Rent a GPU by the hour. Pay in crypto.
H100, H200, B200, A100, L40S, RTX 4090 and RTX 5090, paid only in crypto: Bitcoin, USDT, Monero and four more coins. Whole GPUs with root SSH, metered by the minute.
Pick a GPU and the hours you need below: the calculator finds the cheapest block and links straight to payment.
Estimate
- Rate × GPUs
- $2.39 × 1
- Per hour
- $2.39/hr
- Needed
- 24 GPU-hours
- You pay
- $114.72
Best buy: Session (48 GPU-hours)
Rent H100 · $114.72No charge until the SSH details of your instance appear. Unused GPU-hours stay on your account.
To rent a GPU on Petaflare you buy a block of GPU-hours for one model, from $0.34 an hour for an RTX 4090 to $4.89 for a B200, and pay it in crypto with no account check. You then launch an instance with 1 to 8 whole GPUs and root SSH. The meter only starts when your SSH details appear, counts by the minute, and stops when you stop the machine. See every price or read how renting works.
Seven NVIDIA GPUs to rent, one hourly price each
Datacenter cards for training and serving large models, GeForce cards for image generation and rendering. Each page lists the specs, the market price next to ours and which jobs the card fits. New to this? Here is what a cloud GPU is.
H100
80 GB HBM3 or HBM2e · Hopper
Training and fine-tuning large models, fast LLM serving with FP8
H200
141 GB HBM3e · Hopper
Large-model inference and long context: a 70B model in FP16 fits on two cards
B200
180 GB HBM3e · Blackwell
The fastest training and FP4/FP8 inference on the list
A100 80GB
80 GB HBM2e · Ampere
Proven training and inference in BF16 at the lowest price per GB of fast memory
L40S
48 GB GDDR6 · Ada Lovelace
Image and video generation, 3D rendering and mid-size LLM inference
RTX 4090
24 GB GDDR6X · Ada Lovelace
Stable Diffusion, ComfyUI, Blender and small LLMs at the lowest hourly price
RTX 5090
32 GB GDDR7 · Blackwell
Flux and video models, 30B-class LLMs in 4-bit, fast GPU rendering
Not sure which card?
Compare every block side by side and price a job in the calculator.
What renting a GPU costs here and at the big clouds
Dollars per GPU-hour for one card, read on each provider's own pricing page. Petaflare is below RunPod Secure Cloud on all seven GPUs, and below Lambda on every card Lambda lists. Marketplace and community offers can be cheaper: their hosts are third parties, and you need a card or an identity check to pay them. For each card's lowest price anywhere, see the cheapest cloud GPU, card by card.
| GPU | Petaflare | RunPod Secure | RunPod Community | Lambda | Vast.ai median |
|---|---|---|---|---|---|
| H100 80GB | $2.39 | $2.89 PCIe · $4.39 SXM | $1.99 PCIe · $2.69 SXM | $3.79 PCIe · $4.89 SXM | $1.33 SXM |
| H200 141GB | $3.29 | $5.29 | $3.59 | — | $3.50 |
| B200 180GB | $4.89 | $7.99 | $5.98 | $7.99 | $7.00 |
| A100 80GB | $0.99 | $2.09 PCIe · $2.19 SXM | $1.19 PCIe · $1.39 SXM | $3.19 (8 GPUs only) | $0.76 SXM4 |
| L40S 48GB | $0.87 | $1.09 | $0.79 | — | $0.53 |
| RTX 4090 24GB | $0.34 | $0.89 | $0.34 | — | $0.36 |
| RTX 5090 32GB | $0.49 | $1.39 | $0.69 | — | $0.46 |
Sources: RunPod pricing (updated 12 November 2026), Lambda pricing (plus sales tax), Vast.ai public price feed (daily median for 18 November 2026, cited with attribution as its licence asks). All read on 19 November 2026. A dash means no on-demand single-card offer.
From payment to SSH in four steps
No sign-up form and no card. The full walkthrough, with screenshots of the dashboard, shows each screen.
Get an account number
One click creates a 16-digit account number. It is your login; an email address is optional.
