← all tools

Cheapest place to rent any GPU.

Renting an H100, A100 or RTX 4090 in the cloud? Prices vary wildly, and the big clouds charge two to five times what specialist GPU providers do. This pulls live hourly rates from 40+ providers and ranks them, so you can find the cheapest place to rent the exact GPU you need right now.

Live prices via ComputePrices, refreshed continuously. Spot and marketplace rates change by the minute, so confirm on the provider before you commit.

Search
GPU
Pricing
Currency

GPU VRAM Provider Pricing Commit $/hour
Fetching live GPU prices…
How to read this. The metric is the rental price per GPU hour in USD, lower is better. Pricing: on-demand runs until you stop it; spot is cheaper but can be interrupted, so it suits checkpointed jobs; reserved needs a commitment (shown in the Commit column) but cuts the rate. Filter to the exact GPU you need and the cheapest live option is highlighted. Marketplace and spot rates change minute to minute, so always confirm on the provider before renting. Prices are USD, converted live to your chosen currency.

Common questions

What is the cheapest cloud GPU?

For raw price, consumer cards (RTX 3090/4090) on Vast.ai or Salad can be under $0.50/hr. For training, an A100 or H100 on RunPod, Cudo, Spheron or Vast is far cheaper than AWS/Azure/GCP. Filter to your GPU and sort by price.

Why are AWS/Azure/GCP pricier?

They bundle enterprise support, compliance and reliability, often 2-5x the cost per GPU hour. For solo devs and startups, specialist GPU clouds give the same hardware far cheaper.

On-demand vs spot?

On-demand runs until you stop it; spot is cheaper but can be reclaimed at short notice (as little as 15s on Vast). Spot suits fault-tolerant, checkpointed workloads and can halve costs.

Does per-second billing matter?

Yes for short jobs. RunPod, Salad and AWS bill per second; some enforce per-hour minimums (Paperspace), which raises the real cost of quick runs. Billing is shown per option.