Alphafold running time

[almart7]

What happened?

Dear developers,

I tried to run alphafold from the gget package and compare the execution times to the Alphafold2 implemented in Google Colab.
I submitted the following sequence and used the default parameters in both cases: MAEHGAHFTAASVADDQPSIFEVVAQDSLMTAVRPALQHVVKVLAESNPTHYGFLWRWFDEIFTLLDLLLQQHYLSRTSASFSENFYGLKRIVMGDTHKSQRLASAGLPKQQLWKSIMFLVLLPYLKVKLEKLVSSLREEDEYSIHPPSSRWKRFYRAFLAAYPFVNMAWEGWFLVQQLRYILGKAQHHSPLLRLAGVQLGRLTVQDIQALEHKPAKASMMQQPARSVSEKINSALKKAVGGVALSLSTGLSVGVFFLQFLDWWYSSENQETIKSLTALPTPPPPVHLDYNSDSPLLPKMKTVCPLCRKTRVNDTVLA

The Colab AlphaFold finished in 15 minutes and required 4.6GB RAM and 3.6GB of GPU while gget AlphaFold finished in ~5h in a desktop computer with 8GB GPU, and 64GB RAM.

If I understood correctly, gget alphafold implements the simplified Alphafold version that also uses Google Colab. Why is there such a great difference in execution time?

gget version

0.27.2

Operating System (OS)

Linux

User interface

Command-line

Are you using a computer with an Apple M1 chip?

Not M1

What is the exact command that was run?

gget alphafold MAEHGAHFTAASVADDQPSIFEVVAQDSLMTAVRPALQHVVKVLAESNPTHYGFLWRWFDEIFTLLDLLLQQHYLSRTSASFSENFYGLKRIVMGDTHKSQRLASAGLPKQQLWKSIMFLVLLPYLKVKLEKLVSSLREEDEYSIHPPSSRWKRFYRAFLAAYPFVNMAWEGWFLVQQLRYILGKAQHHSPLLRLAGVQLGRLTVQDIQALEHKPAKASMMQQPARSVSEKINSALKKAVGGVALSLSTGLSVGVFFLQFLDWWYSSENQETIKSLTALPTPPPPVHLDYNSDSPLLPKMKTVCPLCRKTRVNDTVLA

Which output/error did you get?

No response

[lauraluebbert]

Hi Alessandra, gget alphafold probably is not accessing your computer's GPUs, and since it runs locally, the runtime is highly dependent on the speed of your internet connection. (If you run gget alphafold on Google Colab (example ) you should find more similar runtimes.)

[almart7]

Thank you for your quick answer! Do you know how I could check if gget is accessing my GPU? I already checked that the drivers and cuda are correctly installed. El mié, 1 mar 2023 a las 11:11, Laura Luebbert ***@***.***>) escribió:

…

Hi Alessandra, gget alphafold probably is not accessing your computer's GPUs, and since it runs locally, the runtime is highly dependent on the speed of your internet connection. (If you run gget alphafold on Google Colab (example <https://github.com/pachterlab/gget_examples/blob/main/gget_alphafold.ipynb> ) you should find more similar runtimes.) — Reply to this email directly, view it on GitHub <#62 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AVOPCZ3WAFVRLOGHXPAH3LTWZ4OFNANCNFSM6AAAAAAVLXUFWY> . You are receiving this because you authored the thread.Message ID: ***@***.***>

-- Alessandra

[lauraluebbert]

Hm maybe some system monitor, e.g. watch -n 1 nvidia-smi (also see: https://unix.stackexchange.com/questions/38560/gpu-usage-monitoring-cuda). I expect that gget alphafold is not recognizing your GPU (and I am not sure how to fix that), but please let me know what you find

[lauraluebbert]

Re the slower runtime: https://twitter.com/neuroluebbert/status/1555968176348729344?s=46&t=1ZZOf9kQsZAlnUIYgUDEoQ