Catalyst Cloud provides several options for deloying instances with NVIDIA GPU acceleration that can be utilised by selecting the desired GPU flavor when creating the instances.
All GPU instances require suitable drivers to be installed in the operating system before the GPU can be utilised. This applies to both operating system images you upload, and to images provided by Catalyst Cloud.
Instances built using c2-gpu flavors use virtual GPU partitions which have special driver reqirements. Standard NVIDIA drivers supplied with operating systems cannot be used for c2-gpu instances. Please refer to the c2-gpu instructions for details.
All other GPU instances use direct pass-through of the entire GPU hardware, so any driver that is compatible with the GPU hardware can be installed. Catalyst Cloud recommends using drivers supplied by the operating system vendor or downloaded directly from NVIDIA.
These steps do not apply to Kubernetes worker nodes. For more information on using GPU with Kubernetes, please refer to the cluster GPU acceleration documentation.
To create a GPU-backed instance, simply create an instance using a GPU-backed flavor.
The list of GPU-backed types and which GPUs they provide is listed under the Accelerated Types section of the Instance types documentation.
Once the instance is created, suitable GPU drivers will need to be installed before the GPU can be used.
These requirements and instructions are for most GPU instance types except c2-gpu, which has a different process outlined below in Setup for c2-gpu Instance Types.
The minimum requirements for GPU instances are as follows:
A boot/OS disk of at least 30GB (when installing CUDA support). Windows instances should use at least 50GB.
Compatible NVIDIA GPU driver.
The following operating systems have been tested with GPU support:
Ubuntu LTS 20.04 and later.
Rocky Linux 8 and later.
Windows Server 2019 and later.
All other OS images are unsupported and untested.
The simplest method to install drivers is to use the packages supplied in the operating system repositories. The ‘server’ driver packages are recommended:
sudo apt install nvidia-driver-570-server
Note
NVIDIA recently began providing the option of either a proprietary or an open source driver. As of the 570 release the open source driver does not work on g4 (4x GPU) flavours on Catalyst Cloud.
Once the driver is installed, use nvidia-smi to verify that the driver is
loaded and the GPU(s) detected:
$ nvidia-smi
Mon Jun 30 02:40:15 2025
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 570.133.20 Driver Version: 570.133.20 CUDA Version: 12.8 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA L40S Off | 00000000:04:00.0 Off | 0 |
| N/A 29C P0 84W / 350W | 0MiB / 46068MiB | 3% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| No running processes found |
+-----------------------------------------------------------------------------------------+
If that doesn’t work, try running sudo modprobe nvidia to ensure the NVIDIA
driver is loaded, or reboot the instance.
Please refer to the CUDA section for instructions on installing the CUDA toolkit.
To use GPUs in Windows, simply download and install the latest drivers from:
https://www.nvidia.com/en-us/drivers/
Select the GPU model according to the flavour being used:
Compute Flavour |
GPU Model |
|---|---|
c1a-gpu |
NVIDIA RTX A6000 |
c2a-gpu |
NVIDIA A100 40GB |
c3-gpu |
NVIDIA L40S |
Once the driver package is installed, verify that it is installed by checking
the GPU state in Device Manager, or running nvidia-smi from the command
prompt.
For a more streamlined setup of GPU instances, the necessary GPU driver packages can be installed via user data when creating instances. This means the GPU is ready to use within a few minutes of the instance booting up without requiring additional steps.
User data example for automatically installing NVIDIA driver release 570 on Ubuntu 24.04:
#cloud-config
package_upgrade: true
packages:
- nvidia-driver-570-server
Other versions of Ubuntu and other distributions may require a different package name. Please refer to the documentation for the specific distribution for more examples.
Executing the NVIDIA driver installer with the /s argument runs it silently.
The examples below specify the 573 release for Windows Server 2022 and 2025 but
should work with any release if the URL is changed accordingly.
User data for Windows can be supplied as a straight PowerShell script or cloud-init style configuration in YAML format.
Windows Server 2019 and later include curl so regular cloud-init style
config data can be used to run the necessary commands directly:
#cloud-config
runcmd:
- curl -o nvidia.exe https://us.download.nvidia.com/tesla/573.39/573.39-data-center-tesla-desktop-winserver-2022-2025-dch-international.exe
- nvidia.exe /s
Alternatively a PowerShell script can be used directly instead:
#ps1
Invoke-WebRequest https://us.download.nvidia.com/tesla/573.39/573.39-data-center-tesla-desktop-winserver-2022-2025-dch-international.exe -OutFile nvidia.exe
nvidia.exe /s
Refer to the Cloudbase-Init documentation for more information on user data configuration options for Windows.
