skills.rest
dongzhuoyao
dongzhuoyaoCOMMUNITY
Explore all skills by this author →
Website𝕏@vtaohuGitHub

snellius-supercomputer

Submit and manage GPU compute jobs on the SURF Snellius supercomputer.

10|1|Updated Feb 21, 2026
Search Tags:#gpu#conda#hpc#scratch#slurm#sbatch#snellius
One-click install
npx skills add https://github.com/dongzhuoyao/tao-research-skills --skill snellius-supercomputer
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: snellius-supercomputer
Source: https://github.com/dongzhuoyao/tao-research-skills/tree/main/snellius-supercomputer
Command: npx skills add https://github.com/dongzhuoyao/tao-research-skills --skill snellius-supercomputer

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Provides concise, operational guidance to submit and troubleshoot GPU and CPU workloads on the SURF Snellius supercomputer, reducing time lost to environment misconfiguration, incorrect SLURM flags, and storage quota surprises.

Core Features & Use Cases

  • Environment setup: Clear patterns for conda and venv initialization and when to use environment modules to ensure CUDA and Python compatibility.
  • Job submission & partitioning: Partition selection guidance, single-node sbatch templates, interactive srun examples, and account/quotas reminders to avoid rejected jobs.
  • Monitoring & triage: Common monitoring commands, failure causes (CUDA availability, OOM, permission issues), and recommended fixes to recover or resubmit jobs.
  • Use case: Prepare a reproducible training run by copying data to scratch, activating a conda environment, selecting the gpu_h100 partition for large models, and submitting an sbatch script with the correct --account.

Quick Start

Connect to snellius.surf.nl, load the appropriate module year and CUDA module, source conda or activate your venv, and submit your training script with sbatch using the correct --account and partition.

Frequently Asked Questions about snellius-supercomputer

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I submit a GPU job on the Snellius supercomputer using sbatch?▼

To submit a GPU job on Snellius, select the correct partition like gpu_h100, specify your --account, and run sbatch with a script that initializes conda or venv and requests GPU resources through SLURM directives.

Why does my Snellius GPU job fail to find CUDA?▼

CUDA availability errors on Snellius usually stem from missing environment modules or incorrect conda activation, so load the appropriate module year and CUDA module before sourcing your environment to resolve the conflict.

What is the best way to set up a conda environment for HPC training on Snellius?▼

The best way to set up conda on Snellius is loading the appropriate module year, initializing conda or activating your venv, and ensuring Python and CUDA compatibility through environment modules before submitting workloads.

Can I run interactive GPU sessions on Snellius using srun?▼

Yes, you can run interactive sessions on Snellius using srun examples that request GPU partitions, specify your account, and allocate resources properly to avoid rejected jobs during testing or development.

How do I fix out of memory or disk quota errors during GPU compute jobs on Snellius?▼

To fix out of memory or quota errors on Snellius, check disk quotas, copy training data to scratch filesystems, and adjust SLURM memory directives to match your workload requirements before resubmitting.

When should I use the gpu_h100 partition on Snellius?▼

You should use the gpu_h100 partition on Snellius when preparing reproducible training runs for large models, ensuring you select appropriate SLURM resource directives and account parameters for the workload.

Related Skills

Explore More Education & Research Skills→
jmaglyjmagly
175

venv-manager

Create and validate Python virtual environments with dependency installation.

Community
Intermediate
KwongFukKwongFuk

slurm-self-healing-watchdog

Monitor Slurm sbatch jobs via sacct and squeue, apply minimal repairs, and resubmit failures.

Community
Advanced
yo-stevenyo-steven

multi-node-slurm

Convert single-node PyTorch distributed launches into multi-node Slurm sbatch jobs.

Community
Advanced
JKhyroJKhyro

modal-serverless-gpu

Deploy machine learning workloads on Modal serverless GPU compute.

Community
Advanced