Best Cloud GPU for SDXL inference

Minimum 12GB VRAM · Recommended 24GB+ · Runtime: seconds-per-image

Cheapest for SDXL inference: NVIDIA GeForce RTX 3090 24GB on Salad

$0.130/hr/hr · verify on provider site

Cheapest GPU Options — 25 eligible GPUs

Provider	Configuration	Region	Billing	Availability	Price/hr
Saladcheapest	1x RTX 3090	distributed	per-minute	on-demand	$0.130/hr	Rent →
Salad	1x RTX 4080	distributed	per-minute	on-demand	$0.170/hr	Rent →
vast	1x RTX 3090	eu-central	per-second	on-demand	$0.200/hr	Rent →
TensorDock	1x RTX 3090	eu-west	per-minute	on-demand	$0.220/hr	Rent →
Salad	1x RTX 4090	distributed	per-minute	on-demand	$0.240/hr	Rent →
vast	1x RTX 4080	us-west	per-second	on-demand	$0.280/hr	Rent →
genesis	1x RTX 3090	eu-central	per-minute	on-demand	$0.290/hr	Rent →
RunPod	1x RTX 3090	us-east	per-second	on-demand	$0.290/hr	Rent →
vast	1x RTX 4090	eu-west	per-second	on-demand	$0.350/hr	Rent →
TensorDock	1x RTX 4080	us-east	per-minute	on-demand	$0.350/hr	Rent →
FluidStack	1x RTX 4090	eu-central	per-minute	on-demand	$0.440/hr	Rent →
RunPod	1x RTX 4090	us-east	per-second	on-demand	$0.440/hr	Rent →
Replicate	Nvidia T4 (16GB)	us-east	per-second	on-demand	$0.440/hr	Rent →
Hyperstack	1x L4 24GB	uk-london	per-minute	on-demand	$0.490/hr	Rent →
DataCrunch	1x RTX 4090	eu-north	per-minute	on-demand	$0.490/hr	Rent →

GPU Requirements

Minimum VRAM	12 GB
Recommended VRAM	24 GB
Ideal GPUs	NVIDIA GeForce RTX 4090 24GB, NVIDIA A10 24GB
Typical Runtime	seconds-per-image
Billing Pattern	spiky

FAQ

What GPU do I need for SDXL inference?

Requires at least 12GB VRAM. Recommended: 24GB+. Ideal: NVIDIA GeForce RTX 4090 24GB, NVIDIA A10 24GB.

What is the cheapest GPU for SDXL inference?

NVIDIA GeForce RTX 3090 24GB at $0.130/hr/hr on Salad.

How much does SDXL inference cost per hour?

From $0.130/hr/hr. Runtime: seconds-per-image.

GPU-Specific Pages

NVIDIA H100 80GB SXM for SDXL inference

NVIDIA H100 80GB PCIe for SDXL inference

NVIDIA H200 141GB SXM for SDXL inference

NVIDIA A100 80GB SXM for SDXL inference