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

sue-sandbox-quota-calculation

Convert HPC quota units and validate Slurm job consumption across LUMI and Snellius.

Updated Aug 27, 2026
Search Tags:#resource planning#scale-up experiments#hpc quota#lumi billing#snellius sbu#gpu accounting#slurm validation
One-click install
npx skills add https://github.com/dongzhuoyao/deepresearch --skill sue-sandbox-quota-calculation
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: sue-sandbox-quota-calculation
Source: https://github.com/dongzhuoyao/deepresearch/tree/main/.codex/skills/sue-sandbox-quota-calculation
Command: npx skills add https://github.com/dongzhuoyao/deepresearch --skill sue-sandbox-quota-calculation

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Users planning or auditing SUE scale-up experiments frequently mix up different HPC quota units (LUMI billing GPU-hours, software GCD-hours, Snellius SBUs), leading to incorrect resource estimates, failed job submissions, and wasted compute budget.

Core Features & Use Cases

  • Precise Unit Conversion: Convert between LUMI billing GPU-hours, software GCD-hours, and Snellius partition-specific SBUs for A100, H100, MIG, and Vis GPU nodes.
  • Live Quota Comparison: Query real-time remaining allocation across LUMI and Snellius to identify which sandbox has sufficient quota for upcoming experiments.
  • Job Validation: Cross-check Slurm job records to confirm quota consumption matches expected calculations for completed runs.
  • Use Case: If you are planning a 64-node DINOv3 training run, use this skill to calculate exact quota requirements, compare remaining budget across LUMI and Snellius, and validate past job billing to avoid overallocation.

Quick Start

Use the sue-sandbox-quota-calculation skill to calculate the total LUMI billing GPU-hours needed for a 32-node standard-g training run that will run for 10 wall-clock hours, and compare the remaining quota against your Snellius A100 allocation.

Frequently Asked Questions about sue-sandbox-quota-calculation

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

FAQPage Schema
How do I convert LUMI billing GPU-hours to Snellius SBUs for scale-up experiment planning?▼

Converting LUMI billing GPU-hours to Snellius SBUs requires explicit unit labeling and partition-specific SBU conversion for GPU types like A100, H100, MIG, and Vis nodes to eliminate mixed-unit errors.

What is the best way to calculate HPC quota for a multi-node LUMI standard-g training run?▼

Calculating HPC quota for full-node standard-g jobs requires multiplying the node count by wall-clock hours to get exact LUMI billing GPU-hours, preventing overallocation and failed job submissions.

Can I compare remaining compute quota across LUMI and Snellius sandboxes before submitting a Slurm job?▼

Comparing remaining compute quota across LUMI and Snellius sandboxes is supported through live remaining allocation queries, identifying which platform has sufficient budget for upcoming experiments.

How do I validate Slurm job consumption against expected HPC billing calculations?▼

Validating Slurm job consumption involves cross-checking completed job records to confirm that the actual quota usage matches your expected LUMI billing GPU-hours or Snellius SBU calculations.

Why do my Snellius SBU calculations fail when mixing software GCD-hours with HPC billing units?▼

Snellius SBU calculations fail when mixing units because software GCD-hours, LUMI billing GPU-hours, and Snellius SBUs use different scaling factors, requiring explicit unit labeling and partition-specific conversions.

Do I need official HPC documentation to source billing rules for accurate SUE sandbox quota calculation?▼

Sourcing billing rules from official HPC documentation is strictly enforced for accurate SUE sandbox quota calculation, ensuring partition-specific SBU conversions and LUMI billing rates are correctly applied.

Related Skills

Explore More Education & Research Skills→
ovachieverovachiever
52

ray-data

Process distributed ML data pipelines with Ray Data across CPU/GPU clusters.

Community
Advanced
psd401psd401
2

meta-experiment

Design, run, and interpret controlled experiments to validate software changes.

Official
Advanced
niznik-devniznik-dev
16

Run Experiment

Orchestrate SLURM fine-tuning jobs with torchtune and evaluate with inspect-ai.

Community
Advanced
vecearvecear

capacity-planning-helper

Forecast infrastructure needs and estimate compute, database, and cache resources.

Community
Advanced