Overview

This article describes how to setup a GPU-accelerated graphical desktop on Amazon EC2 instances running Amazon Linux 2023 (AL2023) with NVIDIA GRID driver, and access it using Amazon DCV high performance remote desktop protocol.

This article applies to AL2023 only. Similar articles are available for Ubuntu and Windows.

Other install options

• Consider managed EUC (End User Computing) services such as Amazon WorkSpaces
• Use this repository CloudFormation templates to automate setup on AL2023, Ubuntu, Windows or other OSs
• Setup Deep Learning workstation with AWS Deep Learning AMIs
• Install NVIDIA Tesla driver and graphical desktop
• Install desktop environment, NVIDIA GRID driver and Amazon DCV Server from documentation
• Amazon Machine Images (AMIs) with NVIDIA GRID drivers installed are available from AWS MarketPlace

Prerequisites

Verify that the On-Demand Instance quota value for Running On-Demand G and VT instances in your desired AWS Region is equal or greater than your desired instance size vCPUs count. Request a quota increase if otherwise.

Do not proceed until your applied quota value is equal or higher than your desired instance type vCPUs count

Launch EC2 instance

Launch a new EC2 instance with the following attributes

• Amazon Linux 2023
• G7e, G6, G6e, G6f, Gr6, Gr6f, G5 or G4dn EC2 instance type
• Security group that allows inbound TCP and UDP port 8443 from your IP
• More than 15 GB EBS volume size
• EC2 IAM role with AmazonS3ReadOnlyAccess managed policy attached for DCV license validation

Connect to the instance as ec2-user

Installation

Run the below commands

Prepare AL2023

Update OS

sudo dnf update -y

Optional: you may want to upgrade to latest release version (if available) and disable deterministic upgrade

sudo dnf upgrade --releasever=latest
echo latest | sudo tee /etc/dnf/vars/releasever

Restart your EC2 instance

sudo reboot

Install GNOME desktop and Amazon DCV server

sudo dnf clean all
sudo dnf groupinstall "Desktop" -y
sudo sed -i '/^\[daemon\]/a WaylandEnable=false' /etc/gdm/custom.conf
sudo systemctl set-default graphical.target

cd /tmp

sudo rpm --import https://d1uj6qtbmh3dt5.cloudfront.net/NICE-GPG-KEY
curl -L -O https://d1uj6qtbmh3dt5.cloudfront.net/nice-dcv-amzn2023-$(arch).tgz
tar -xvzf nice-dcv-amzn2023-$(arch).tgz && cd nice-dcv-*-amzn2023-$(arch)
sudo dnf install -y ./nice-dcv-server-*.rpm
sudo dnf install -y ./nice-dcv-web-viewer-*.rpm
sudo dnf install -y ./nice-xdcv-*.rpm
sudo dnf install -y ./nice-dcv-gltest-*.rpm

sudo systemctl enable dcvserver

sudo sed -i "/^\[session-management\/automatic-console-session/a owner=\"ec2-user\"\nstorage-root=\"%home%\"" /etc/dcv/dcv.conf
sudo sed -i "s/^#create-session/create-session/g" /etc/dcv/dcv.conf

sudo dnf install -y cups
sudo usermod -a -G sys dcv
sudo systemctl enable --now cups

Use of DCV is subject to DCV EULA (end user license agreement)

Install NVIDIA GRID driver

We will install NVIDIA GRID (also known as NVIDIA RTX Virtual Workstation (vWS)) driver. NVIDIA GRID drivers are certified to provide optimal performance for professional visualization applications that render content such as 3D models or high-resolution videos, and provide access to four 4K displays per GPU.

sudo dnf clean all
sudo dnf install -y vulkan-devel libglvnd-devel elfutils-libelf-devel xorg-x11-server-Xorg
sudo dnf install -y glx-utils vulkan-tools
sudo dnf install -y dkms 
sudo systemctl enable --now dkms

KVER=$(uname -r)
case $KVER in
  6.18.*)
    sudo dnf install -q -y kernel6.18-headers-$(uname -r) kernel6.18-devel-$(uname -r) kernel6.18-modules-extra-$(uname -r) kernel6.18-modules-extra-common-$(uname -r) --allowerasing
    ;;
  6.12.*)
    sudo dnf install -q -y kernel6.12-headers-$(uname -r) kernel6.12-devel-$(uname -r) kernel6.12-modules-extra-$(uname -r) kernel6.12-modules-extra-common-$(uname -r) --allowerasing
    ;;
  *)
    sudo dnf install -q -y kernel-headers-$(uname -r) kernel-devel-$(uname -r) kernel-modules-extra-$(uname -r) kernel-modules-extra-common-$(uname -r) --allowerasing
    ;;
esac

cd /tmp

aws s3 cp --recursive s3://ec2-linux-nvidia-drivers/latest/ . --no-sign-request
chmod +x NVIDIA-Linux-x86_64*.run
sudo ./NVIDIA-Linux-x86_64*.run -s

echo "options nvidia NVreg_EnableGpuFirmware=0" | sudo tee --append /etc/modprobe.d/nvidia.conf

sudo nvidia-xconfig  --enable-all-gpus --connected-monitor=DFP-0,DFP-1,DFP-2,DFP-3

Note that NVIDIA GRID driver is available for AWS customers only. By downloading and installing the software, you are bound by the terms of the NVIDIA GRID Cloud EULA (End User License Agreement).

