Manually install the Tesla driver on Windows - Elastic GPU Service

For general-purpose computing workloads such as deep learning and AI, a GPU requires the Tesla driver to use its high-performance computing capabilities. GPU-accelerated compute-optimized instances that run Windows do not support automatic driver installation during instance creation. You must follow the steps in this topic to manually install the Tesla driver.

Procedure

Note

This topic applies to all GPU-accelerated compute-optimized Windows instances. For more information, see GPU-accelerated compute-optimized instance families (gn, ebm, and scc series).

Step 1: Download the Tesla driver

Connect to your Windows GPU-accelerated instance.
For more information, see Connect to a Windows instance using Workbench.
On your remote desktop, visit the NVIDIA drivers download page.
Note
For more information about how to install and configure NVIDIA drivers, see NVIDIA CUDA Installation Guide for Linux.

Set the search criteria and click Search.

驱动搜索.jpg

The following table describes the filters.

Filter	Description	Example value
Product Type Product Series Product	Select the product type, series, and product that correspond to the GPU model of your instance type. Note To view details about your GPU instance, such as the instance ID, instance type, and operating system, see View instance information.	Data Center / Tesla A-Series NVIDIA A10
Operating System	Select the Windows operating system version that matches the image used by your instance.	Windows 10 64-bit
CUDA Toolkit	Select the CUDA Toolkit version.	11.2
Language	Select a language for the driver.	Chinese (Simplified)

GPU information and supported Tesla driver and CUDA Toolkit versions for different GPU-accelerated compute-optimized instance families.

Instance family	gn8is	gn7i	gn7e	gn7	gn6e	gn6i	gn6v	gn5i	gn5
Product type	Data Center / Tesla	Data Center / Tesla	Data Center / Tesla	Data Center / Tesla	Data Center / Tesla	Data Center / Tesla	Data Center / Tesla	Data Center / Tesla	Data Center / Tesla
Product series	L-Series	A-Series	A-Series	A-Series	V-Series	T-Series	V-Series	P-Series	P-Series
Recommended Tesla driver version	Version 538.67 or later	Version 462.31 or later	Version 452.77 or later		Version 426.00 or later
Recommended CUDA Toolkit version	CUDA Toolkit 12.2 Update 2	CUDA Toolkit 11.2	CUDA Toolkit 11.0 Update 1		CUDA Toolkit 10.1 Update 2

Note

The preceding table lists GPU information for only some common GPU-accelerated compute-optimized instance types. Instances that use the same GPU model, such as ebmgn6e and gn6e which both use NVIDIA V100, share the same product type, product series, and product.
When you manually install the Tesla driver and CUDA Toolkit, you must ensure the driver and toolkit versions are compatible. For more information, see CUDA Compatibility.

On the driver search results page, click Show More Versions.
Find the driver that you want to download and click View.
For example, you can select Data Center Driver for Windows with driver version 462.31 and CUDA Toolkit version 11.2.
On the driver details page, click Download.

Step 2: Install the Tesla driver

Open the folder containing the downloaded driver and double-click the installation file.
Follow the on-screen instructions to complete the installation.
After the installation, restart the GPU instance if prompted. For more information, see Restart instances.
Important
After you install the Tesla driver on Windows, the built-in Remote Desktop Protocol (RDP) does not support applications that rely on DirectX or OpenGL. To run these applications, you must install a VNC service and client or use another remote connection client that supports protocols such as PCoIP or XenDesktop HDX 3D.
In the lower-left corner of your remote desktop, type Device Manager and press Enter.
On the Device Manager page, click Display adapters to check whether the driver is installed.
If the driver information (for example, NVIDIA A10) appears, the installation was successful.

References

If you have a GPU-accelerated compute-optimized instance that runs a Linux operating system, you can install a Tesla driver for general-purpose computing workloads such as deep learning and AI. For more information, see Automatically install or load the Tesla driver when you create a GPU-accelerated instance or Manually install the Tesla driver on a GPU-accelerated compute-optimized Linux instance.
If you need to uninstall the current Tesla driver for any reason, see Uninstall Tesla drivers.
If the installed driver is unsuitable, or if an incorrect driver causes the instance to malfunction, you can either uninstall it and install a new one, or perform a direct upgrade. To upgrade a driver, see Upgrade the NVIDIA Tesla or GRID driver.