g4dn EC2 refusing to use T4

Question

set up the instance, installed the nvidia drivers, installed dcv, on connect it's using some amazon display adapter (aws indirect display device according to dxdiag), and not the nvidia one. Windows Server 2025 Nvidia smi output:

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
+-----------------------------------------------------------------------------------------+

Seems fine. but how can I set the T4 to the default used one?

MikeLim · Answer

From your nviidia-smi output, your driver is running in TCC and not WDDM mode. TCC mode does not use GPU for graphics acceleration. For example, your Process List is empty

Do uninstall and install NVIDIA GRID (aka NVIDIA RTX Virtual Workstation) driver as per documentation. Or you can refer to this blog for installation using CloudFormation. Example screenshot of nvidia-smi and nvidia-smi -q | findstr "Product License" below

Enter image description here

Do uninstall and install NVIDIA GRID (aka NVIDIA RTX Virtual Workstation) driver as per [documentation](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/install-nvidia-driver.html#nvidia-GRID-driver). Or you can refer to this [blog](https://aws.amazon.com/blogs/compute/building-a-high-performance-windows-workstation-on-aws-for-graphics-intensive-applications/) for installation using CloudFormation. 
Example screenshot of `nvidia-smi` and `nvidia-smi -q | findstr "Product License"` below

![Enter image description here](/media/postImages/original/IMcOKvq1VGTcq0z1QFeQwc-w)

re:Post Agent · Answer

The issue you're experiencing is that your g4dn instance has the NVIDIA T4 GPU properly installed (as shown by your nvidia-smi output), but it's not being used for display rendering in your DCV session.

This is actually expected behavior. The NVIDIA T4 GPU on g4dn instances is configured in TCC (Tesla Compute Cluster) mode as shown in your nvidia-smi output, which optimizes it for compute workloads rather than graphics display. The AWS indirect display device is handling the display rendering while the T4 is available for compute tasks.

To use the GPU for graphics acceleration in your DCV session, you need to:

1. Make sure you have the DCV-GL package installed (nice-dcv-gl)
2. Ensure the NVIDIA driver is properly configured for DCV
3. Run the following commands to enable GPU acceleration for DCV:
- Stop the X server
- Run `dcvgladmin disable`
- Run `dcvgladmin enable`
- Restart the X server

You can verify if OpenGL acceleration is working by running the `dcvgltest` tool from the nice-dcv-gltest package.

If you're using a DLAMI (Deep Learning AMI), make sure you're using the correct one for your g4dn instance. AWS provides specific DLAMIs that use the NVIDIA OSS driver which supports G4dn instances.  
**Sources**  
[Issues gdscheck -p - receiving: CUDA_ERROR_SYSTEM_NOT_READY | AWS re:Post](https://repost.aws/questions/QUMUqas3dpR_mf3tb6NFoFSQ/issues-gdscheck-p-receiving-cuda-error-system-not-ready)  
[DCV cannot access the 3D X Server [:0.0 :0.1]. | AWS re:Post](https://repost.aws/questions/QU50McuwM_S8Gor8akifvRPw/dcv-cannot-access-the-3d-x-server-0-0-0-1)  
[Specifications for Amazon EC2 accelerated computing instances - Amazon EC2](https://docs.aws.amazon.com/ec2/latest/instancetypes/ac.html)  
[Important NVIDIA driver changes to DLAMIs - AWS Deep Learning AMIs](https://docs.aws.amazon.com/dlami/latest/devguide/important-changes.html)

g4dn EC2 refusing to use T4

添加回答

相关内容