ANSWERS · COMPUTE COST ANSWERS
What does rented GPU capacity actually cost: six answers you can quote
Every answer is a number, and behind every number is a formula we show in full. Prices come from our public observations of real listed machines (latest quote per machine); compute requirements use published baselines and the 6ND formula — labelled separately on every page, because one is a fact and the other is an assumption.
Data as of 2026-11-05 (UTC) · every figure updates as the crawler runs
What does an 8-GPU H100 server cost per month?
8-GPU month $13,948
Where is an RTX 4090 cheapest to rent?
global median $0.458/h
A100 or H100: which is cheaper per unit of compute?
median gap 2.83x
What does Llama-3 70B inference cost per million tokens?
about $2.95/M tokens
What does it cost to fine-tune a 7B model?
about $1.57/run
How many H100s and how much money to train a 70B model?
~412k GPU-hours
Caliber, upfront: Every price here is an asking price, not a transaction price, and an L0 listing fingerprint — consistent across snapshots, but not evidence for collateral or underwriting. Compute requirements (GPU-hours, throughput, MFU) are published-baseline assumptions, not measurements by us. We only measure prices and machine counts. The split matters: it tells anyone quoting this page which half is a fact and which half must travel with its assumption.