Marketplace de location de GPU
Louez des GPU cloud à l'heure. by the hour.
Les tarifs horaires les moins chers de 8 fournisseurs sur 39 modèles de GPU — pour l'entraînement, le fine-tuning et l'inférence.
Right now ·
Nvidia Nvidia L4
from
$0.016/hr
on
Vast.ai
· refreshed 2 minutes ago
01
The market right now
What costs what.
Cheapest median hourly rate available in each memory band. Read it as a floor — your actual cost depends on quant, GPU count, and host reliability.
Loading...
Ad
02
Pick your workload
Que faites-vous tourner ?
Or browse every GPU →
Train large language models
≥80GB
Fine-tune LLMs (LoRA / QLoRA)
≥24GB
Run LLMs (inference / serving)
≥12GB
Image generation (Stable Diffusion, Flux, …)
≥8GB
Video generation
≥24GB
Speech and audio AI
≥8GB
3D rendering and content creation
≥12GB
General ML research and experiments
≥12GB
Scientific compute (HPC, simulation)
≥24GB
Text embeddings
≥2GB
Code generation
≥16GB
03
Top picks by tier
Datacenter
Frontier iron
Workstation
Multi-GPU desk + workstation
Consumer
Hobbyist + edge
04
Cheapest per VRAM bucket
Sized by what you need to fit.
See all →
06
Biggest inference providers
By models hosted on each API.
See all →
Every price on this page comes from each provider's own API, refreshed hourly. No editorial favourites, no affiliate sleight of hand on the headline rate. Pick the GPU that fits your job — we just keep the prices honest.