How does this skill handle limited GPU availability?

It identifies idle CPU cores across multiple accounts and creates a secondary processing pool, allowing jobs to run concurrently on CPUs with an automated 5x time multiplier to ensure completion.

What is the benefit of multi-account resource calculation?

It aggregates resource limits (QOS) from all authorized accounts, providing truly additive concurrency rather than being limited to a single account's hardware quota.

Does CPU processing affect the quality of the output?

No, quality parameters such as iterations and tolerances remain identical between GPU and CPU modes; only the compute device and execution speed differ.

How are Out-Of-Memory (OOM) kills prevented in this workflow?

The skill avoids oversubscribed burst queues, instead utilizing regular account QOS with guaranteed memory enforcement and resource allocation for every job.

Slurm Concurrent GPU+CPU Processing

Name: Slurm Concurrent GPU+CPU Processing
Author: smith6jt-cop

bysmith6jt-cop

•

数据科学与机器学习

Maximizes HPC throughput by orchestrating concurrent GPU and CPU batch processing across multiple SLURM accounts.

This skill provides a specialized framework for KINTSUGI SLURM batch processing, designed to eliminate resource bottlenecks in high-performance computing environments. By implementing a multi-account resource calculation strategy, it enables users to leverage both GPU and CPU slots simultaneously across different accounts, effectively increasing concurrency beyond limited GPU counts. It includes automated device mode detection, independent resource pools to prevent preemption, and calibrated time multipliers for CPU-bound tasks, ensuring maximum efficiency and reliable job completion without manual intervention in headless environments.

主要功能

011 GitHub stars

02Automated device mode exporting and detection

03Multi-account resource calculation for additive concurrency

04Independent GPU and CPU partition management

05Smart resource querying via sacctmgr integration

06Empirically-derived CPU time multipliers for job reliability

使用场景

01Scaling heavy image processing tasks like stitching and deconvolution on HPC clusters

02Orchestrating complex Snakemake workflows across federated SLURM accounts

03Maximizing batch throughput when hardware GPU availability is constrained

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add smith6jt-cop/skills_registry slurm-concurrent-processing

For use in Claude.ai and ChatGPT

Download Skill