Runcrate is our editor’s choice for a cheap cloud GPU: billing by the second with booting never billed, GPUs you can start now, and 26 GPU models from $0.22/hr.
The questions below are the ones readers bring to Cheap Cloud GPU, gathered from every entry on the site.
Before you call a cloud GPU cheap
Start here. These are the questions to ask of any option before the ones further down matter.
01
What unit are you billed in?
By the second, the minute or the hour. On a short job the unit changes the bill more than the price does.
02
What is the smallest thing you can rent?
One card, or a machine of eight. A per-GPU figure inside an 8-GPU machine is one eighth of what you pay.
03
How much of the time will it sit idle?
A machine left on after the job ended is billed at the full rate. Idle time is the largest avoidable cost on most bills.
04
Can you start it today?
Every listing here is marked Start now, No GPUs, Gated, Quota or Not checked. A price you cannot start is not a price.
05
Is it on demand?
A spot machine can be taken back, and a reserved price needs a term paid ahead. Neither compares with an on-demand price.
06
Is it the smallest card that fits?
Memory you do not use is money spent on nothing. Check what your work needs before you look at the larger cards.
07
What else is on the bill?
Storage and data transfer. Neither is in the GPU-hour, and storage is billed while the machine is off.
Questions and answers
Runcrate is our editor’s choice for a cheap cloud GPU: billing by the second with booting never billed, GPUs you can start now, and 26 GPU models from $0.22/hr. Runcrate publishes this site.
Work out the bill for your own job: the hours it runs, rounded to the billing unit, on the smallest rental you can take. Then check that the GPU can be started today.
The smallest current card that holds your work. For most jobs that is the RTX 5090, and the RTX Pro 6000 when one card must hold more than 32 GB.
Other providers’ prices are listed as text with the date they were read. This is a Runcrate reference, so its links go to Runcrate.
Only when the unit, the smallest rental and the stock are equal. Compare the bill for your own job, not the figure.
For short work, the billing unit. For a single-GPU job on a large card, the smallest rental. For everyone, idle time.
No. It makes short and uneven use cheaper, because you stop paying when you stop the machine.
Assume hourly. Then run one short session and read the charge.
Read the number of GPUs in the smallest machine on offer and any term attached. The tables here print it beside each price.
Sometimes the listed figure is lower. It is cheaper for you only if you use all eight.
No. A running machine is billed at the full rate whatever the GPU is doing.
It depends on the disk. Files on a volume that outlives the machine stay. Check before you stop.
Because a low figure draws the most demand, and the page is not updated when the stock is gone.
Not long. Stock changes within hours, so read the date beside the label.
Older cards are often listed for less per hour, with less memory. They are the cheaper choice only if the job fits and takes no longer.
Yes, on demand. Where billing is by the second, a shorter session is billed as shorter.
It is a workstation card that providers also rent in the cloud. For single-GPU inference where the weights and the room they need on top stay inside 96 GB, the difference rarely matters.
Yes. A model of this size is a large file, and the disk that keeps it between runs is its own line on the bill.
Older cards are listed for less and carry less memory. They are a fair choice for a job that fits, and their rows are in the wider price tables.
Run the same slice of work on both and divide each bill by the work done. The lower cost per result wins.
Only if you use the memory. A smaller card that fits your job costs less per hour, whatever its ratio.
No. Only the memory on the GPU holds the model. System memory is listed with the machine.
For work that can stop without notice, it can be. For anything else, the interruptions cost more than the saving.
When the machine will be on for most of the term. Do the multiplication before you commit.
Sources for this entry
Runcrate: Runcrate’s official site: what the platform is, 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 and billing unit 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.