Methodology

How prices are collected

Once a day we read each provider's public pricing page (and, for Vast.ai, its public marketplace API). We respect robots.txt and identify ourselves as GPURentWatchBot/0.2 (+https://gpurentwatch.com/methodology/; [email protected]). Each provider has its own parser; prices are converted to US dollars per GPU per hour (multi-GPU instance prices are divided by the GPU count).

How prices are filtered

• We list the cheapest price per GPU model and price type (on-demand, spot, reserved, marketplace) for each provider.
• Partial or shared GPUs (MIG slices, fractional GPUs) are labelled and never count as the cheapest price.
• Vast.ai is a marketplace: we show the cheapest rentable whole-GPU offer, and only list a GPU when at least 3 different hosts offer it.
• Prices far below or above other providers, or that move more than 25% in a day, are flagged and held back from "cheapest" until reviewed.
• If a provider's page can't be read on a given day, we keep showing its last good prices for up to 7 days, marked "stale since DATE".
• Price trends compare only providers present on both dates, so a provider dropping in or out doesn't create fake moves.

Sponsorship policy

Sponsorship never affects sort order. Tables are always sorted by price, and paid placements are clearly labelled as sponsored. Sponsors cannot change, hide or delay prices.

Spotted a wrong price? Email [email protected]. Always confirm on the provider's site before renting.