Inquiry on 5090 GPU Support in Triton Inference Server

[Weihong-Liu]

Description:
I encountered the following error while deploying a model on a 5090 GPU:

Invalid argument: instance group face_detect_fp16_0 of model face_detect_fp16 specifies invalid or unsupported gpu id 0. GPUs with at least the minimum required CUDA compute compatibility of 6.000000 are ...

Is the Triton Inference Server currently supporting the 5090 GPU? If not, could you please let me know if there are any plans to support it in the near future?

Triton Information:

• Triton Version: Please confirm the recommended version.
• Deployment Environment: Using the official container.

To Reproduce:

• Model Name: face_detect_fp16
• Model Type: FP16
• GPU Model: NVIDIA 5090
• Configuration: instance group face_detect_fp16_0, gpu id 0

Expected Behavior:
The 5090 GPU should be correctly recognized and supported for model deployment without the aforementioned error.

[whoisj]

This seem like it could be a driver or missing CUDA update issue.

Please provide OS information as well the output of nvidia-smi.

[Weihong-Liu]

This seem like it could be a driver or missing CUDA update issue.

Please provide OS information as well the output of nvidia-smi.

Thanks for your response!
I’m running Triton inside a Docker container using the official image:

nvcr.io/nvidia/tritonserver:25.09-py3

I’ll attach screenshots showing the following details shortly:
•Host OS information

•nvidia-smi output

•nvcc --version output inside the container

Thanks!