GPUs with Snakemake and Nextflow

GPUs can significantly accelerate some pipelines, but their scarce availability requires architecting around low availability. FlowDeploy includes a "low availability" assumption for GPUs to help fix that issue, but you can work with the team to create a plan that works best for your workload.

GPU and spot/preemptible instances

For GPU instances, there's almost no spot/preemptible availability at the moment. Because of this, requesting a GPU overrides preemptible if both are requested.

Usage

FlowDeploy supports GPUs through both Nextflow and Snakemake, but you'll have to enable GPU instances first. Once GPU instances are enabled, you can mark workflow steps as requiring GPUs.

Snakemake

Basic usage

Add the gpus directive to the rule's resources:

rule RULE_NAME:
  resources:
    gpus=1
  ...

Nextflow

Basic usage

Add the accelerator directive to the process definition:

process {
  accelerator 1
  ...
}