Model fit · prices from 2026-09-10

The cheapest GPU that fits your model

Pick a model class and precision; get the GPU count and this week's cheapest tracked price for every accelerator — with a disclosed-capacity column for anything that spans nodes. VRAM calculators tell you the memory; nobody connects it to live prices. This does.

Method: VRAM ≈ parameters × bytes/parameter × 1.2 (20% headroom for KV cache and activations at ~8K context). Longer context or high concurrency needs more; training needs far more (optimizer states: ~4–8× weights). GPU counts assume even sharding. Above one 8-GPU node, the interconnect becomes the constraint — restrict to Grade A/B listings. Prices are this week's cheapest tracked on-demand listing — check the GPU page for every seller and its grade.