AWS Graphic Drivers Help

Question

Hi,We've been attempting to install and use the NVIDIA public driver (as here: https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/install-nvidia-driver.html#preinstalled-nvidia-driver) on our G5xlarge instance running Windows 11 in order to visualize a 3D model created using downloaded software (RealityCapture). The NVIDIA driver we chose is the following: Data Center/Tesla -->A-Series --> NVIDIA A10 --> Windows 11 --> Any CUDA Toolkit Version. While the driver does seem to be recognised as being installed when checking the drivers on Windows' Device Management console, the NVIDIA app shows a loading screen which never loads. The CMD window shows no load going through the GPU when we enter Nvidia-SMI and the task manager does not give GPU statistics. I have also attempted to test the GPU using https://www.ocbase.com/ to understand if the GPU is actually being utilized which showed the same statistics as CMD.I repeated the above tests with the following instance: Data Center/Tesla -->A-Series --> NVIDIA A10 --> G5.16xlarge --> Windows 10 --> Any CUDA Toolkit Version. In case it is relevant, the latter instance was launched from a custom AMI with an associated snapshot which we created ourselves. I haven't yet attempted to use GRID drivers. I also haven't attempted to use an AMI with the drivers pre-installed because of wanting to avoid any additional charges from AMI subscriptions, but if such an AMI does exist and is truly free, I would be grateful if anyone could point me to that option.When trying to run GPU intensive software such as Reality capture, we are experiencing extreme software slowdown and the PC is not able to visualise the model despite computing it quickly. Is there something we are doing wrong in our work flow causing no load to pass through the GPU? From our research a G5 instance should be suitable for reality Capture.

MikeLim · Answer

The NVIDIA public driver (aka Tesla driver) that you downloaded is designed for **compute** and not **graphics** workloads. The driver is not being used by your applications for hardware graphics acceleration.  More info at [Types of NVIDIA drivers](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/install-nvidia-driver.html#nvidia-driver-types).

You need to install GRID driver which is licensed for use in your G5 EC2 instance. Refer to [documentation](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/install-nvidia-driver.html#nvidia-GRID-driver) for installation instructions.

Alternatively, you can use the CloudFormation template mentioned on blog [Building a high-performance Windows workstation on AWS for graphics intensive applications](https://aws.amazon.com/blogs/compute/building-a-high-performance-windows-workstation-on-aws-for-graphics-intensive-applications/) to create your EC2 instance.

Below screenshot is G5 with GRID driver on Windows Server 2022 created using mentioned [CloudFormation template](https://github.com/aws-samples/amazon-ec2-nice-dcv-samples/tree/main/cfn). Notice `nvidia-smi` lists applications using NVIDIA GPU for graphics as indicated by G in type.

![G5](/media/postImages/original/IMI1OJNY5NQSqxGkiUXsmWkg)

re:Post Agent · Answer

Based on your description, it appears you're encountering issues with NVIDIA driver installation and GPU utilization on your G5 instances running Windows. Here are some suggestions to address your concerns:

1. Driver Selection: For G5 instances, which use NVIDIA A10G GPUs, you should use GRID drivers instead of public NVIDIA drivers. GRID drivers are specifically designed for virtualized GPU environments like AWS EC2 and provide better compatibility and performance.

2. GRID Driver Installation: To install GRID drivers on Windows:
   - Connect to your Windows instance
   - Download the GRID driver from AWS S3 (the exact link depends on your Windows version)
   - Extract the downloaded zip file
   - Install the driver using the included setup application

3. AMI Options: There are AWS-provided AMIs with pre-installed NVIDIA GRID drivers that don't incur additional charges. These can simplify the setup process and ensure compatibility.

4. Verification: After installation, you can verify GPU functionality by:
   - Checking Device Manager for "NVIDIA GRID A10-8Q" under Display adapters
   - Running nvidia-smi in Command Prompt
   - Using GPU-Z or other diagnostic tools

5. Software Configuration: Ensure that RealityCapture and other GPU-intensive applications are configured to use the GPU. Some applications may need manual configuration to utilize CUDA or OpenCL.

6. Instance Type: G5.xlarge has only one GPU, which might not be sufficient for very demanding 3D modeling tasks. Consider using a larger instance type like G5.4xlarge or G5.8xlarge for more GPU power if needed.

7. Windows Version: While both Windows 10 and Windows 11 are supported, ensure you're using the latest Windows builds and have all updates installed.

If you're still experiencing issues after trying these steps, it may be helpful to review the EC2 instance logs and NVIDIA driver logs for any error messages or conflicts. Additionally, testing with a clean instance (not from a custom AMI) could help isolate any potential configuration issues.  
**Sources**  
[NVIDIA drivers for your Amazon EC2 instance - Amazon Elastic Compute Cloud](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/install-nvidia-driver.html)  
[AMD drivers for your EC2 instance - Amazon Elastic Compute Cloud](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/install-amd-driver.html)  
[Running Windows on a g6e instance | AWS re:Post](https://repost.aws/questions/QUBpXpVYgqTNSe_zdCkTzI3Q/running-windows-on-a-g6e-instance)

AWS Graphic Drivers Help

Add your answer

Relevant content