Start X Windows and DCV server

sudo systemctl isolate multi-user.target && sudo systemctl isolate graphical.target
sudo systemctl restart dcvserver

Configure password

Configure a strong ec2-user password

sudo passwd ec2-user

Optional: GPU-accelerated container workloads

Besides GPU-accelerated graphics, you may want to run GPU-accelerated container workloads.

Install Docker and NVIDIA Container Toolkit

sudo dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/amzn2023/x86_64/cuda-amzn2023.repo

sudo dnf install -y docker
sudo systemctl enable docker
sudo usermod -aG docker ec2-user

sudo dnf install -y nvidia-container-toolkit
sudo nvidia-ctk runtime configure --runtime=docker
sudo systemctl restart docker

Refer to Install NVIDIA GPU driver, CUDA toolkit, NVIDIA Container Toolkit on Amazon EC2 instances running Amazon Linux 2023 (AL2023) (method 1, option 1) for more details including CUDA toolkit and other NVIDIA software installation.

Connect to AL2023 using DCV client

Download and install native client from https://www.amazondcv.com/.

Connect your macOS , Windows or Linux client to your EC2 instance public IP address. Login as ec2-user.

While you can connect using web browser client, native clients provide better user experience with additional features such as QUIC UDP transport protocol, multi-channel audio, and up to four 4K displays support.

Using Amazon DCV

Refer to DCV User Guide for usage guidance, e.g. transferring files.

EC2 Install Script

You can use the below as install script (or user data) to install NVIDIA GRID driver, docker with NVIDIA Container Toolkit, and Amazon DCV server on a new AL2023 GPU instance preferably with latest updates applied and with at least 15 GB storage.

#!/bin/bash
sudo dnf clean all
sudo dnf groupinstall "Desktop" -y
sudo sed -i '/^\[daemon\]/a WaylandEnable=false' /etc/gdm/custom.conf
sudo systemctl set-default graphical.target

cd /tmp

sudo rpm --import https://d1uj6qtbmh3dt5.cloudfront.net/NICE-GPG-KEY
curl -L -O https://d1uj6qtbmh3dt5.cloudfront.net/nice-dcv-amzn2023-$(arch).tgz
tar -xvzf nice-dcv-amzn2023-$(arch).tgz && cd nice-dcv-*-amzn2023-$(arch)
sudo dnf install -y ./nice-dcv-server-*.rpm
sudo dnf install -y ./nice-dcv-web-viewer-*.rpm
sudo dnf install -y ./nice-xdcv-*.rpm
sudo dnf install -y ./nice-dcv-gltest-*.rpm

sudo systemctl enable dcvserver

sudo sed -i "/^\[session-management\/automatic-console-session/a owner=\"ec2-user\"\nstorage-root=\"%home%\"" /etc/dcv/dcv.conf
sudo sed -i "s/^#create-session/create-session/g" /etc/dcv/dcv.conf

sudo dnf install -y cups
sudo usermod -a -G sys dcv
sudo systemctl enable --now cups

sudo dnf install -y vulkan-devel libglvnd-devel elfutils-libelf-devel xorg-x11-server-Xorg
sudo dnf install -y glx-utils vulkan-tools
sudo dnf install -y dkms 
sudo systemctl enable --now dkms

KVER=$(uname -r)
case $KVER in
  6.18.*)
    sudo dnf install -q -y kernel6.18-headers-$(uname -r) kernel6.18-devel-$(uname -r) kernel6.18-modules-extra-$(uname -r) kernel6.18-modules-extra-common-$(uname -r) --allowerasing
    ;;
  6.12.*)
    sudo dnf install -q -y kernel6.12-headers-$(uname -r) kernel6.12-devel-$(uname -r) kernel6.12-modules-extra-$(uname -r) kernel6.12-modules-extra-common-$(uname -r) --allowerasing
    ;;
  *)
    sudo dnf install -q -y kernel-headers-$(uname -r) kernel-devel-$(uname -r) kernel-modules-extra-$(uname -r) kernel-modules-extra-common-$(uname -r) --allowerasing
    ;;
esac

cd /tmp

aws s3 cp --recursive s3://ec2-linux-nvidia-drivers/latest/ . --no-sign-request
chmod +x NVIDIA-Linux-x86_64*.run
sudo ./NVIDIA-Linux-x86_64*.run -s

echo "options nvidia NVreg_EnableGpuFirmware=0" | sudo tee --append /etc/modprobe.d/nvidia.conf

sudo nvidia-xconfig --enable-all-gpus --connected-monitor=DFP-0,DFP-1,DFP-2,DFP-3

sudo dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/amzn2023/x86_64/cuda-amzn2023.repo

sudo dnf install -y docker
sudo systemctl enable docker
sudo usermod -aG docker ec2-user

sudo dnf install -y nvidia-container-toolkit
sudo nvidia-ctk runtime configure --runtime=docker
sudo systemctl restart docker

sudo reboot

Do set ec2-user user password and reboot EC2 instance before using.