> ## Documentation Index > Fetch the complete documentation index at: https://docs.poolside.ai/llms.txt > Use this file to discover all available pages before exploring further. # Prepare Ubuntu > Prepare an Ubuntu host before you install Poolside model inference on-premises. export const nvidiaDriverVersion = "580.126.20"; Complete these steps on an Ubuntu host before you run the [on-premises installation](/deployment/on-prem/install). For requirements that apply to every host, see [On-premises installation prerequisites](/deployment/on-prem/prerequisites/overview). ## Supported versions and architectures | Item | Supported values | | ---------------- | --------------------------------------- | | Operating system | Ubuntu 22.04 LTS, Ubuntu 24.04 LTS | | CPU architecture | `amd64` (`x86_64`), `arm64` (`aarch64`) | The steps on this page apply to both Ubuntu 22.04 LTS and Ubuntu 24.04 LTS. ## Step 1: Install required tools * Install `kubectl` using `sudo snap install kubectl --classic` * Install `jq` using `sudo apt install -y jq` * Install `yq` (version `v4.49.2` or later) from the [yq releases page](https://github.com/mikefarah/yq/releases/tag/v4.49.2) * Download [`yq_linux_amd64.tar.gz`](https://github.com/mikefarah/yq/releases/download/v4.49.2/yq_linux_amd64.tar.gz) or [`yq_linux_arm64.tar.gz`](https://github.com/mikefarah/yq/releases/download/v4.49.2/yq_linux_arm64.tar.gz) and install it to `/usr/local/bin/yq` * Install `terraform` (version `1.8.5`) from the [Terraform 1.8.5 releases page](https://releases.hashicorp.com/terraform/1.8.5) * Download [`terraform_1.8.5_linux_amd64.zip`](https://releases.hashicorp.com/terraform/1.8.5/terraform_1.8.5_linux_amd64.zip) or [`terraform_1.8.5_linux_arm64.zip`](https://releases.hashicorp.com/terraform/1.8.5/terraform_1.8.5_linux_arm64.zip) and install the binary to `/usr/local/bin/terraform` * `unzip` is required to extract the Terraform binary * Install `skopeo` (version `v1.18` or later) from the [skopeo-binary releases page](https://github.com/lework/skopeo-binary/releases/tag/v1.20.0) * Download [`skopeo-linux-amd64`](https://github.com/lework/skopeo-binary/releases/download/v1.20.0/skopeo-linux-amd64) or [`skopeo-linux-arm64`](https://github.com/lework/skopeo-binary/releases/download/v1.20.0/skopeo-linux-arm64) and install it to `/usr/local/bin/skopeo` ## Step 2: Configure the containers trust policy Ensure the containers trust policy at `/etc/containers/policy.json` allows `skopeo` to access the RKE2 registry with the minimum required permissions. Replace `` with the fully qualified path to the root of the unpacked installation bundle. This allows the bundled images to be loaded into the local RKE2 registry during [Step 2 of the on-premises installation](/deployment/on-prem/install#step-2-install-supporting-infrastructure-services). ```json theme={null} { "default": [ { "type": "reject" } ], "transports": { "docker": { "localhost:30500": [{"type": "insecureAcceptAnything"}] }, "dir": { "": [{"type": "insecureAcceptAnything"}] } } } ``` ## Step 3: Configure kernel parameters Poolside file watchers can exceed the Ubuntu default for inotify instances. Set the following parameter to `65535` or higher: ```text theme={null} fs.inotify.max_user_instances = 65535 ``` To apply the setting, add the parameter under `/etc/sysctl.d/` and reload: ```bash theme={null} echo "fs.inotify.max_user_instances = 65535" | sudo tee /etc/sysctl.d/99-poolside.conf sudo sysctl --system ``` ## Step 4: Disable the `nouveau` driver Confirm that the `nouveau` graphics driver is not loaded. For instructions, see [Disable the nouveau driver in the NVIDIA documentation](https://docs.nvidia.com/ai-enterprise/deployment/vmware/latest/nouveau.html). Run the following command to check whether the `nouveau` driver is loaded. If the command returns output, follow the next steps to turn off the driver and reboot. ```bash theme={null} lsmod | grep nouveau ``` If the `nouveau` driver is loaded: ```bash theme={null} cat < The driver compiled by the GPU Operator is not signed and is not compatible with Secure Boot. If you require Secure Boot, preinstall the driver on the host. ### Use external Ubuntu package repositories If the host can reach the Ubuntu package repositories and Secure Boot is disabled, do not preinstall an NVIDIA driver. The GPU Operator installs the driver during [Step 2 of the on-premises installation](/deployment/on-prem/install#step-2-install-supporting-infrastructure-services). ### Use the host APT cache On an air-gapped host, the GPU Operator can compile the driver if you ran `apt update` before disconnecting the host and the APT cache contains these packages for the kernel version reported by `uname -r`: * `linux-headers-` * `linux-image-` * `linux-modules-` If the cache does not contain all required packages, use an internal APT mirror or preinstall the driver. ### Use an internal APT mirror If the host can reach an internal APT mirror, configure the GPU Operator driver container to use it. Before you run Step 2 of the installation, add the following configuration to `02-infra-services/terraform.tfvars`: ```hcl theme={null} gpu_operator_driver_container_config = { package_repository_config_b64 = "" ca_certificate_config_b64 = "" environment_variables = {} } ``` * `package_repository_config_b64`: A base64-encoded APT repository configuration in the format described in [NVIDIA's air-gapped GPU Operator documentation](https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/26.3/install-gpu-operator-air-gapped.html) * `ca_certificate_config_b64`: An optional base64-encoded CA chain, including intermediate certificates, when the mirror uses a TLS certificate signed by an internal CA * `environment_variables`: An optional map of `NAME = VALUE` environment variables passed to the driver container, such as `PROXY` and `NO_PROXY` ### Preinstall the NVIDIA driver Preinstall the driver if you require Secure Boot or the GPU Operator cannot access the required kernel packages. Follow the [NVIDIA driver installation instructions for Ubuntu](https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#ubuntu). Install NVIDIA driver version {nvidiaDriverVersion}, which is the version fully validated with this installation bundle. Earlier driver versions are not supported. Later versions might work, but can cause compatibility issues because they have not been validated against this bundle. After you install the driver, run `nvidia-smi` and confirm that it reports the host GPUs. During installation, the GPU Operator detects the host driver and skips driver compilation. ## Step 6: Verify prerequisites Your host is ready when the following checks pass: * Confirm that the required tools are installed and on the path: ```bash theme={null} kubectl version --client jq --version /usr/local/bin/yq --version /usr/local/bin/terraform version /usr/local/bin/skopeo --version ``` * Confirm that the `nouveau` driver is not loaded. This command returns no output: ```bash theme={null} lsmod | grep nouveau ``` ## Next steps * [Install on-premises](/deployment/on-prem/install) ## Related resources * [On-premises installation prerequisites](/deployment/on-prem/prerequisites/overview) * [Install on-premises](/deployment/on-prem/install) * [On-premises deployment](/deployment/on-prem/overview) * [Supported configurations](/deployment/supported-configurations)