Brian Loyal

Try running with Python 3.8 - seems like an issue with newer versions

I'm seeing the same thing in us-east-2 with the SKLearn, TensorFlow, and XGBoost estimators as well

@milot-mirdita Yeah, tried that too and got the same result. For my own sanity, have you been able to successfully run GPU search against the colabfold profile databases?

Update: The script you shared generates the same error that I saw before, both from within the `ghcr.io/soedinglab/mmseqs2:17-b804f-cuda12` container and directly on the host [docker_stdout.txt](https://github.com/user-attachments/files/18525879/docker_stdout.txt) I'll try it on a...

Good news! The script works on both the L4 and L4S, both directly on the host and from the container. Maybe there's an issue running on Ampere? My original script...

Closing this for now, but you may want to update the [wiki](https://github.com/soedinglab/MMseqs2/wiki#compile-from-source-for-linux-with-gpu-support) to strongly encourage Lovelace-gen GPUs for best results

Ok, I just tried on an A100 and it also works fine, both inside and outside the container. So, at least so far, it seems specific to an A10G. It...

One more update. I switched over to a Ubuntu AMI running CUDA 12.4 on a A10G and it worked. I've seen some [reports](https://discuss.pytorch.org/t/cuda-driver-initialization-failed-torch-cuda-is-available-false/213691) in the past of CUDA version inconsistencies...

Yep, that's right. The one that worked on A10G was 550.144.03 + CUDA 12.4

> [@brianloyal](https://github.com/brianloyal) would you mind sharing the AMI (I didn't find it on AWS cuda 12.4) and how you started the docker ultimately? Sure thing - this was the winner...