GPUPerHour solves a narrow but expensive problem: cloud GPU pricing is scattered across dozens of provider dashboards, in different units and with different availability. The site periodically scrapes the public pricing and inventory endpoints of 31 cloud GPU providers, normalises each listing to a price per GPU per hour (dividing a multi-GPU instance price by its GPU count) and tags it with its pricing type and security tier. The current-price summary deliberately uses only secure on-demand offers observed within roughly 15 minutes of the last refresh, excluding spot, reserved and peer-to-peer marketplace listings because those prices are not directly comparable.
The result is a single live table covering H100 SXM5 and PCIe, H200, B200, B300, A100 80GB and 40GB, L40S, RTX PRO 6000, RTX 5090, RTX 4090 and RTX A6000, alongside the provider, a 30-day historical catalog average and the number of providers with stock. A daily price index with CSV downloads lets you track how pricing moves over time, while buying guides and hardware explainers cover practical topics such as comparing like-for-like GPUs, choosing by VRAM for a workload, accounting for storage and egress, and renting by the hour. The site also points to a companion console, DeployGPU, for provisioning on supported providers from one place.
The key caveat is stated openly on the page: offer counts describe configurations, not the number of physical GPUs bookable, and availability can change before you complete a booking. Prices are per GPU-hour but providers bill for the whole instance, and storage, network transfer and commitment terms affect the final bill. GPUPerHour is a research and comparison layer, not a hosting provider itself.
