sodcalculator

10 Tech How-To

NVIDIA Tesla Driver Install: Step-by-Step

Guide 10-43497 · Updated · Step by step

This guide walks you through installing NVIDIA Tesla drivers on a Linux system (Ubuntu/Debian or RHEL/CentOS) or Windows Server. Tesla drivers are optimized for data center GPUs (A100, V100, T4, etc.) and differ from GeForce drivers. You’ll need administrator/root access, an internet connection, and the exact GPU model to get the correct driver package.

What You’ll Need

  • A Tesla GPU (check with nvidia-smi or the system’s device manager)
  • Root or sudo access on Linux, or Administrator on Windows
  • Internet connection to download the driver from NVIDIA’s official site
  • For Linux: the kernel headers and development tools (gcc, make)
  • For Windows: the correct driver for your OS version (Server 2019/2022 or Windows 10/11)

Step-by-Step

  1. Identify your GPU model. Run lspci | grep -i nvidia on Linux or check Device Manager on Windows. Note the exact model (e.g., Tesla T4, V100, A100).
  2. Find the correct driver version. Go to NVIDIA’s official driver download page (nvidia.com/drivers). Select “Data Center / Tesla” as the product type, choose your GPU model, OS, and distribution. Note the driver version number (e.g., 550.54.15).
  3. Download the driver. On the page, click “Download” and save the file. For Linux, it will be a .run file; for Windows, an .exe.
  4. Prepare Linux system (if applicable). Install prerequisites: sudo apt update && sudo apt install build-essential linux-headers-$(uname -r) on Ubuntu/Debian, or sudo yum groupinstall "Development Tools" && sudo yum install kernel-devel on RHEL/CentOS.
  5. Disable the open-source Nouveau driver (Linux). Create a blacklist file: sudo bash -c "echo 'blacklist nouveau' >> /etc/modprobe.d/blacklist-nouveau.conf". Then run sudo update-initramfs -u (Ubuntu) or sudo dracut --force (RHEL). Reboot.
  6. Stop any running X server (Linux). If you have a GUI, switch to a text console with Ctrl+Alt+F3, then log in and run sudo systemctl isolate multi-user.target or sudo service gdm stop (depending on your display manager).
  7. Run the installer (Linux). Make the file executable: chmod +x NVIDIA-Linux-*.run. Then run sudo ./NVIDIA-Linux-*.run. Follow the prompts: accept the license, choose “Install” when asked about 32-bit compatibility, and allow it to update the X configuration if asked.
  8. Reboot (Linux). After installation, reboot the system. Then verify with nvidia-smi. You should see the GPU model and driver version.
  9. Windows installation. Double-click the downloaded .exe. If prompted by UAC, click “Yes”. Choose “Custom” installation, uncheck everything except the driver, and click “Next”. Let it install; the screen may flicker. Reboot when prompted.
  10. Verify on Windows. Open Command Prompt as Administrator and run nvidia-smi. If it shows the GPU, the driver is installed correctly.

Common Mistakes

  • Installing GeForce drivers instead of Tesla drivers – they are different packages.
  • Not blacklisting Nouveau on Linux, causing conflicts and installation failure.
  • Missing kernel headers – the installer will fail with “Unable to find the kernel source tree”.
  • Downloading the wrong architecture (32-bit vs 64-bit) – always choose the 64-bit version for modern servers.
  • Skipping the reboot after installation – the driver won’t load until then.

Tips

  • Use the latest stable driver branch (e.g., 550 series) for production, but check your GPU’s support matrix for the minimum version.
  • On Linux, if you have Secure Boot enabled, you’ll need to sign the driver or disable Secure Boot in BIOS – otherwise the kernel will reject the module.
  • Keep a backup of your current driver – you can always reinstall the old one if the new version causes issues.
  • For cluster or container use, consider the NVIDIA Container Toolkit after the driver is installed.
  • Check the NVIDIA documentation for your specific GPU – some models (e.g., A100) require a specific minimum driver version.

Troubleshooting

  • nvidia-smi not found: The driver isn’t in PATH. On Linux, check /usr/bin/nvidia-smi or /usr/local/bin. On Windows, it’s usually in C:\Program Files\NVIDIA Corporation\NVSMI.
  • Installation fails with “Nouveau is running”: You didn’t blacklist it properly. Re-check the blacklist file and run sudo update-initramfs -u again, then reboot.
  • Kernel module build error: Your kernel headers don’t match the running kernel. Update them: sudo apt install linux-headers-$(uname -r) or sudo yum install kernel-devel-$(uname -r).
  • GPU not detected after install: Check if the card is properly seated in the PCIe slot. Run lspci | grep -i nvidia – if it shows nothing, the card isn’t detected at all.
  • Windows driver won’t install: Uninstall any previous NVIDIA driver using Display Driver Uninstaller (DDU) in Safe Mode, then retry.
  • High power usage or fan noise: Tesla GPUs are designed for data centers; they may not have active cooling. Check that your system has adequate airflow.

Frequently Asked Questions

Can I use GeForce drivers on a Tesla GPU?

Technically, GeForce drivers may work for basic compute, but they lack the optimizations and features (like MIG, NVLink) that Tesla drivers provide. For production, always use Tesla drivers.

How do I update the driver later?

Download the new driver and follow the same steps. On Linux, you can run sudo ./NVIDIA-Linux-*.run --uninstall first, then install the new one. On Windows, just run the new installer over the old one.

Do I need to install CUDA separately?

No, the driver includes the CUDA runtime libraries. But for development, you may want to install the CUDA Toolkit separately to get compilers and libraries.

Why does nvidia-smi show 0% GPU utilization?

That’s normal if no compute job is running. The GPU is idle. When you start a CUDA workload, utilization will increase.

Can I install Tesla drivers on a laptop?

No, Tesla GPUs are not available in laptops. They are data center cards. For laptops, use GeForce or Quadro drivers.

What is the difference between Tesla and Quadro drivers?

Tesla drivers are for data center compute (AI, HPC), while Quadro drivers are for professional workstations (CAD, 3D rendering). They are not interchangeable for their intended hardware.

Related

General steps for current mainstream versions; menus can differ by device and version. See the disclaimer or report an error.

More in this bed

The whole Tech How-To bed →