search

nanoporetech / dorado

Oxford Nanopore's Basecaller
https://nanoporetech.com/
Other
488 stars 59 forks source link

Support for older NVIDIA Chips #345

Closed onkeld closed 4 months ago

onkeld commented 1 year ago

Dear Dorado Team,

I was just wondering about your plans to support older generation NVIDIA GPUs in dorado. Reason is that the High Performance Cluster I have access to currently still runs NVIDIA Tesla P100 GPUs and there is no plan yet to upgrade the systems. Since Guppy basecaller will be replaced by dorado in the future, I'm wondering if I will still be able to basecall my data with my current hardware once Guppy is replaced.

ymcki commented 1 year ago

Pascal or before has no Tensor Core units. I believe dorado requires Tensor Core to run.

If you are short on $$$, you can run it on cloud with T4 GPUs.

tijyojwad commented 1 year ago

Hi @onkeld , we don't support P100s right now. It's a fair amount of work to add that in. Let me discuss it with the team and get back to you.

JMencius commented 10 months ago

Hi @tijyojwad I am using dual-RTX2070 for basecalling and the basecalling rate is very slow. I estimate using and 32-core EPYC CPU and dual-RTX2070 takes more than a week to finish basecalling a 100G fast5 file data. I wonder is this speed normal?

tijyojwad commented 10 months ago

Hi @JMencius - Converting fast5 format to pod5 will yield better performance.

vellamike commented 4 months ago

We will not officially support Pascal generation and older because these GPUs perform poorly compared to their modern counterparts. We understand that this is not ideal for people who have clusters with older hardware, but there are not many such clusters any longer and the engineering effort which would be required to add this support is very high.