TensorRT-LLM icon indicating copy to clipboard operation
TensorRT-LLM copied to clipboard

cuDDN minor version diffrences (TensorRT was linked against cuDNN 8.9.X but loaded cuDNN 8.9.X)

Open enochlev opened this issue 1 year ago • 2 comments

System Info

I got this warning when trying to compile the model and I can imagine it may reduce performance

TensorRT was linked against cuDNN 8.9.6 but loaded cuDNN 8.9.4

the problem here is there is in the read me it provides an option the change MAJOR.MINOR but not "extra_minor" version

this is how I compiled it make -C docker release_build CUDA_ARCHS="89-real" (i have a 4090)

should I not be wored about the extra_minor" version or is there a command to specify it? I am aware of this issue, but ultimately the problem may be the build command

Who can help?

No response

Information

  • [ ] The official example scripts
  • [ ] My own modified scripts

Tasks

  • [ ] An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
  • [ ] My own task or dataset (give details below)

Reproduction

make -C docker release_build CUDA_ARCHS="89-real"

compile this with a WSL2 machine having a 4090

Expected behavior

adding the third option may be nice

make -C docker release_build CUDA_ARCHS="896-real"

actual behavior

TensorRT was linked against cuDNN 8.9.6 but loaded cuDNN 8.9.4

additional notes

If thats not possible, then I would greatly appreciate it if you could point me to the solution

enochlev avatar Jan 25 '24 16:01 enochlev

just to note... I am able to compile model successfully. Just curiouse if it has a performance impact

enochlev avatar Jan 26 '24 22:01 enochlev

Could you please provide more detailed logs? BTW, I think there will be no compatibility issues between minor versions.

Shixiaowei02 avatar Feb 02 '24 09:02 Shixiaowei02