Minimum wager
per GPU-hour.
On-demand reference
Loading.
Fetching on-demand rates.
Unit
USD per GPU-hour
GPU
Sources & method
Loading sources.
Observed index
Formed
Calculation
Rates are USD per GPU-hour, displayed to 6 decimals. The geometric mean uses unrounded provider means; rounded inputs may differ in the last digit.
Equal provider weight
Each provider’s price is the arithmetic mean of its distinct qualifying regional rates, including AWS—not its cheapest region. The index is the geometric mean of those prices: equal logarithmic weights of 1/5 on-demand or 1/4 spot, not an arithmetic sum. More regions never increase a provider’s weight.
Scope & limits
All 5 on-demand or all 4 spot providers are required; never a reduced basket. Standard U.S. regions where separately quoted (GovCloud, Local Zones and non-U.S. regional quotes excluded); broader or unspecified public rates stay labeled, not verified U.S. quotes or stock. Not a listing median or market-share average.
Rate normalization
Node rates divide by GPU count (including bundled CPU/RAM); unit rates apply directly. Excludes tax, discounts, commitments, and extras. Checks confirm prices, not stock.