The cheapest GPU in the cloud, found for you.

gpu1.ai tracks live rental prices for the GPUs people actually train and serve models on — across Vast.ai's marketplace and RunPod — normalized per GPU, per hour, so an 8-GPU rig and a single card compare fairly.

See today's prices

13GPUs tracked

342Vast.ai offers sampled

2Providers compared

$0.12Cheapest GPU/hr

gpu1 — cheapest per GPU, USD/hr

Cheapest right now

The four GPUs people ask about most, at the prices below. Marketplace prices move constantly — the pill in the top bar always shows which set of prices you're looking at.

Full price comparison

Every price is per GPU per hour. Vast.ai figures come from live marketplace offers (cheapest and median); RunPod shows its current lowest available rate. “Best” is whichever provider wins that row — and that's where the Rent button points.

Min VRAM Sort

GPU	VRAM	Vast.ai from	Vast.ai median	RunPod from	Best

Inside the marketplace

The actual cheapest Vast.ai machines behind the headline numbers, sorted by per-GPU price. Watch the trade-off the headline hides: the rock-bottom offer sometimes sits on a host with a lower reliability score — the table note gives each GPU's “reliable floor”, the cheapest offer from a host rated 98% or better.

$/GPU/hr	Total $/hr	GPUs	Location	Reliability	Host CPU	RAM	Disk	Down / Up Mbps

What will your run cost?

Pick a workload — or dial in your own GPU, count, and hours — and see the bill on each provider at the prices above. Then check the rent-vs-buy line before you reach for a credit card.

Your run

GPU

GPU count

Hours: 100

The bill

Best case = the cheapest single offer in the current data; it may be gone by the time you click. Typical = median of live offers — the safer planning number. Storage, egress, and your own time aren't included, because they're never included in anyone's headline price either.

The providers, honestly

Two very different animals. gpu1.ai links to both through affiliate programs — disclosed in plain sight, because a comparison site you can't trust is worth nothing.

Vast.ai

A marketplace of independent GPU hosts bidding for your job.

Cheapest headline prices anywhere — hosts compete, so a 4090 dips to about $0.38/GPU/hr and an H100 to $1.34.
Quality varies by host. Filter by the reliability score; the cheapest offer isn't always the one that finishes your training run.
Best for: batch jobs, fine-tuning, experiments — anything checkpointed that can survive a host hiccup.

How gpu1.ai earns: Vast.ai pays 3% of everything you spend, for the life of your account. Your price doesn't change — the cut comes out of their side.

Rent on Vast.ai

RunPod

A managed GPU cloud — Secure Cloud datacenters plus a Community Cloud.

Predictable availability. A fixed rate card instead of a bidding war; an H100 SXM is $2.69/hr as listed above.
Per-second billing on pods, one-click templates, persistent volumes.
Best for: inference servers and long-running pods where “is it still up?” matters more than the last 20 cents.

How gpu1.ai earns: new accounts through this link get a $5–$500 credit bonus, and gpu1.ai earns credits too. A 12% cash affiliate tier unlocks after 25 paying referrals.

Get the RunPod credit bonus

On the roadmap: Lambda, CoreWeave, Vultr, and DigitalOcean rate cards join as curated rows next. Fixed-price clouds don't expose live-offer APIs the way a marketplace does, so their published rates get verified by hand before they enter the table.