Hamish Tomlinson

Hi @hezhiqiang8909 thank you for the update. @fengsidu does your machine have a CUDA driver?

Hi thanks for raising this. Can you please provide more details of the test run so this can be reproduced? On Thu, 22 Sept 2022 at 17:12, Luca Jovine ***@***.***>...

Thanks @lucajovine. Are you running this with the docker image? It should have all of the prerequisites installed. pdbfixer isn't available on pypi, so needs to be installed with conda....

Hi thanks very much for this. I will investigate and get back to you.

Hi again, I'm not seeing the same performance issues on the fasta sequence you provided. Below are the results I have obtained: ``` 2.2.3 "model_1_multimer_v2_pred_0": 0.7026557510461339, "model_2_multimer_v2_pred_0": 0.4509017402688641, "model_3_multimer_v2_pred_0": 0.380578570853769,...

Hi thanks for this. Would it be possible to reverse the experiment? Also, what cuda versions are on these machines? (run nvidia-smi)

Thanks for raising this. For now, if you are using an affected GPU we recommend pinning to previous versions of the repo. We will investigate the issue further.

Hi all. We have run AF multimer on several different GPUs and are not seeing any significant differences between predictions. @lucajovine would you mind checking to see if there are...

Thanks -- sounds like its not a data issue. Its been difficult to reproduce internally and we have tried on several different GPUs. We will continue to look into this...

Hi thanks for raising this. Have you updated the pdb_seqres file more recently than the PDB mmcif files? This error is likely to happen when they are out of sync....