Skip to main content

NVIDIA Driver Installation and Verification for A100 GPUs

The NVIDIA A100 GPU requires installation of the correct driver stack to enable compute workloads, virtualization, and advanced features such as Multi-Instance GPU (MIG). This article outlines the driver installation process, along with key steps to verify operational readiness in a private cloud environment.

Downloading the NVIDIA Driver​

Obtain the latest stable production driver that supports the A100 GPU from the official NVIDIA website or your software repository. Ensure that the driver version matches the kernel and CUDA toolkit requirements of your environment.

For bare metal installations, OpenMetal recommends downloading the .run installer package for maximum control:

wget https://us.download.nvidia.com/tesla/<VERSION>/NVIDIA-Linux-x86_64-<VERSION>.run

Replace <VERSION> with the specific driver release number required for your deployment.

Preparing the System for Installation​

Before running the installer:

Stop the display manager if running a GUI environment:

sudo systemctl stop gdm

Blacklist the nouveau driver to prevent conflicts:

echo "blacklist nouveau" | sudo tee -a /etc/modprobe.d/blacklist-nouveau.conf
sudo update-initramfs -u

Reboot the system to apply changes.

Ensure that kernel headers and development packages are installed:

sudo apt install -y build-essential dkms linux-headers-$(uname -r)

Installing the NVIDIA Driver​

Run the installer in text mode to begin installation:

chmod +x NVIDIA-Linux-x86_64-<VERSION>.run
sudo ./NVIDIA-Linux-x86_64-<VERSION>.run --silent

Options during installation:

Allow the installer to compile the kernel module.

Select "yes" to install the 32-bit compatibility libraries if required by your applications.

Skip driver signing unless Secure Boot is enabled and configured.

After installation, verify that the NVIDIA kernel module is loaded:

lsmod | grep nvidia

Validating the Driver and GPU Status​

The primary validation tool is nvidia-smi, which provides real-time information about GPU status and driver installation:

nvidia-smi

The output should display:

  • Driver version
  • GPU model (A100)
  • PCIe bus ID
  • Memory usage
  • Compute processes (if active)

Example output snippet:

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 535.104.12    Driver Version: 535.104.12    CUDA Version: 12.2   |
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
|  0   A100-PCIE-40GB     On    | 00000000:81:00.0 Off |                    0 |
+-----------------------------------------------------------------------------+

Use the following command to check for available Multi-Instance GPU profiles if planning to configure MIG:

nvidia-smi mig -lgip

Verifying PCIe Performance​

Confirm the GPU is linked at the correct PCIe generation and width to avoid performance bottlenecks:

sudo nvidia-smi topo --matrix

This output provides insight into interconnect bandwidth, which is critical for high-throughput AI/ML workloads.

Enabling Persistent Mode (Optional)​

For environments where GPU power cycling is undesired, enable persistence mode to keep the GPU initialized:

nvidia-smi -pm 1

Persistence mode is recommended for systems running continuous or long-duration workloads.

Next Steps​

With the driver installed and verified, the system is ready to support A100 workloads. The next phase involves configuring SR-IOV and Multi-Instance GPU (MIG) features for fine-grained resource sharing and workload isolation.

OpenMetal's private cloud platforms are built to support full GPU virtualization, allowing customers to run AI/ML workloads with dedicated hardware performance and predictable resource availability.