Different cuda kernels reports 'unspecified launch failure' crashes at the same time


In my application I have 2 different threads that does different types of GPU calculations.
However from time to time I have encountered a ‘unspecified launch failure’ reported in both threads in the same time.

I don’t believe that there is anything wrong with the actual kernels but I’m more starting to think that I might be using to much resources on the GPU at the same time what causing the issue OR its caused by some texture issue since we previously used textures defined in the same file as the kernel but when upgrading the code we moved to texture objects.

We have encountered this issue on GTX1080, RTX2070 and A4500

If someone have had similar experience or tips for what to check it would be great to get some feedback on this.

Thanks in advance

Thank you Robert_Crovella.

I looked at this post earlier but I was more wondering if one failed kernel can cause issues for the other kernels as well, since I get this for the other kernels running on the same GPU at the same time.

Yes, that is nature of CUDA error handling. The first kernel fault corrupts the context, and after that every CUDA runtime API call or kernel launch will report the same error. You can learn more about CUDA error handling in unit 12 of this online training series.

1 Like