Pay for a block of GPU-hours
Choose a GPU and a block, then send the exact amount in the coin you hold. Hours are added once the payment is confirmed.
Launch an instance
Pick 1, 2, 4 or 8 GPUs, a PyTorch, CUDA or Docker image, and paste your SSH public key.
Connect and work
Our team prepares the machine. When the SSH address shows in your dashboard, the meter starts; stop the instance and it stops.
Pick a GPU by what you run
The card that fits depends on how much memory the job needs, more than on raw speed. Each guide gives the memory sums and the card to start with.
Stable Diffusion and Flux
SDXL runs in 12 GB; Flux and video models want 24 to 32 GB.
RTX 4090 · $0.34/hr02Serving an LLM
A 70B model in FP8 is 70 GB of weights before the cache: one H200 or two H100s.
H100 or H20003Fine-tuning a model
QLoRA on a 70B model fits on one 80 GB card; full fine-tuning does not.
A100 80GB · $0.99/hr043D rendering
Blender Cycles, Redshift and Octane scale with CUDA cores and VRAM.
RTX 5090 · $0.49/hr05A whole multi-GPU server
Up to eight cards in one machine for distributed training.
8 × H100Paid in crypto, with no account check
Petaflare takes the coins below at their network's normal fee. There is no ID, no card and no email required, as our no-KYC GPU cloud page details. The smallest payment is $25, and every block costs more than that, so you pay the exact block price; paying with USDT or Bitcoin shows each step. Mining, password cracking and attacks are not allowed on any instance. Paying for a GPU with crypto, coin by coin.
- ₿BitcoinBitcoin
- TTetherTron (TRC-20), Ethereum (ERC-20)
- CUSD CoinEthereum (ERC-20)
- ΞEthereumEthereum
- SSolanaSolana
- LLitecoinLitecoin
- MMoneroMonero
Renting a GPU: what people ask first
How much does it cost to rent a GPU?
On Petaflare, an RTX 4090 costs $0.34 per GPU-hour, an RTX 5090 $0.49, an L40S $0.87, an A100 80GB $0.99, an H100 $2.39, an H200 $3.29 and a B200 $4.89. The two largest blocks of each card are 5% and 10% cheaper per hour. A 24-hour fine-tuning run on one H100 comes to $57.36.
Which GPU should I rent?
Start from memory. Image generation and small LLMs fit on a 24 GB RTX 4090; Flux, video models and 30B models in 4-bit want the 32 GB RTX 5090 or the 48 GB L40S; serving or tuning 70B models needs 80 GB or more (A100, H100, H200, B200). The pricing page puts all seven side by side.
How does renting a GPU work here?
You buy GPU-hours for one model, launch an instance from the dashboard with your SSH key, and our team prepares the machine. Your instance draws from your GPU-hours by the minute while it runs: one hour on two GPUs uses two GPU-hours.
How long until my instance is ready?
A person on our team prepares every instance, so it is not instant. We do not promise a delay we have not measured; the meter only starts when your SSH details appear, so the wait never costs you GPU-hours.
Do I pay while the instance is stopped?
No GPU-hours are used while an instance is stopped. Its disk is kept until you delete the instance; copy your data out before you delete it, because deletion wipes the disk.
What happens to GPU-hours I do not use?
They stay on your account with no expiry date and work for any later instance of the same GPU model. If an instance fails to start, the hours it used come back; the refund policy has the details.
Can I rent several GPUs in one machine?
Yes: 1, 2, 4 or 8 whole GPUs per instance, never a slice of a card. Multi-GPU machines and what to check for distributed training are on the multi-GPU server page.
What is not allowed?
Cryptocurrency mining, password or hash cracking, attacks on other systems, malware, spam and any illegal content. The acceptable use policy lists every rule and what happens when one is broken.
Where are the terms, and how do I reach support?
The terms of service and the privacy policy set out the rules in plain English. Read about Petaflare for how we set and check prices, or contact support from your dashboard or by email.
Price your job, then rent the GPU
Every block of every card, with the cost of a real job worked out for you.