answered with live data · 2026-07-08

What is the cheapest GPU for Stable Diffusion?

Stable Diffusion runs well on modest hardware, so the cheapest sensible pick is a consumer card with at least 12GB to 24GB of VRAM, like an RTX 3090 or 4090. The cheapest verified 4090 in our index is $0.69/hr. You do not need a datacenter card for image generation unless you are training or fine-tuning at scale.

GPU	VRAM	$/hr	Where
RTX A4000	16 GB	$0.15	Hyperstack on-demand	Rent →
Tesla V100	16 GB	$0.17	DataCrunch on-demand	Rent →
RTX A5000	24 GB	$0.27	RunPod secure cloud	Rent →
L4	24 GB	$0.39	RunPod secure cloud	Rent →
A40	48 GB	$0.44	RunPod secure cloud	Rent →
RTX 3090	24 GB	$0.46	RunPod secure cloud	Rent →
RTX A6000	48 GB	$0.49	RunPod secure cloud	Rent →
RTX 4090	24 GB	$0.69	RunPod secure cloud	Rent →
RTX 6000 Ada	48 GB	$0.77	RunPod secure cloud	Rent →
L40	48 GB	$0.82	RunPod secure cloud	Rent →

Base Stable Diffusion models fit in 8GB to 12GB of VRAM for inference, and SDXL is comfortable at 16GB or more. A 24GB card gives you room for larger batches, higher resolutions, and running a ControlNet or LoRA stack without swapping. For pure generation, spending more on an H100 buys speed you rarely need and pay a large premium for.

If you are training a LoRA or fine-tuning a checkpoint, VRAM matters more and a 24GB card is the practical floor. Full fine-tuning of SDXL benefits from more memory, at which point an A100 or similar starts to make sense.

Because generation is bursty, a spot or community instance is a good fit. You can start it, run a batch, and stop it, paying only for the minutes you use. The table below lists the cheapest cards we track that comfortably handle Stable Diffusion.

What is the cheapest GPU for Stable Diffusion?

Related questions