Unlike other GPU-backed types, c2-gpu instances are provided with a partition of an NVIDIA A100 GPU rather than the entire capacity of the card, using NVIDIA vGPU services. The partition size provided is “GRID A100D-20C”, which provides two compute pipelines and 20GB of video RAM from the underlying GPU.
vGPUs are isolated in hardware between different consumers, so there is no risk of information leaking or performance problems from other users of the same physical GPU.
For c2-gpu, the absolute minimum requirements are as follows:
A boot/OS disk of at least 30GB (when installing CUDA support)
Compatible NVIDIA vGPU driver. This is currently version 535.154.05.
The version of the driver loaded into your virtual server must be exactly this version, and not any other. From time to time we will update the version needed, and inform you when this updated will be required on your virtual servers.
Note
Drivers provided by OS or distribution vendors should not be installed. Only the drivers specified here will function with the vGPUs available.
Installing Ubuntu HWE kernel packages on Ubuntu is not recommended.
In addition, NVIDIA support only the following server operating systems for your vGPU virtual server while running in Catalyst Cloud:
Ubuntu LTS 20.04, 22.04 and 24.04
Tested by Catalyst Cloud, but not supported by NVIDIA are the following server operating systems:
Rocky Linux 8, 9
All other OS images are unsupported or untested.
To create a GPU-enabled virtual server, create an instance using a flavor prefixed with “c2-gpu”.
To help with streamlining c2-gpu server builds we’ve provided examples on using Packer to build custom images that include GPU drivers and software. This process is recommended for bulk GPU compute deployments.
Once you have created an Ubuntu virtual server using a version supported by the NVIDIA drivers, you will need to perform the following steps.
First, ensure all packages are up to date on your server and it is running the latest kernel (which will require a reboot):
sudo apt update
sudo apt dist-upgrade -y
sudo reboot
Then download and install the GRID driver package.
sudo apt install -y dkms
curl -O https://object-storage.nz-por-1.catalystcloud.io/v1/AUTH_483553c6e156487eaeefd63a5669151d/gpu-guest-drivers/nvidia/grid/16.7/linux/nvidia-linux-grid-535_535.183.06_amd64.deb
sudo dpkg -i nvidia-linux-grid-535_535.183.06_amd64.deb
Note
If you get a 404 response to this download, contact Catalyst Cloud support as the driver versions may have been updated making this documentation outdated.
Next, you will need to install the client license for vGPU support.
Download and save the license to /etc/nvidia/ClientConfigToken on
your virtual server, using the following steps:
(cd /etc/nvidia/ClientConfigToken && curl -O https://object-storage.nz-por-1.catalystcloud.io/v1/AUTH_483553c6e156487eaeefd63a5669151d/gpu-guest-drivers/nvidia/grid/licenses/client_configuration_token_12-29-2022-15-20-23.tok)
Edit the GRID driver configuration file /etc/nvidia/gridd.conf and
ensure that FeatureType is set to 1. Then restart the
nvidia-gridd service. The following commands apply the setting and
restart the service:
sudo sed -i -e '/^\(FeatureType=\).*/{s//\11/;:a;n;ba;q}' -e '$aFeatureType=1' /etc/nvidia/gridd.conf
sudo systemctl restart nvidia-gridd
After the service has been restarted, check the license status of the vGPU:
nvidia-smi -q | grep 'License Status'
This should return a line stating it is “Licensed” with an expiry in the future.
(Optional) Install the CUDA toolkit, if CUDA support is needed:
curl -O https://developer.download.nvidia.com/compute/cuda/12.2.2/local_installers/cuda_12.2.2_535.104.05_linux.run
sudo sh cuda_12.2.2_535.104.05_linux.run --silent --toolkit
This will run without any visible output for a while, before returning to a command prompt.
Note
We do not recommend using Debian or Ubuntu packages for the installation of CUDA toolkit. Those packages conflicts with required driver versions and will break your vGPU support.
To complete CUDA tookit installation, ensure that the CUDA libraries are available for applications to link and load:
sudo tee /etc/ld.so.conf.d/cuda.conf <<< /usr/local/cuda/lib64
sudo ldconfig
Linux distributions derived from RHEL, such as Rocky Linux, need the following steps to install the drivers.
