GPU servers, paid in crypto. Monthly, no meter, no KYC.
A dedicated NVIDIA GPU with its own vCPU, memory and NVMe, CUDA and drivers pre-installed, paid from a crypto-funded balance. RTX 5090 to 8× H100, in Frankfurt, Amsterdam, New York and Singapore.
How renting a GPU with crypto works
No quota request, no card, no ID. Fund the balance, pick the card, SSH in.
-
01
Create an account with an e-mail address
The same account rents a $3 VPS and an 8× H100 node.
-
02
Top up in the coin you hold
An RTX 5090 is $449 a month, an H100 SXM $979. Fund the balance in one deposit; a large deposit in USDT avoids any rate question. Up to $5,000 per top-up.
-
03
Deploy and connect
Pick the GPU, a location and an image (vLLM, Ollama, PyTorch, ComfyUI or plain CUDA), then SSH in. Renewals draw from the same balance.
- MonthlyBillingNo hourly meter, no idle charges, no surprise bill
- Yours aloneGPUNo time-slicing, no shared VRAM
- Under 60 sReady inUbuntu 24.04, CUDA 12.8 and driver 570 pre-installed
- NoneIdentity checksPay in BTC, ETH, XMR or USDT
Seven GPUs, one flat monthly price
Every plan is a dedicated GPU with its own vCPU, memory and NVMe. No hourly meter, no sharing, no setup fee.
| GPU | VRAM | vCPU / RAM | NVMe | Network | Monthly | Order |
|---|---|---|---|---|---|---|
| NVIDIA RTX 4090 | 24 GB GDDR6X | 16 vCPU64 GB RAM | 500 GBNVMe Gen4 | 1 Gbpsunmetered | $259/mo | Sold out |
| NVIDIA RTX 5090Low stock · 3 left | 32 GB GDDR7 | 16 vCPU96 GB RAM | 1 TBNVMe Gen4 | 1 Gbpsunmetered | $449/mo | Order |
| NVIDIA L40S | 48 GB GDDR6 | 24 vCPU128 GB RAM | 1 TBNVMe Gen4 | 1 Gbpsunmetered | $539/mo | Sold out |
| NVIDIA A100 | 80 GB HBM2e | 24 vCPU192 GB RAM | 2 TBNVMe Gen4 | 10 Gbpsunmetered | $789/mo | Sold out |
| NVIDIA H100 SXMPopular5 in stock | 80 GB HBM3 | 32 vCPU256 GB RAM | 2 TBNVMe Gen4 | 10 Gbpsunmetered | $979/mo | Order |
| NVIDIA H200 SXM | 141 GB HBM3e | 32 vCPU384 GB RAM | 4 TBNVMe Gen4 | 10 Gbpsunmetered | $1,559/mo | Sold out |
| NVIDIA 8× H100 SXMLow stock · 1 left | 640 GB HBM3 | 192 vCPU2 TB RAM | 30 TBNVMe Gen4 | 25 Gbpsunmetered | $9,300/mo | Order |
Monthly GPUs for workloads that run all month
GPU clouds bill by the hour and ask for a card, and often an ID, before they let you near an H100. That model fits bursty experiments; it fits a model that serves traffic all month badly, because an hourly price multiplied by 720 hours is usually far above a monthly one, and the meter runs while you sleep. We rent GPUs by the month at a fixed USD price, paid in crypto, with the whole card to yourself.
Every plan is a dedicated GPU: no time-slicing, no shared VRAM, no noisy neighbour. RTX 5090 with 32 GB for image generation and 8B to 14B model inference, L40S and A100 for fine-tuning, H100 and H200 with HBM for 70B-class serving and training, and an 8× H100 SXM node with NVLink for the rest. Images ship with CUDA 12.8 and the usual stacks; serving a 70B model in FP8 on one H100 is documented step by step.
Paying in crypto does not change the hardware; it changes the sign-up. There is no card, no identity verification and no quota request. A researcher in a country the card clouds refuse, an anonymous open-source project, or a company that simply wants a fixed monthly cost get the same server at the same price.
