AWS AWSEC2 documentation change
Summary
Restructured documentation by removing 'Install an additional version of CUDA' section, consolidating installation options into separate pages, and reorganizing content hierarchy
Security assessment
The changes primarily involve content reorganization and removal of redundant installation instructions. No security vulnerabilities, patches, or security-related configuration changes are mentioned. The modifications focus on improving documentation structure rather than addressing security concerns.
Diff
diff --git a/AWSEC2/latest/UserGuide/install-nvidia-driver.md b/AWSEC2/latest/UserGuide/install-nvidia-driver.md index b84439dc4..410039e7c 100644 --- a//AWSEC2/latest/UserGuide/install-nvidia-driver.md +++ b//AWSEC2/latest/UserGuide/install-nvidia-driver.md @@ -5 +5 @@ -Types of NVIDIA driversAvailable drivers by instance typeInstallation optionsInstall an additional version of CUDA +Types of NVIDIA driversAvailable drivers by instance typeInstallation options @@ -13 +13 @@ To install AMD drivers on an instance with an attached AMD GPU, such as a G4ad i -###### Contents +###### Topics @@ -15 +15 @@ To install AMD drivers on an instance with an attached AMD GPU, such as a G4ad i - * [Types of NVIDIA drivers](./install-nvidia-driver.html#nvidia-driver-types) + * Types of NVIDIA drivers @@ -17 +17 @@ To install AMD drivers on an instance with an attached AMD GPU, such as a G4ad i - * [Available drivers by instance type](./install-nvidia-driver.html#nvidia-driver-instance-type) + * Available drivers by instance type @@ -19 +19 @@ To install AMD drivers on an instance with an attached AMD GPU, such as a G4ad i - * [Installation options](./install-nvidia-driver.html#nvidia-installation-options) + * Installation options @@ -21 +21 @@ To install AMD drivers on an instance with an attached AMD GPU, such as a G4ad i - * [Option 1: AMIs with NVIDIA drivers installed](./install-nvidia-driver.html#preinstalled-nvidia-driver) + * [Use AMIs that include NVIDIA drivers](./preinstalled-nvidia-driver.html) @@ -23 +23 @@ To install AMD drivers on an instance with an attached AMD GPU, such as a G4ad i - * [Option 2: Public NVIDIA drivers](./install-nvidia-driver.html#public-nvidia-driver) + * [Install NVIDIA public drivers](./public-nvidia-driver.html) @@ -25 +25 @@ To install AMD drivers on an instance with an attached AMD GPU, such as a G4ad i - * [Configure Fabric Manager for P6-B200 instance types](./install-nvidia-driver.html#nvidia-public-install-fabman) + * [Install NVIDIA GRID drivers (G6, Gr6, G6e, G6f, Gr6f, G5, G4dn, and G3 instances)](./nvidia-GRID-driver.html) @@ -27,5 +27 @@ To install AMD drivers on an instance with an attached AMD GPU, such as a G4ad i - * [Option 3: GRID drivers (G6, Gr6, G6e, G6f, Gr6f, G5, G4dn, and G3 instances)](./install-nvidia-driver.html#nvidia-GRID-driver) - - * [Option 4: NVIDIA gaming drivers (G6, G6e, G5, and G4dn instances)](./install-nvidia-driver.html#nvidia-gaming-driver) - - * [Install an additional version of CUDA](./install-nvidia-driver.html#gpu-instance-install-cuda) + * [Install NVIDIA gaming drivers (G6, G6e, G5, and G4dn instances)](./nvidia-gaming-driver.html) @@ -107,1384 +103 @@ Use one of the following options to get the NVIDIA drivers required for your GPU -###### Options - - * Option 1: AMIs with NVIDIA drivers installed - - * Option 2: Public NVIDIA drivers - - * Option 3: GRID drivers (G6, Gr6, G6e, G6f, Gr6f, G5, G4dn, and G3 instances) - - * Option 4: NVIDIA gaming drivers (G6, G6e, G5, and G4dn instances) - - - - -### Option 1: AMIs with NVIDIA drivers installed - -AWS and NVIDIA offer different Amazon Machine Images (AMIs) that come with the NVIDIA drivers installed. - - * [Marketplace offerings with the Tesla driver](https://aws.amazon.com/marketplace/search/results?page=1&filters=VendorId&VendorId=e6a5002c-6dd0-4d1e-8196-0a1d1857229b%2Cc568fe05-e33b-411c-b0ab-047218431da9&searchTerms=tesla+driver) - - * [Marketplace offerings with the GRID driver](https://aws.amazon.com/marketplace/search/results?&searchTerms=NVIDIA+quadro) - - * [Marketplace offerings with the Gaming driver](https://aws.amazon.com/marketplace/search/results?searchTerms=NVIDIA+gaming) - - - - -To review considerations that are dependent on your operating system (OS) platform, choose the tab that applies to your AMI. - -Linux - - -To update the driver version installed using one of these AMIs, you must uninstall the NVIDIA packages from your instance to avoid version conflicts. Use this command to uninstall the NVIDIA packages: - - - [ec2-user ~]$ sudo yum erase nvidia cuda - -The CUDA toolkit package has dependencies on the NVIDIA drivers. Uninstalling the NVIDIA packages erases the CUDA toolkit. You must reinstall the CUDA toolkit after installing the NVIDIA driver. - -Windows - - -If you create a custom Windows AMI using one of the AWS Marketplace offerings, the AMI must be a standardized image created with Windows Sysprep to ensure that the GRID driver works. For more information, see [Create an Amazon EC2 AMI using Windows Sysprep](./ami-create-win-sysprep.html). - -### Option 