Note
NVIDIA do not support RHEL-derived Linux distributions on Catalyst Cloud
First, ensure all packages are up to date on your server and it is running the latest kernel:
sudo dnf update -y && sudo reboot
Then install kernel source and related development tools:
sudo dnf install -y kernel-devel make
(Optional) Next, enable EPEL repositories and install DKMS support. This will automatically rebuild the drivers on kernel upgrades, rather than forcing you to re-install the GRID drivers every time the kernel is updated.
sudo dnf install -y epel-release
sudo dnf install -y dkms
Then install the GRID driver package:
curl -O https://object-storage.nz-por-1.catalystcloud.io/v1/AUTH_483553c6e156487eaeefd63a5669151d/gpu-guest-drivers/nvidia/grid/16.7/linux/NVIDIA-Linux-x86_64-535.183.06-grid.run
sudo sh NVIDIA-Linux-x86_64-535.183.06-grid.run -s -Z
Note
If you get a 404 response to this download, contact Catalyst Cloud support as the driver versions may have been updated making this documentation outdated.
This may produce errors or warnings related to missing X libraries and Vulkan ICD loader. These warnings can be safely ignored.
It may also produce an error about failing to register with DKMS, if you installed DKMS support above. This can be safely ignored, the modules will be rebuilt automatically despite the error message.
Next, you will need to install the client license for vGPU support.
Download and save the license to /etc/nvidia/ClientConfigToken on
your virtual server, using the following steps:
(cd /etc/nvidia/ClientConfigToken && curl -O https://object-storage.nz-por-1.catalystcloud.io/v1/AUTH_483553c6e156487eaeefd63a5669151d/gpu-guest-drivers/nvidia/grid/licenses/client_configuration_token_12-29-2022-15-20-23.tok)
Edit the GRID driver configuration file /etc/nvidia/gridd.conf and
ensure that FeatureType is set to 1. Then restart the nvidia-
gridd service. The following commands apply the setting and restart
the service:
sudo sed -i -e '/^\(FeatureType=\).*/{s//\11/;:a;n;ba;q}' -e '$aFeatureType=1' /etc/nvidia/gridd.conf
sudo systemctl restart nvidia-gridd
After the service has been restarted, check the license status of the vGPU:
nvidia-smi -q | grep 'License Status'
This should return a line stating it is “Licensed” with an expiry date in the future.
(Optional) Install the CUDA toolkit, if CUDA support is needed:
curl -O https://developer.download.nvidia.com/compute/cuda/12.2.2/local_installers/cuda_12.2.2_535.104.05_linux.run
sudo sh cuda_12.2.2_535.104.05_linux.run --silent --toolkit
This will run without any visible output for a while, before returning to a command prompt.
Note
We do not recommend using distribution-provided packages for the installation of CUDA toolkit. Those packages conflicts with required driver versions and will break your vGPU support.
To complete CUDA tookit installation, ensure that the CUDA libraries are available for applications to link and load:
sudo tee /etc/ld.so.conf.d/cuda.conf <<< /usr/local/cuda/lib64
sudo ldconfig
The CUDA version supported is specific to the NVIDIA driver release version. For most GPU flavors, simply download the appropriate CUDA toolkit version from NVIDIA to match the driver release used in the instance:
https://docs.nvidia.com/cuda/cuda-installation-guide-linux/
Some operating systems (e.g. Ubuntu) include CUDA packages in their repositories that can also be used instead, although they are usually older versions.
In the case of c2-gpu instances the CUDA toolkit version is currently limited by driver release 535 which officially supports CUDA 12.2.
NVIDIA provide compatibility libraries to allow applications compiled against newer CUDA releases to work. There are some caveats to this. Please refer to the NVIDIA CUDA compatibility guide for more information:
https://docs.nvidia.com/deploy/pdf/CUDA_Compatibility.pdf
Catalyst Cloud suggests the following approach to enable CUDA 12.4 compatibility (for example) with c2-gpu on Ubuntu instances.
Add the NVIDIA CUDA repo and signing keys and update the APT cache:
wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu$(lsb_release -rs | tr -d .)/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt update
Install the CUDA 12.4 compatibility package:
sudo apt install cuda-compat-12-4
When running your application you’ll need to set the LD_LIBRARY_PATH
environment variable to the location of the CUDA compatibility libraries, which
in this case is /usr/local/cuda-12.4/compat. For example:
LD_LIBRARY_PATH=/usr/local/cuda-12.4/compat /path/to/application
If a different CUDA compatibility level is required then this can be substituted in the steps above, provided NVIDIA have provided it.
NVIDIA provide documentation on supporting vGPU access from Docker containers here:
https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html