search

WGLab / DeepMod

DeepMod: a deep-learning tool for genomic-scale, strand-sensitive and single-nucleotide based detection of DNA modifications
Other
99 stars 35 forks source link

GPU recommendations? #51

Open chuanli11 opened 2 years ago

chuanli11 commented 2 years ago

Hi, authors

Thanks for open-source the code. Just curious what is the GPU recommendation (e.g. minimal memory size) for running DeepMod?

Here says GPU can speed up detection.

But Here suggests GPU won't benefit if there is IO bottleneck. Can you please clarify how to avoid such bottleneck?

Thank you.

liuqianhn commented 2 years ago

@chuanli11 Thank you for pointing this out. If there is no IO bottleneck, you can speed up the detection with the GPU. Unfortunately, whether there is an IO bottleneck depends on the design of your storage system. If the storage system has lower bandwidth, you will have IO bottleneck. I wish I could help to avoid this type of bottleneck, but only your admin can help this.

But the way, since DeepMod do NOT improve the released trained models, based on our recent test, the trained models does NOT work on the data basecalled by albacore v2 and guppy. We will release more information after the test by the lab members.