Closed mruffalo closed 3 years ago
I made some (preliminary, untested) progress in https://github.com/hubmapconsortium/codex-pipeline/tree/mruffalo/split-cpu-gpu-pipelines -- this probably needs some work and I haven't tried to script what's required to call each sub-workflow independently in sequence.
This pipeline now contains a mix of CPU and GPU processing, and some of the CPU-oriented processing is memory-intensive also. All pipeline steps are executed from one top-level workflow, and all steps must be executed on a GPU node because of the Cytokit step.
Split this into three "top-level" workflows: