure Boot enabled. They require the user to add a new Machine Owner Key for the system to boot. Drivers from Ubuntu are signed by Canonical and will work with Secure Boot.

1. Install ubuntu-drivers utility:

   sudo apt update && sudo apt install -y ubuntu-drivers-common
   

2. Install the latest NVIDIA drivers:

   sudo ubuntu-drivers install
   

   Reboot the VM after the GPU driver is installed.

3. Download and install the CUDA toolkit from NVIDIA:

   Note

   The example shows the CUDA package path for Ubuntu 24.04 LTS. Replace the path specific to the version you plan to use.

   Visit the NVIDIA Download Center or the NVIDIA CUDA Resources page for the full path specific to each version.

   wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/x86_64/cuda-keyring_1.1-1_all.deb
   sudo apt install -y ./cuda-keyring_1.1-1_all.deb
   sudo apt update
   sudo apt -y install cuda-toolkit-12-5
   

   The installation can take several minutes.

4. Reboot the VM after installation completes:

   sudo reboot
   

5. Verify that the GPU is correctly recognized (after reboot):

   nvidia-smi
   

NVIDIA driver updates

We recommend that you periodically update NVIDIA drivers after deployment.

sudo apt update
sudo apt full-upgrade

CentOS or Red Hat Enterprise Linux

1. Update the kernel (recommended). If you choose not to update the kernel, ensure that the versions of kernel-devel, and dkms are appropriate for your kernel.

   sudo yum install kernel kernel-tools kernel-headers kernel-devel
   sudo reboot
   

2. Install the latest Linux Integration Services for Hyper-V and Azure. Check if LIS is required by verifying the results of lspci. If all GPU devices are listed as expected, installing LIS isn't required.

   LIS is applicable to Red Hat Enterprise Linux, CentOS, and the Oracle Linux Red Hat Compatible Kernel 5.2-5.11, 6.0-6.10, and 7.0-7.7. Refer to the Linux Integration Services documentation for more details. Skip this step if you plan to use CentOS/RHEL 7.8 (or higher versions) as LIS is no longer required for these versions.

   wget https://aka.ms/lis
   tar xvzf lis
   cd LISISO
   
   sudo ./install.sh
   sudo reboot
   

3. Reconnect to the VM and continue installation with the following commands:

   sudo rpm -Uvh https://dl.fedoraproject.org/pub/epel/epel-release-latest-7.noarch.rpm
   sudo yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel7/x86_64/cuda-rhel7.repo
   sudo yum clean all
   sudo yum -y install nvidia-driver-latest-dkms cuda-drivers
   

   The installation can take several minutes.

   Note

   Visit Fedora and Nvidia CUDA repo to pick the correct package for the CentOS or RHEL version you want to use.

For example, CentOS 8 and RHEL 8 need the following steps.

sudo rpm -Uvh https://dl.fedoraproject.org/pub/epel/epel-release-latest-8.noarch.rpm
sudo yum install dkms

sudo wget https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/cuda-rhel8.repo -O /etc/yum.repos.d/cuda-rhel8.repo

sudo yum install cuda-drivers

4. To optionally install the complete CUDA toolkit, type:

   sudo yum install cuda
   

   Note

   If you see an error message related to missing packages like vulkan-filesystem then you may need to edit /etc/yum.repos.d/rh-cloud , look for optional-rpms and set enabled to 1

5. Reboot the VM and proceed to verify the installation.

Verify driver installation

To query the GPU device state, SSH to the VM and run the nvidia-smi command-line utility installed with the driver.

If the driver is installed, Nvidia SMI lists the GPU-Util as 0% until you run a GPU workload on the VM. Your driver version and GPU details may be different from the ones shown.

RDMA network connectivity

RDMA network connectivity can be enabled on RDMA-capable N-series VMs such as NC24r deployed in the same availability set or in a single placement group in a virtual machine (VM) scale set. The RDMA network supports Message Passing Interface (MPI) traffic for applications running with Intel MPI 5.x or a later version:

Distributions

Deploy RDMA-capable N-series VMs from one of the images in the Azure Marketplace that supports RDMA connectivity on N-series VMs:

• Ubuntu 16.04 LTS - Configure RDMA drivers on the VM and register with Intel to download Intel MPI:

  1. Install dapl, rdmacm, ibverbs, and mlx4

     sudo apt-get update
     
     sudo apt-get install libdapl2 libmlx4-1
     
     

  2. In /etc/waagent.conf, enable RDMA by uncommenting the following configuration lines. You need root access to edit this file.

     OS.EnableRDMA=y
     
     OS.UpdateRdmaDriver=y
     

  3. Add or change the following memory settings in KB in the /etc/security/limits.conf file. You need root access to edit this file. For testing purposes you can set memlock to unlimited. For example: <User or group name> hard memlock unlimited.

     <User or group name> hard    memlock <memory required for your application in KB>
     
     <User or group name> soft    memlock <memory required for your application in KB>
     

  4. Install Intel MPI Library. Either purchase and download the library from Intel or download the free evaluation version.

     wget http://registrationcenter-download.intel.com/akdlm/irc_nas/tec/9278/l_mpi_p_5.1.3.223.tgz
     

     Only Intel MPI 5.x runtimes are supported.

     For installation steps, see the Intel MPI Library Installation Guide.

  5. Enable ptrace for non-root non-debugger processes (needed for the most recent versions of Intel MPI).

     echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope
     

• CentOS-based 7.4 HPC - RDMA drivers and Intel MPI 5.1 are installed on the VM.

• CentOS-based HPC - CentOS-HPC 7.6 and later (for SKUs where InfiniBand is supported over SR-IOV). These images have Mellanox OFED and MPI libraries pre-installed.

Troubleshooting

• You can set persistence mode using nvidia-smi so the output of the command is faster when you need to query cards. To set persistence mode, execute nvidia-smi -pm 1. Note that if the VM is restarted, the mode setting goes away. You can always script the mode setting to execute upon startup.
• If you updated the NVIDIA CUDA drivers to the latest version and find RDMA connectivity is no longer working, reinstall the RDMA drivers to reestablish that connectivity.
• During installation of LIS, if a certain CentOS/RHEL OS version (or kernel) is not supported for LIS, an error "Unsupported kernel version" is thrown. Please report this error along with the OS and kernel versions.
• If jobs are interrupted by ECC errors on the GPU (either correctable or uncorrectable), first check to see if the GPU meets any of Nvidia's RMA criteria for ECC errors. If the GPU is eligible for RMA, please contact support about getting it serviced; otherwise, reboot your VM to reattach the GPU as described here. Less invasive methods such as nvidia-smi -r don't work with the virtualization solution deployed in Azure.

Next steps

• To capture a Linux VM image with your installed NVIDIA drivers, see How to generalize and capture a Linux virtual machine.