Back home

GPU cloud · 13 plans · 6 providers

Compare GPU cloud pricing, per hour

Unlike everything else on this site, GPU cloud is priced and rented by the hour, not the month. Filter by minimum VRAM or budget-per-hour, then sort by price-per-GB-VRAM to see who's actually cheapest for the model you're running.

Configurator — GPU cloud

Tell it your workload, it tells you the GPU

Pick a preset or set your own minimums. Ranked by price-per-GB-VRAM across RunPod, Lambda and Paperspace on-demand instances.

Preset

24 GB
$/mo ~1

Providers

Min VRAM (GB)

Max $/hr

No cap

13 of 13 plans match

Ru
RunPod
RTX 4090
NVIDIA RTX 4090 · 24 GB VRAM · Community Cloud (spot, can be interrupted)
30+ regions
$0.34/hrSource
Ru
RunPod
A100 PCIe
NVIDIA A100 PCIe · 80 GB VRAM · Community Cloud (spot, can be interrupted)
30+ regions
$1.19/hrSource
La
Lambda
GH200 (1x)
NVIDIA GH200 · 96 GB VRAM · On-demand
US
$2.29/hrSource
Ru
RunPod
H100 PCIe
NVIDIA H100 PCIe · 80 GB VRAM · Community Cloud (spot, can be interrupted)
30+ regions
$1.99/hrSource
Co
CoreWeave
A100 80GB (8x)
NVIDIA A100 SXM · 640 GB VRAM · On-demand
US
$21.60/hrSource
Pa
Paperspace
RTX A6000approx
NVIDIA RTX A6000 · 48 GB VRAM · On-demand
US
$1.89/hrSource
Co
CoreWeave
L40S (8x)
NVIDIA L40S · 384 GB VRAM · On-demand
US
$18/hrSource
La
Lambda
A100 (1x, 40GB)
NVIDIA A100 · 40 GB VRAM · On-demand
US
$1.99/hrSource
La
Lambda
H100 SXM (1x)
NVIDIA H100 SXM · 80 GB VRAM · On-demand
US
$4.29/hrSource
Pa
Paperspace
H100approx
NVIDIA H100 · 80 GB VRAM · On-demand
US
$5.95/hrSource
Co
CoreWeave
HGX H100 (8x)
NVIDIA H100 SXM · 640 GB VRAM · On-demand
US
$49.24/hrSource
Te
Tencent Cloud
GN10X.4XLARGE80approx
NVIDIA V100 NVLink (×2) · 64 GB VRAM · On-demand
Singapore
$4.98/hrSource
Al
Alibaba Cloud
ecs.gn7i-c48g1.12xlargeapprox
NVIDIA A10 · 24 GB VRAM · On-demand
Southeast Asia (Singapore)
$3.95/hrSource

Methodology: figures last checked 2026-08-05. All prices are hourly on-demand rates (shown with a "/hr" suffix), not monthly — reserved/committed pricing is typically much lower and isn't shown here. RunPod's rate is its cheaper "Community Cloud" tier, not its pricier "Secure Cloud" tier for the same hardware. VRAM figures use each GPU model's standard published spec rather than reproducing a vendor page's own text verbatim, since at least one source page had an internally inconsistent VRAM number for a well-documented chip. Marketplace/spot providers with highly dynamic, bid-based pricing (e.g. Vast.ai) aren't included yet — a fixed number would misrepresent how that pricing actually moves.