gpufind Finds the Cheapest GPU That Fits Your Model

Field notes

The calculator combines model memory estimation with a catalog of provider hardware rather than ranking cards by headline specifications alone. Its example for Llama 3.3 70B at FP16 and 32k context estimates 169 GB, then filters for machines that can pool enough VRAM in one node. The page records 787 prices from 38 providers and explicitly declines to invent throughput rankings because it has measured none.