Dear Organizers, I built and tested our docker on RTX 3060 already, but when I submit the docker images to the challenge, I received that logs. And also have error about the cuda capabilities. Tesla K80 with CUDA capability sm_37 is not compatible with the current PyTorch installation. The current PyTorch install supports CUDA capabilities sm_52 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 compute_86. How can I fix it?

Created by Quan Dung Pham pqdung
@pqdung , Many apologies for the frustrations caused. After some discussion, we have moved towards updating the infrastructure hardware to a V100 to better emulate the specs used by the participants. I am not sure if this will solve your particular issue but I am hoping that this GPU will be more compatible with your PyTorch version?
@vchung It seems like our docker image can not run with other Pytorch versions and NVCC versions. According to Nvidia, to run our docker, GPU Requirements Release 21.02 supports CUDA compute capability 6.0 and higher. This corresponds to GPUs in the Pascal, Volta, Turing, and NVIDIA Ampere GPU architecture families. Do you have any solution for our team?
Hi @pqdung , Apologies for the delay. Unfortunately, given our current license, we are able to only use the Tesla K80 GPU driver. If it helps any, the driver and CUDA versions are 470.57.02 and 11.4, respectively. If you were to use an older version of Pytorch, say 1.6, would the image still run?
Any update @TimothyBeck @vchung
Hi @trberg, Our model was based on CLARA, a product of Nvidia for training deep neural network. In the docker of Clara, they use cuda version 11.2 (a demo which optimized with CLARA) and pytorch 1.8 (with modifications). I tried to reinstall the package to pytorch 1.8 with cuda version 11.1 but it can not run anything. That is the reason why I can not do anything to make my docker run with CUDA capability sm_37. Can you update to another GPU which has the higher CUDA capability. I tested my docker on RTX 3060 and it run well. Thank you.
Hi @pqdung, At the moment, the only solution would be to use an older version of PyTorch. I'm not sure which version exactly you'd have to use, but that's been our experience so far. Apologies for the inconvenience,
@trberg Any update on how to fix it?

Tesla K80 is not support by container page is loading…