Cheapest GPU to generate 1,000 images.
Prices checked
RTX A5000 · Community Cloud at RunPod does this job for $0.32 per 1,000 images, the cheapest of 162 machines that can, checked .
Cheapest 10
all 162 rentals- 1
RTX A5000 · Community Cloud
Marketplace
RunPod · 1x RTX A5000 · 24 GB$0.32per 1,000 images - 2
RTX 3090 · Community Cloud
MarketplaceRunPod · 1x RTX 3090 · 24 GB$0.44per 1,000 images - 3
1V100.4V · 1x Tesla V100 16GB
Own machines
Verda (DataCrunch) · $0.24 on spot$0.47per 1,000 images - 4
2V100.10V · 2x Tesla V100 16GB
Own machinesVerda (DataCrunch) · $0.24 on spot$0.47per 1,000 images - 5
4V100.20V · 4x Tesla V100 16GB
Own machinesVerda (DataCrunch) · $0.24 on spot$0.47per 1,000 images - 6
8V100.40V · 8x Tesla V100 16GB
Own machinesVerda (DataCrunch) · $0.24 on spot$0.47per 1,000 images - 7
RTX A5000 · Secure Cloud
Own machinesRunPod · 1x RTX A5000 · 24 GB$0.54per 1,000 images - 8
RTX A6000 · Community Cloud
MarketplaceRunPod · 1x RTX A6000 · 48 GB$0.66per 1,000 images - 9
RTX 4090 · Community Cloud
MarketplaceRunPod · 1x RTX 4090 · 24 GB$0.68per 1,000 images - 10
RTX 5090 · Community Cloud
MarketplaceRunPod · 1x RTX 5090 · 32 GB$0.77per 1,000 images
Other ways to pick
What matters
- 16 GB is the floor
- An SDXL-sized model and one 1024-pixel image fit in 16 GB. Everything in this index clears it, so the job is decided on price and speed rather than on whether it runs.
- Throughput, not memory
- Once the model fits, more memory buys nothing. What you are paying for is how fast the card pushes the denoising steps, which is why the cheap boards do well here and badly everywhere else.
- One GPU at a time
- Images are generated one per GPU, so a node of eight is eight times the work and eight times the price. The cost below is per GPU, which makes node size irrelevant.
- Cold starts cost money
- The model has to load before the first image. On a run this short, a provider that bills by the second and keeps the weights on local disk is worth more than a few cents on the rate.
How we count
Image generation is the one job on this site where the cheap cards win outright, because the model fits on all of them and only throughput separates them. We count only machines with 16 GB of GPU memory or more, and 162 of the 162 rentals we read qualify. The cheapest of them is $0.32 per 1,000 images at RTX A5000 · Community Cloud, which rents for $0.16 per GPU-hour. They are ranked by what the work costs rather than by the hourly rate, because a faster card at twice the price can still be the cheaper way to buy the job.
An image is one 1024 by 1024 picture from a Stable Diffusion XL sized model. Boards are placed in four speed classes by the memory bandwidth their maker publishes, and counted at 1,400, 900, 500 or 180 images an hour from the fastest class to the slowest. Those four numbers are assumptions for planning, not something measured here, so treat the ranking as the shape of the market rather than as a benchmark. Each price is that work at the machine's on-demand rate: RTX A5000 · Community Cloud rents for $0.16 an hour, and $0.32 per 1,000 images is what the work costs on it.
- Marketplace
- The provider matches you with someone else's machine.
- Own machines
- The provider owns or leases the machines.
Frequently asked
- What is the cheapest GPU to generate 1,000 images?
- RTX A5000 · Community Cloud at RunPod, $0.32 per 1,000 images. That is 1x RTX A5000 with 24 GB of memory each, renting for $0.16 per GPU-hour. Read on 1 Nov 2026.
- How is the cost worked out?
- An image is one 1024 by 1024 picture from a Stable Diffusion XL sized model. Boards are placed in four speed classes by the memory bandwidth their maker publishes, and counted at 1,400, 900, 500 or 180 images an hour from the fastest class to the slowest. Those four numbers are assumptions for planning, not something measured here, so treat the ranking as the shape of the market rather than as a benchmark.
- Why do only 162 machines qualify?
- The job needs 16 GB of GPU memory or more. The other 0 rentals we read are still in the index; they just cannot do this job as described. Serverless rates are left out entirely, because they buy a container that runs when it is called rather than a machine you hold.
- Can I pay less on spot?
- 68 of the 162 qualifying machines publish an interruptible rate, and each of those rows shows what the same work would cost on it if the run were never preempted. It can be, so treat the spot figure as a floor rather than a quote.
- How current are these prices?
- Prices are read every 3 days from 10 providers' own public pages and keyless APIs, and this ranking recomputes with them. Last refresh: 1 Nov 2026.