While using Triton with CUDA 12.6, I receive the following error.
E RuntimeError: Internal Triton PTX codegen error:
E ptxas /tmp/tmpqgu1f9dk.ptx, line 5; fatal : Unsupported .version 8.6; current version is '8.5'
E ptxas fatal : Ptx assembly aborted due to errors
Per the table here, both CUDA 12.5 and 12.6 use PTX ISA 8.5, making the calculation below no longer correct.
While using Triton with CUDA 12.6, I receive the following error.
Per the table here, both CUDA 12.5 and 12.6 use PTX ISA 8.5, making the calculation below no longer correct.
https://github.com/triton-lang/triton/blob/2d38ffab6c13699d5f13a1979dde772d476f9d3f/third_party/nvidia/backend/compiler.py#L45-L57