What people run on a crypto-paid GPU
- LLM inferenceServe open-weight models with vLLM or Ollama behind an OpenAI-compatible API; 70B in FP8 fits one H100.
- Fine-tuningLoRA and QLoRA on a 5090 in an afternoon; full fine-tunes on H100 and H200.
- Image and video generationComfyUI, Stable Diffusion and FLUX with models on a persistent volume.
- Rendering and transcodingBlender, Unreal pixel streaming and NVENC farms on RTX cards.
- Speech and audioWhisper servers, XTTS and voice conversion with room for concurrency.
- Private AIRun models on hardware nobody else touches, paid without a card or a name.
Everything a GPU node needs, nothing to add on
- Whole GPU, no sharingNo vGPU slicing and no time-sharing: the full card and all of its VRAM are yours.
- CUDA 12.8 readyDrivers, cuDNN, NCCL, Docker and the NVIDIA Container Toolkit are pre-installed.
- NVMe scratch spaceLocal NVMe Gen4 for datasets and checkpoints, 500 GB to 30 TB depending on the plan.
- Unmetered network1 to 25 Gbps with no traffic cap. Pull models and datasets as often as you like.
- Jupyter in one clickJupyterLab, Ollama, vLLM and Hugging Face TGI images ready at checkout.
- Private networkFree VLAN between your nodes in the same location for multi-node jobs and shared storage.
- DDoS protectionAlways-on volumetric filtering up to 3 Tbps, at no extra cost.
- Monthly, no commitmentKeep the node for a month or a year. No setup fee, no minimum term, same price at renewal.
GPU nodes in 4 locations, same price everywhere
- Core network site
- Data center
Europe3
FRAFrankfurtGPU
AMSAmsterdamGPU
HELHelsinki
North America2
NYCNew YorkGPULAXLos Angeles
Asia-Pacific1
SINSingaporeGPU
Other ways to pay, other things to rent
Bitcoin, Ethereum, Monero or Tether fund one USD balance. Each coin has its own page with confirmation times, fees and pitfalls.
GPU server questions, answered
Why monthly instead of hourly?
A GPU that serves traffic all month is far cheaper on a fixed monthly price than on a metered one, and a fixed price cannot surprise you. If you need a GPU for two hours, an hourly cloud is the right tool; if you need it for two weeks or more, this is.
Can I pay for an H100 with crypto without verification?
Yes. The same e-mail-only account that rents a $3 VPS rents an H100 or an 8× H100 node. Top up the balance and order.
Which GPUs are in stock?
The stock shown on the GPU page is the real free pool, updated every minute. Sold-out cards show as such; nothing is listed that cannot be provisioned now.
What is pre-installed?
Ubuntu 24.04 with CUDA 12.8 and driver 570 by default, or Ubuntu 22.04 with CUDA 12.4. Templates add vLLM, Ollama, TGI, PyTorch, TensorFlow, JupyterLab, ComfyUI and more; Docker with the NVIDIA Container Toolkit is on every image.
Can I connect several GPU servers?
Yes. Servers in the same location can share a private VLAN for multi-node training or a serving cluster; the 8× H100 node has NVLink internally and a 25 Gbps uplink.
How do refunds work for expensive plans?
The same as everywhere: unused balance is refundable within 14 days of the top-up in the coin you paid with. A cancelled server runs to the end of its paid month.
From the people who run the racks
GPU & AIServing an open-weight LLM with vLLM on an H100From a bare Ubuntu image to an OpenAI-compatible endpoint for Llama 3.1 70B in FP8, with the flags that matter and the throughput we measured.
GPU & AILoRA fine-tuning Llama 3.1 8B on a single RTX 5090 in an afternoonDataset formatting, QLoRA settings that fit in 32 GB, how long an epoch really takes and how to evaluate before you ship the adapter.
GPU & AIComfyUI on a rented GPU: a setup that survives rebootsA systemd unit, models on a persistent volume, a password in front of the UI and the RTX 4090 vs 5090 numbers for SDXL and FLUX. Your GPU node from $259 a month, online in under a minute.
Top up from $25 in BTC, ETH, XMR or USDT. Your balance never expires and unused funds are refundable.