aqlaboratory / openfold

Trainable, memory-efficient, and GPU-friendly PyTorch reproduction of AlphaFold 2
Apache License 2.0
2.72k stars 511 forks source link

Hardware requirement #127

Open zyc573823770 opened 2 years ago

zyc573823770 commented 2 years ago

Such a good work!!!! I am so interested to deploy your work. Could you please list the requirements of your hardware such as: GPU: 32G V100 * 1 CPU: 12 cores xxxx Intel RAM: 32GB Disk: 4TB SSD

Any recommended or minimum configurations will help!!

gahdritz commented 2 years ago

We trained OF on 44 A100s with 8 CPU cores per GPU and lots of RAM (idk how much it turned out to be exactly; it was hundreds of GB). From start to finish, training took us ~60k A100 hours. However; respectable performance can be obtained even after just a couple of thousand (<5k) (see Mohammed's tweet on the OF release for more details).