CUDA and cuDNN version mismatch HelixFold3

PaddlePaddle / PaddleHelix

Bio-Computing Platform Featuring Large-Scale Representation Learning and Multi-Task Deep Learning “螺旋桨”生物计算工具集

Other

1.02k stars 225 forks source link

CUDA and cuDNN version mismatch HelixFold3 #341

Closed ahmedselim2017 closed 2 months ago

ahmedselim2017 commented 2 months ago

Hi, the environment part of the readme for the HelixFold3 suggests that CUDA 12.0 should be installed alongside cuDNN 8.40. However, the cuDNN archive of Nvidia (https://developer.nvidia.com/rdp/cudnn-archive) has cuDNN 8.4.0 only for CUDA 10.2 and 11.x. The first cuDNN release that supports CUDA 12 is cuDNN 8.8. Should the CUDA version that supports cuDNN 8.4.0 or the cuDNN version that supports CUDA 12 be used?

RyanGarciaLI commented 2 months ago

Hi, sorry for the unclear content in the README. You may choose CUDA 12 with cuDNN 8.9 or CUDA 11.x with cuDNN 8.4. You can select any combination of CUDA/cuDNN versions that Paddle 2.6.1 supports. Thanks.

ahmedselim2017 commented 2 months ago

Thank you very much!