Using TRT 8.0.1 on RTX 4070 - Unsupported SM Error

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

Description

We are getting an error when serializing a TensorRT engine using RTX 4070:

caskUtils.cpp::trtSmToCask: Error Code 1: Internal Error (Unsupported SM: 0x809)

The documentation for 8.0.1 say the card supports >SM5.0, and the card is SM8.9. But this post implies cards also have minimum supported TensorRT/CUDA versions.

What are the supported versions for the RTX 4070, and where can I find this information in the future?

Environment

TensorRT Version: 8.0.1
GPU Type: RTX 4070
Nvidia Driver Version: 525
CUDA Version: 11.4.2
CUDNN Version: 8
Operating System + Version: ubuntu20.04

4

Hi,

We recommend that you try TensorRT version 8.5 or above. Before version 8.5, SM 8.9 was not supported.
Please refer to the release notes for details.

Thank you.

1 Like
5

Thanks for your quick response!

I’ve read the release notes, but I cannot see anywhere that explains forward compatibility of TRT with the latest GPUs.

For example, will TRT8.5 be supported on the next gen 50-series cards? Does a new SM version always require a new release of TRT?

This is important to us as this affects the compatibility of our product releases.

Thanks,
Magnus

6

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.