Cost of the whole run

Compute for the entire job, not the hourly rate. Set the shape of your run once and every provider that publishes a rate for that card is priced against it.

GPU model

GPUs

Run length

Total GPU-hours

Massed Compute

$53—

compute $53 · $6.60/GPU/hr× 8 GPU-hoursstorage and egress not tracked

Gcore

$55+4%

compute $55 · $6.85/GPU/hr× 8 GPU-hoursstorage and egress not tracked

RunPod Community Cloud

$56+5%

compute $56 · $6.94/GPU/hr× 8 GPU-hoursstorage and egress not tracked

Modal

$57+8%

compute $57 · $7.10/GPU/hr× 8 GPU-hoursstorage and egress not tracked

Enverge

$60+14%

compute $60 · $7.50/GPU/hr× 8 GPU-hoursstorage and egress not tracked

RunPod Secure Cloud

$63+20%

compute $63 · $7.89/GPU/hr× 8 GPU-hoursstorage and egress not tracked

Latitude.sh

$64+21%

compute $64 · $8.00/GPU/hr× 8 GPU-hoursstorage and egress not tracked

Verda

$65+23%

compute $65 · $8.12/GPU/hr× 8 GPU-hoursstorage and egress not tracked

Fal

$68+29%

compute $68 · $8.50/GPU/hr× 8 GPU-hoursstorage and egress not tracked

Scaleway

$68+30%

compute $68 · $8.56/GPU/hr× 8 GPU-hoursstorage and egress not tracked

Nebius AI Cloud

$76+44%

compute $76 · $9.50/GPU/hr× 8 GPU-hoursstorage and egress not tracked

Spheron

$82+55%

compute $82 · $10.21/GPU/hr× 8 GPU-hoursstorage and egress not tracked

OOracle Cloud Infrastructure

$120+127%

compute $120 · $15.00/GPU/hr× 8 GPU-hoursstorage and egress not tracked

Amazon Web Services

$142+170%

compute $142 · $17.80/GPU/hr× 8 GPU-hoursstorage and egress not tracked

How this is worked out

The arithmetic

1 GPU × 8 h=8 GPU-hours

8 GPU-hours × $6.60/GPU/hr=$53

We run that sum once for each provider. The GPU-hours stay the same — only their rate changes. Cheapest is at the top, and the percentage next to each row is how much more it costs than the cheapest one. Right now the same 8 GPU-hours cost $53 at the cheapest provider and $142 at the most expensive.

Which prices are counted

Basisper GPU only

Billingon demand only

Left outspot · reserved · committed · quote-only

Per providertheir cheapest rate for this card

Models listed38 of 63 tracked

Spot prices are left out because the provider can take the machine back mid-run, so they are not a price for an 8-hour job. The picker only lists cards that at least two providers price the same way — with one provider there is nothing to compare.

What this does not know

Minimum order sizes aren't checked. Some providers only sell this card in fixed bundles — a node of 8, say — so a listed rate for 1 GPU is not always something you can actually buy at every provider on this list.
The run occupies the GPUs continuously for all 8 hours. No queueing, no idle time, no checkpoint restarts, no per-second billing below the hour.
Availability is not checked. We hold published prices, not stock levels, so this will quote you 1 GPU from a provider who may not have 1 free right now.
Compute only. Storage and data transfer are not included — we don't collect those prices. This makes providers with free data transfer look worse than they are, and providers who charge for it look better. Moving 4 TB out at $0.02/GB would add $80 that this table never shows.