cpu-gpu-performance

What problem does it solve?

Running heavy CPU/GPU jobs without proper planning can lead to wasted resources, long execution times, and impact shared infrastructure. This skill provides a guardrail workflow to keep CPU/GPU usage lean by measuring load, narrowing scope, and choosing efficient validation strategies.

Core Features & Use Cases

• Baseline Establishment: Captures current CPU/GPU utilization and sets per-task budgets.
• Scope Narrowing: Refuses "whole world" jobs, preferring diff-based or tag-based selective testing.
• Instrumentation: Guides on choosing the right profiler (perf, nvidia-smi, nsys) to identify bottlenecks.
• Throttling & Sequencing: Uses nice, ionice, or quotas to prevent starvation and stagger heavy tasks.
• Use Case: Before running a full machine learning model training, use this skill to check current GPU utilization, narrow the scope to a small dataset for a smoke test, and instrument with nvidia-smi dmon to monitor resource usage.

Quick Start

Capture your current CPU/GPU utilization using uptime and nvidia-smi, then set a per-task budget.