Closed adi1bioinfo closed 1 year ago
Hi @adi1bioinfo
Sorry for the delayed response. I do not have SLURM set up to test this. I assume that since this is a shared cluster maybe the GPU is used by several others simultaneously, RAM and storage could also be limiting factors for slow performance. Try with CPU instead of GPU and you may have better performance. I would need more hardware related information to further comment on this.
I have installed alphafold2 using the non-docker method on HPC, I am running the script using GPU (V100 with 16 GB of memory). For a sequence of around 200 amino acids, It is taking around 8 hours for structure determination. In Alphafold2 paper (https://www.nature.com/articles/s41586-021-03819-2.pdf), it is quoted that "Representative timings for the neural network using a single model on V100 GPU are 4.8 min with 256 residues, 9.2 min with 384 residues and 18 h at 2,500 residues". I have pasted the outputed logs below. Any idea why it is running very slow in my case although I am using the same GPU?
Components of out file generated by HPC:
timings.json file output
Thank you, Aditi