GPU Support

Table of Contents

- [Introduction](#introduction) - [Filter and Reserve a GPU Node](#filter-and-reserve-a-gpu-node) - [Filter Nodes](#filter-nodes) - [Reserve a Node](#reserve-a-node) - [Deploy a VM with GPU](#deploy-a-vm-with-gpu) - [Install the GPU Driver](#install-the-gpu-driver) - [AMD Driver](#amd-driver) - [Nvidia Driver](#nvidia-driver) - [With an AI Model](#with-an-ai-model) - [Troubleshooting](#troubleshooting) - [GPU Support Links](#gpu-support-links) *** ## Introduction This section covers the essential information to deploy a node with a GPU. We also provide links to other parts of the manual covering GPU support. To use a GPU on the TFGrid, users need to rent a dedicated node. Once they have rented a dedicated node equipped with a GPU, users can deploy workloads on their dedicated GPU node. ## Filter and Reserve a GPU Node You can filter and reserve a GPU node using the [Dedicated Nodes section](dashboard@@node_finder) of the **ThreeFold Dashboard**. ### Filter Nodes * Filter nodes using the vendor name * In **Filters**, select **GPU's vendor name** * A new window will appear below named **GPU'S VENDOR NAME** * Write the name of the vendor desired (e.g. **nvidia**, **amd**) ![image](./img/gpu_8.png) * Filter nodes using the device name * In **Filters**, select **GPU's device name** * A new window will appear below named **GPU'S DEVICE NAME** * Write the name of the device desired (e.g. **GT218**) ![image](./img/gpu_9.png) ### Reserve a Node When you have decided which node to reserve, click on **Reserve** under the column **Actions**. Once you've rented a dedicated node that has a GPU, you can deploy GPU workloads. ![image](./img/gpu_2.png) ## Deploy a VM with GPU Now that you've reserverd a dedicated GPU node, it's time to deploy a VM to make use of the GPU! There are many ways to proceed. You can use the [Dashboard](fullvm.md), [Go](developers@@grid3_go_gpu), [Terraform](terraform_gpu_support.md), etc. For example, deploying a VM with GPU on the Dashboard is easy. Simply set the GPU option and make sure to select your dedicated node, as show here: ![image](./img/gpu_3.png) ## Install the GPU Driver Once you've deployed a VM with GPU, you want to SSH into the VM and install the GPU driver. - SSH to the VM and get your system updated ```bash dpkg --add-architecture i386 apt-get update apt-get dist-upgrade reboot ``` - Find your Driver installer - [AMD driver](https://www.amd.com/en/support/linux-drivers) - [Nvidia driver](https://www.nvidia.com/download/index.aspx) - You can see the node card details on the ThreeFold Dashboard or by using the following command lines: ```bash lspci | grep VGA lshw -c video ``` ### AMD Driver - Download the GPU driver using `wget` - For example: `wget https://repo.radeon.com/amdgpu-install/23.30.2/ubuntu/focal/amdgpu-install_5.7.50702-1_all.deb` - Install the GPU driver using `apt-get`. Make sure to update ``. ```bash apt-get install ./amdgpu-install_.deb amdgpu-install --usecase="dkms,graphics,opencl,hip,rocm,rocmdev,opencl,hiplibsdk,mllib,mlsdk" --opencl=rocr --vulkan=pro --opengl=mesa ``` - To verify that the GPU is properly installed, use the following command lines: ```bash rocm-smi rocminfo ``` - You should something like this: ![image](./img/gpu_4.png) ![image](./img/gpu_5.png) ### Nvidia Driver For Nvidia, you can follow [those steps](https://linuxize.com/post/how-to-nvidia-drivers-on-ubuntu-20-04/#installing-the-nvidia-drivers-using-the-command-line). - To verify that the GPU is properly installed, you can use `nvidia-smi`. You should something like this: ![image](./img/gpu_6.png) ### With an AI Model You can also try this [AI model](https://github.com/invoke-ai/InvokeAI#getting-started-with-invokeai) to install your driver. ## Troubleshooting Here are some useful links to troubleshoot your GPU installation. - [Steps to install the driver](https://amdgpu-install.readthedocs.io/en/latest/index.html) - Changing kernel version - [Link 1](https://linux.how2shout.com/how-to-change-default-kernel-in-ubuntu-22-04-20-04-lts/) - [Link 2](https://gist.github.com/chaiyujin/c08e59752c3e238ff3b1a5098322b363) > Note: It is recommended to use Ubuntu 22.04.2 LTS (GNU/Linux 5.18.13-051813-generic x86_64) ## GPU Support Links You can consult the [GPU Table of Contents](gpu_toc.md) to see all available GPU support links on the ThreeFold Manual.