USD per GPU-hour
Is your GPU quote unusually high or low?
Historical snapshot through 2026-11-19 UTC · not a live availability check
Compare hourly quotes for the same GPU, provider and tier.
This checks price history, not whether a rental is a good fit. Location, host quality, CPU/RAM, storage, network and exact instance configuration may differ.
This difference is a hypothetical comparison, not savings you can necessarily obtain. Past quotes do not promise a future price or available capacity.
Historical quotes
- Median recorded quote
- Middle half (P25–P75)
- Recorded minimum–maximum
Seen in a check does not mean uptime, availability or a successful rental. A provider-level success can hide a failed request for an individual GPU. Missing rows are not treated as zero-price hours.
Methodology
The reference is the Hostinger collector's hourly observations, not the separate daily snapshots on the trends page. Each series keeps GPU, provider and tier separate. Windows end at midnight UTC; today's partial day is excluded.
- RunPod secure/community and Verda on-demand/spot: these are provider API quotes, not completed transactions. Normalization can collapse multiple instance configurations. Spot prices do not account for interruption costs.
- Percentiles: exact nearest-rank quantiles of observed hourly prices, with each observation weighted equally. The median is P50. P25–P75 is the middle half of observations, not a confidence interval. Tied prices can collapse this range. No arbitrary quote percentile is inferred from the summary points.
- Missingness: prices are not zero-filled or interpolated. Catalogue changes, filters, absent offers and partial per-GPU request failures may remove rows. A successful provider check does not prove every GPU request succeeded. Observed-only prices can therefore be biased; observation frequency is not availability.
To avoid a strong label on weak evidence, the relative-price verdict is withheld when there are fewer than 168 observations, presence below 75% of successful checks, fewer than 90% of expected checks succeeded, no observation in the final 48 hours, or the window ended more than 72 hours ago. These are conservative editorial gates, not statistical guarantees. Historical values remain visible.
Before renting, confirm GPU count and memory, location, CPU/RAM, bandwidth and storage charges, interruption policy, minimum billing and current capacity with the provider. Compare current GPU listings →
Snapshot generated 2026-11-20T12:00:58.268Z · underlying completed windows shown above. This page does not claim continuous refresh.