The honest broker for GPU compute

The actual floor price
for every GPU,
across every provider.

Eleven providers. One screen. Real-time pricing. Built by an operator who runs the rigs, not a marketer who writes blog posts. We show you the cheapest option, even when it isn't ours.

11
Providers tracked
50K+
GPU offerings indexed
Hourly
Price refresh cadence
$0
Cost to use

The floor, right now.

GPU Cheapest provider $/hr Listings 24h Δ
Loading current floor prices…

Get told before the price moves.

GPU pricing shifts hourly. Spot prices can drop 40% overnight. The provider you used last month may now be the most expensive option for your workload.

We watch eleven providers continuously. You get one email per week (or instant alerts on threshold crossings) with the actual cheapest option for your specs. No promotion, no spam, no upselling.

  • 01 Weekly digest. Friday morning. Price changes across all providers, ranked by impact on your saved configurations.
  • 02 Threshold alerts. Tell us your target price. We email when an A100 drops below $0.70/hr or a 4090 below $0.30/hr.
  • 03 Provider movement. When a provider raises prices materially, we tell you what's now cheaper for the same workload.
  • 04 Outage and trust signals. If a host goes dark or quality drops, we surface it. We're operators ourselves.
━ Free · No card · Unsubscribe anytime
Watch the floor.

Pick a GPU, set your price, get notified. Or just take the weekly digest — it's the most useful 60-second email you'll get all week.

No spam. No selling your address. Real talk only.
✓ done.
You're on the list.

Check your inbox for a confirmation. First digest hits Friday morning. If a price moves materially before then, you'll hear from us sooner.

Eleven providers, one screen.

We track every major GPU rental marketplace and managed cloud. No favorites. No payola. We earn when you book — through whichever provider you pick.

Vast.ai
Marketplace · peer-to-peer
ModelHosts + DCs
BillingPer second
SpotYes (interruptible)
RunPod
Hybrid · community + secure
ModelDual-tier cloud
BillingPer second
ServerlessYes
TensorDock
Marketplace · global DCs
ModelVM-based
BillingPer minute
Locations20+ countries
Hyperbolic
Marketplace · supplier-mix
ModelAggregated supply
BillingPer minute
Inference APIYes
Lambda
Managed · enterprise
ModelOwned hardware
BillingHourly
ReservedYes
Thunder Compute
Managed · low-cost focus
ModelVirtualized GPUs
BillingPer second
VS CodeNative
Hyperstack
Managed · NVIDIA partner
ModelNVIDIA-certified
BillingHourly
InfiniBandYes
GMI Cloud
Managed · H200/B200 focus
ModelDatacenter
BillingHourly
Latest genB200 NVL72
Paperspace
Managed · DigitalOcean
ModelNotebooks + VMs
BillingHourly
NotebooksNative
Akash
Decentralized · permissionless
ModelOpen bidding
BillingPer block
Crypto-nativeAKT token
SaladCloud
Marketplace · consumer-first
ModelDistributed
BillingPer second
Best forInference
━ Why this site exists ━

The cheapest GPU on the market is almost never the one you're using.

Pricing shifts hourly. Hosts come and go. The provider that won last month is overpriced this week. Most renters check two sites and book on whichever loaded first.

ComputeFloor was built by someone who runs GPU rigs for a living, who got tired of watching friends overpay because the comparison was tedious. It's free, it's fast, and it'll show you the actual floor — even when that floor isn't us.

Signed, the operator · Fort Worth, TX · 2026