CUDA Errror: no kernel image is available

[AlexPreviero]

I was testing the (Meituan/YOLOv6 ) repo on the AGX Orin Dev Kit with Jetpack 5.0.2 (35.1.0), with the Pytorch-L4T containers found here: NVIDIA L4T PyTorch | NVIDIA NGC.

None of the 3 supported containers for Jetpack 5.0.2 worked due to the error:

RuntimeError: CUDA error: no kernel image is available for execution on the device
CUDA kernel errors might be asynchronously reported at some other API call,so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.

I found this answer by Dusty: CUDA error: no kernel image is available for execution on the device - #5 by i_love_nvidia, which fixed it, and I could run the Pytorch container from Docker Hub dustynv/l4t-pytorch:r35.1.0-pth1.13-py3.

Will there be a similar patch to the NVIDIA L4T Pytorch containers? These are widely advertised, unlike the Pytorch container from Dusty.

Thank you

[AastaLLL]

Hi,

Confirmed that we can reproduce this issue error with the NGC l4t-pytorch container.
We are double-checking with our internal team to see if we can include the patch in the future release.

Thanks for the feedback.

[dusty_nv]

Hi @AlexPreviero, yes the patch from CUDA error: no kernel image is available for execution on the device - #8 by dusty_nv will be included in the next update of l4t-pytorch container on NGC. For now please use the mirror on DockerHub that you pointed to above.