Home / Value
Explicit formulas · no overall winner
Local AI GPU value explorer
Pick one objective at a time. RigForAI does not collapse VRAM, price, model coverage, and speed into a single score. Current prices are the lowest fresh Amazon-matched board-partner SKU in this catalog — not MSRP and not the entire US market.
Ranking modes
- Cheapest full-VRAM fit — requires a model. MIN(fresh price) WHERE FITS_IN_VRAM.
- Most advertised VRAM — MAX(VRAM) among GPUs with a fresh offer, optional budget.
- Most full-fit supported models — COUNT of catalog models with FITS_IN_VRAM at the selected quantization and context.
- Lowest current $/GB — fresh price / advertised VRAM GB. Not a value score.
- llama.cpp decode per $1,000 — only when the same workload is published at MEDIUM or HIGH confidence for a selected model. Formula: decode tok/s / price × 1000.
Single-GPU rankings only. For 2–4 identical cards see multi-GPU feasibility.
Indexable budget landings: under $1,000 · under $1,500 · under $2,000 · under $500
The landing is indexable. Applying filters creates a query state that is not indexed. Defaults are Q4 and 8K when you do not change them.