hardware-calculator

name	hardware-calculator
description	Quick VRAM/RAM calculations, hardware recommendations, feasibility checks for AI models
short_desc	VRAM/RAM calc + GPU sizing for AI models
keywords	["VRAM","GPU memory","RAM requirements","model footprint","GPU recommendation","which GPU","GPU sizing","fit on GPU","memory requirements","H100","A100","RTX 4090","consumer GPU"]
model	haiku

Hardware Calculator (Haiku)

Purpose: Quick VRAM/RAM calculations, hardware recommendations, feasibility checks for AI models.

Model: Haiku 4.5

When to Invoke Autonomously

1. "Can I run X?": User asks if model fits hardware
2. Hardware Shopping: "Which GPU for [model]?"
3. Quick VRAM Check: Before loading model
4. Multi-Model Planning: "Can I run 2 models simultaneously?"
5. Quantization Math: "How much VRAM saves Q4 vs Q8?"

DO NOT invoke for

• Complex architecture decisions (use /architect skill)
• Performance optimization (use /performance-optimizer)
• Already know hardware fits

What This Skill Does

VRAM Calculations:

• Calculate model VRAM requirements from parameters + quantization
• Account for context overhead and batch size
• Apply 20% safety margin
• Formula: (params × bytes_per_param × 1.2) + context + batch

GPU Recommendations:

• Match budget to appropriate GPU tier ($300-5000+)
• Match VRAM needs to GPU options (8-80GB range)
• Consider price/performance tradeoffs
• Warn about overspending or underpowered options

Feasibility Checks:

• Quick yes/no: Will model fit on GPU?
• Account for OS overhead (~2GB)
• Warn if <10% headroom (tight fit, unstable)

Multi-Model Planning:

• Calculate combined VRAM for running multiple models
• Suggest offloading strategies if tight
• Account for shared context when applicable

See: examples/calculations.md for formulas, examples/gpu-specs.md for hardware details, scripts/vram_calculator.py for automated calculations

Quick Workflow Reference

Before calculating: Search for hardware specs and benchmarks

.claude/scripts/kg-search search "hardware" --type hardware

For deep research: Ask user "Use hybrid_search to research [GPU comparison]"

Development env: Python 3.12, Weaviate:8081, Ollama:11435, venv: source claude_mcp_servers/.venv/bin/activate

Success Metrics

• ✅ VRAM estimates within ±1GB of actual usage
• ✅ GPU recommendations fit user's budget and needs
• ✅ Calculations complete in <2 seconds
• ✅ Users don't run into OOM errors after following recommendations
• ✅ Hardware purchases are successful (not over/under-powered)