We are trying to run nf-core-scrnaseq pipeline on OCI’s GPU compute but it seems entire pipeline only running on CPUs. Any thoughts?
Hi @Deepak_Khopade, sorry we haven’t been able to get you an answer yet. Were you able to solve the problem? If not, we’ll need you to share more details about your command and configuration. Did you use -profile gpu in your command? You also need to use the appropriate clusterOptions for your job scheduler.
You can try asking Seqera AI to provide specific recommendations; be sure to give it as much information as possible about the infrastructure you’re using. Seqera AI Chat
The only process in nf-core/scrnaseq which supports GPUs is cellbender. It should be pretty straightforward to configure cellbender to use GPUs by adding the profile gpu to your profiles (e.g. -profile docker,gpu) and making sure your containers can use the GPU hardware (the containerOptions Geraldine refers to above).
Support for GPUs in bioinformatics is sparse, largely because the development costs are still relatively high compared to the benefits. CPUs are basically all the same in every university all over the world, so why write specialized code that might not work. Furthermore, any new GPU enabled tool needs to prove it can beat the legacy code in performance and cost to encourage people to adopt it. And those costs aren’t just hardware running, it’s also set up and CUDA management which is not insignificant.
With the increasing adoption of GPUs and improved developer tooling, I expect to see newer bioinformatics tools offer a GPU mode. For nf-core/scrnaseq specifically, you could reach out to the authors to request this feature, or, if you’re interested, contribute an implementation yourself; the community would appreciate it!