GPU cloud prices compared
Hourly on-demand prices for the cards people actually rent, side by side. Our prices are the ones you are billed (read 2026-09-15); RunPod's are from its pricing page dated 2026-09-13, Lambda's from its GPU cloud page read 2026-09-15. Vast.ai is a marketplace where each host sets a price, so there is no single number to quote.
All prices are USD per GPU per hour. We round up to the minute; one hour is pre-authorised at start and unused minutes are refunded on stop.
| GPU | cloudgpu.app | RunPod (community / secure) | Lambda | Vast.ai |
|---|---|---|---|---|
| RTX 3090 24 GB | $0.21/h | $0.22 / $0.50 | not offered | host-set, varies |
| RTX 4090 24 GB | $0.33/h | $0.34 / $0.74 | not offered | host-set, varies |
| RTX 5090 32 GB | $0.59/h | $0.69 / $0.99 | not offered | host-set, varies |
| RTX A6000 48 GB | $0.44/h | $0.33 / $0.53 | $1.09 | host-set, varies |
| A100 80 GB | $1.49/h | $1.19 / $1.59 (PCIe) | $1.99 (PCIe), $2.79 (SXM) | host-set, varies |
| H100 80 GB | not offered — H800 80 GB $1.79/h and H20 96 GB $1.29/h are the closest we have | $1.99 / $2.89 (PCIe) | $3.29 (PCIe), $3.99 (SXM) | host-set, varies |
Sources: cloudgpu.app pricing page (2026-09-15); runpod.io/pricing (page dated 2026-09-13); lambda.ai/service/gpu-cloud (2026-09-15). Prices on every side move; check the live pricing pages before you budget.
How to read this table
Consumer cards are where we are cheaper. RTX 3090, 4090 and 5090 cards are plentiful in Chinese data centres, so an RTX 5090 here costs less than RunPod's community tier and well under its secure tier. If your job fits in 24 or 32 GB of VRAM, which covers fine-tuning 7B to 14B models, Stable Diffusion, FLUX and quantised 32B inference, this is the saving.
Data-centre cards are not. Our A100 80G sits between RunPod's community and secure tiers and below Lambda. We do not offer H100s; H800 and H20 are the closest. If you need A100s or H100s in quantity and can pay by card, RunPod's community tier or Lambda may be the better choice, and we say so.
The other differences are not in the table: our machines are in mainland China (plus some in Singapore and the US), so expect 150 to 250 ms from Europe or North America; we accept card, USDT and bank transfer; there is no subscription and balance does not expire.
Questions people ask before choosing
What is the cheapest way to rent an RTX 5090 by the hour?
As of 2026-09-15, an RTX 5090 32 GB on cloudgpu.app is $0.59 per hour, billed per minute. RunPod's pricing page dated 2026-09-13 lists the same card at $0.69 (community) and $0.99 (secure). Lambda does not offer consumer cards. Vast.ai hosts set their own prices, sometimes lower, without a fixed rate.
How much does a 20-minute experiment cost?
Because billing is per minute, a 20-minute run on an RTX 4090 at $0.33/h costs about $0.11 and on an RTX 5090 at $0.59/h about $0.20. One hour is pre-authorised when the machine starts and the unused minutes are refunded the moment you stop it.
Is cloudgpu.app cheaper than RunPod for A100 or H100?
Not necessarily. Our A100 80G is $1.49/h; RunPod lists the A100 PCIe at $1.19 (community) and $1.59 (secure). We do not offer H100s at all. For data-centre cards, compare both pricing pages on the day; the saving here is on consumer cards.
Where are the machines and what latency should I expect?
Supplier data centres are mainly in mainland China (Shanghai, Beijing, Inner Mongolia, Guangdong), with some capacity in Singapore and the US. From Europe or North America expect 150 to 250 ms round trip: fine for training, fine-tuning and batch inference, noticeable for interactive chat.
How do I pay?
Visa or Mastercard (any amount from $5 to $2,000, handled by Creem as merchant of record), USDT on TRC-20, or bank transfer for business customers. The balance is prepaid, does not expire, and there is no subscription.
Do the prices include tax?
Our hourly prices are what is deducted from your balance. When you top up by card, the amount you choose is the amount charged: any VAT or sales tax for your country is included in it, not added on top, and the full amount is credited to your balance. USDT and bank transfers are credited 1:1.