What is a GPU hour?
A GPU hour is one GPU running for one hour. It is the basic unit cloud providers use to bill and to measure the size of a job. Eight GPUs running for one hour is eight GPU hours, and one GPU running for eight hours is also eight GPU hours, so the two are directly comparable.
The GPU hour is useful because it separates the size of a job from how fast you finish it. If a training run needs one thousand GPU hours, you can do it on one GPU for a thousand hours or on a hundred GPUs for ten hours, and the compute bill for the GPUs is roughly the same either way. What changes is your wall-clock time and any overhead from splitting the work.
When you see an hourly price on this site, that is the price of one GPU hour on that provider and tier. Multiply it by the GPU hours your job needs to estimate the compute cost. Remember that the total bill also includes storage, data transfer, and any per-instance CPU and memory that comes bundled with the card.
Related questions
- How much does it cost to train an LLM?
- Should I look at GPU price per month or per hour?
- How does multi-GPU node pricing work?
Numbers on this page come from today's verified snapshot. Full table on the homepage; method in the methodology.