2: Public NVIDIA drivers - -If the AWS Marketplace AMIs described in Option 1: AMIs with NVIDIA drivers installed don't fit your use case, you can install the public drivers and bring your own license. To install a public driver, download it from the NVIDIA site as described in this section. - -###### Step 1: Download a public NVIDIA driver - -Log on to your instance and download the 64-bit NVIDIA driver appropriate for the instance type from [http://www.nvidia.com/Download/Find.aspx](http://www.nvidia.com/Download/Find.aspx). For **Product Type** , **Product Series** , and **Product** , use the options in the following table. - -Instance | Product type | Product series | Product | Minimum driver version ----|---|---|---|--- -G3 | Tesla | M-Class | M60 | \-- -G4dn | Tesla | T-Series | T4 | \-- -G5 | Tesla | A-Series | A10 | 470.00 or later -G5g1 | Tesla | T-Series | NVIDIA T4G | 470.82.01 or later -G6 | Tesla | L-Series | L4 | 525.0 or later -G6e | Tesla | L-Series | L40S | 535.0 or later -Gr6 | Tesla | L-Series | L4 | 525.0 or later -P2 | Tesla | K-Series | K80 | \-- -P3 | Tesla | V-Series | V100 | \-- -P4d | Tesla | A-Series | A100 | \-- -P4de | Tesla | A-Series | A100 | \-- -P5 | Tesla | H-Series | H100 | 530 or later -P5e | Tesla | H-Series | H200 | 550 or later -P5en | Tesla | H-Series | H200 | 550 or later -P6-B2002 | Tesla | HGX-Series | B200 | 570 or later -P6e-GB200 | Tesla | HGX-Series | B200 | 570 or later - -1 The operating system for G5g instances is Linux aarch64. - -2 For P6-B200 instance types, there are additional installation requirements to configure NVIDIA Fabric Manager. - -###### Step 2: Install the public driver - -Follow the install instructions for your instance type operating system. - -Linux - - - 1. To install the NVIDIA Tesla driver on a Linux distribution, see the [NVIDIA Driver Installation Guide](https://docs.nvidia.com/datacenter/tesla/driver-installation-guide/index.html). - - 2. P6-B200 instance types require installation and configuration of additional packages that come bundled with the NVIDIA CUDA Toolkit. For more information, see Configure Fabric Manager for P6-B200 instance types. - - - - -Windows - - -To install the NVIDIA driver on Windows, follow these steps: - - 1. Open the folder where you downloaded the driver and launch the installation file. Follow the instructions to install the driver and reboot your instance as required. - - 2. Disable the display adapter named **Microsoft Basic Display Adapter** that is marked with a warning icon using Device Manager. Install these Windows features: **Media Foundation** and **Quality Windows Audio Video Experience**. - -###### Important - -Don't disable the display adapter named **Microsoft Remote Display Adapter**. If **Microsoft Remote Display Adapter** is disabled your connection might be interrupted and attempts to connect to the instance after it has rebooted might fail. - - 3. Check Device Manager to verify that the GPU is working correctly. - - 4. To achieve the best performance from your GPU, complete the optimization steps in [Optimize GPU settings on Amazon EC2 instances](./optimize_gpu.html). - - - - -#### Configure Fabric Manager for P6-B200 instance types - -The P6-B200 platform is unique in that it exposes Mellanox ConnectX 7 network interface cards (NICs) to the instance as PCIe devices. These CX7 NICs do not act as typical network interfaces but instead function as NVSwitch bridges providing a control path to initialize and configure the NVFabric, which is the NVLink topology of the GPU interconnect. - -To fully initialize the system, the NVIDIA Fabric Manager must configure `NVFabric` and establish the NVSwitch topology. This enables InfiniBand kernel modules to communicate with the CX7 devices. - -Installation and configuration steps vary slightly depending on the instance operating system. Follow the instructions on the _NVIDIA Developer_ website to get the [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads) as either a local or network installer. package for your instance. After you've installed the toolkit, configure NVIDIA Fabric Manager as follows: - - 1. Install NVIDIA Link Subnet Manager and `ibstat`. - - sudo dnf install nvlink5 - - 2. Enable automatic loading of the Infiniband module on startup. - - echo "ib_umad" | sudo tee -a /etc/modules-load.d/modules.conf - - 3. Enable the NVIDIA Fabric Manager service on startup. - - sudo dnf install nvidia-fabric-manager - sudo systemctl start nvidia-fabricmanager.service - - 4. Reboot the instance. - - sudo reboot - - - - - 1. Install the `mlx5_ib` kernel module. - - sudo apt install linux-modules-extra-$(uname -r) -y - - 2. Install NVIDIA Link Subnet Manager. - - sudo apt install nvlsm - - 3. Install `ibstat`. - - sudo apt install infiniband-diags - - 4. Enable automatic loading of the Infiniband module on startup. - - echo "ib_umad" | sudo tee -a /etc/modules-load.d/modules.conf - - 5. Enable the NVIDIA Fabric Manager service on startup. - - sudo apt install nvidia-fabricmanager - sudo systemctl enable nvidia-fabricmanager - - 6. Reboot the instance. - - sudo reboot - - - -