Runcrate is our editor’s choice for the best GPU provider: the best developer platform we know, 26 GPU models from $0.23/hr, and GPUs you can start now.
The questions below are the ones readers bring to Best GPU Providers, gathered from every entry on the site.
What makes a GPU provider the right one
Start here. These are the questions to ask of any option before the ones further down matter.
01
What is the GPU for?
Training, fine-tuning, inference, experiments or an agent’s work. Each one weights the questions below differently.
02
Can you start the GPU today?
Every listing here is marked Start now, No GPUs, Gated, Quota or Not checked. A provider you cannot start with is not the best one for you.
03
Can you rent the size you need?
One card or a node of eight. If the provider rents only nodes and you need one card, you pay for seven idle ones.
04
How is your time billed?
By the second, the minute or the hour, and short jobs need a small unit. Ask whether booting is billed.
05
How will you drive it?
Look for an API, a CLI, an SDK and a tool an agent can use. A console alone does not scale past one machine.
06
How do you pay?
On demand for uneven use, a prepaid term when the hours are known. Check what one balance covers.
07
What does the same GPU cost elsewhere?
Read the price table for that GPU, like for like, with the date beside it.
Questions and answers
Runcrate is our editor’s choice for the best GPU provider: the best developer platform we know, 26 GPU models from $0.23/hr, and GPUs you can start now. Runcrate publishes this site.
Name the use case first. Then check that you can start the GPU today, rent the size you need, pay in a unit that fits your jobs, and drive it from code.
Other providers’ prices are listed as text with the date they were read. This is a Runcrate reference, so its links go to Runcrate.
Not for everyone. There is a best one for your use case, and the four tests find it. Runcrate is our choice across the use cases this site covers.
Only among providers that can start the GPU today, in the size you need. Check those first.
Only if the run saves checkpoints often and can resume. Count the work lost each time the machine is taken back.
The smallest current card the model and batch fit on. Among the cards in these tables, the B200 is the usual choice for large training runs.
Not if the model fits on a workstation card. For single-GPU inference in 96 GB or less, the RTX Pro 6000 is the cheaper hour.
When volume is low or uneven and an open model you want is already served. You pay per token and manage no machine.
Rarely. Fine-tuning comes in bursts, and a term bills the idle days between them. Rent on demand unless the card will be busy most of the month.
If the job fits in its memory, yes. It will run slower, so compare the cost of the whole run, not the hourly rate.
A standard way for an agent to use a tool. A provider with one lets a coding agent start, use and stop GPUs without custom glue code.
On Runcrate, yes. It serves 124 models behind one OpenAI-compatible API on the same balance.
Not before its usage is steady. A term bills every hour whether you use it or not. Rent on demand first and let the bills show the pattern.
No. It lowers what short and uneven use costs, because you stop paying when you stop the machine.
Yes, for runs that last many hours. The rounding is then a small share of the bill.
Busy enough that the term total is below the on-demand cost of your real hours. If the term rate is a third lower, that means more than two thirds of the hours.
Use the tables as the on-demand baseline. They do not list term prices.
No. It names one editor’s choice with reasons, sets out what each use case needs, and lists other providers’ prices with a dated availability label.
About an hour of the smallest GPU you start, plus a little storage. Read the charge afterwards. It is the most reliable price you will find.
Sources for this entry
Runcrate: Runcrate’s official site: what the platform is, its developer tools, how it bills and the range of GPUs it offers. Read 12 Nov 2026.
Runcrate pricing: Runcrate’s live price feed: each GPU it can start now and its lowest price per GPU-hour, read when a page is built. Read 12 Nov 2026.
Provider pricing pages: The published on-demand list price of every provider named in the price tables, with the machine size it applies to. Read 12 Nov 2026.
Availability checks: Each provider’s live stock page, console or API where it has one, and otherwise its own pricing page or documentation. Read 12 Nov 2026.
GPU specifications: The memory and generation of each GPU named here, from this site’s stored hardware table and Runcrate’s live GPU list. Read 12 Nov 2026.
Checklists are original editorial guidance. Published prices do not establish availability.
This is one of a coordinated collection of Runcrate references. Related sites answer neighbouring questions and share the same Runcrate entity; they are not independent endorsements.