New NVIDIA B300 · 288 GB HBM3e servers — from $4,019/mo per GPU

See the B300

Deploy

Deploy your GPU server.

Four steps: the server, optional AI software sized to your model, your account, then the first month in crypto. Online in under 10 minutes, no KYC.

Server

GPU model

AI flagship

Pro

Creator

GPUs per server

48 GB GDDR6 ECC · 12 vCPU · 96 GB RAM · 1 TB NVMe

Software

Optional · included in the price
Operating system
GPU software

NVIDIA driver, CUDA toolkit, Fabric Manager on NVLink servers, Docker with the NVIDIA Container Toolkit.

AI software needs the GPU-ready layer. Switch GPU software to GPU-ready to add it.

Serve a model optional
Also install optional
Options

Account

Account type

No KYC: we never ask for ID documents and run no identity verification.

Billing address

Server accessCreate an SSH key

Add a note for our team

    Payment

    How crypto payment works
    Pay with

    Next, you get the address and the exact amount in the coin you picked, valid for a limited time. You pay the network fees. Once the payment is confirmed, your server is online in under 10 minutes.

    Before you pay.

    Something else? Ask us before ordering.

    How fast will my server be ready?

    In under 10 minutes once your payment is confirmed on-chain. If you asked for large models, their weights can keep downloading in the background after your server is online.

    Do you ask for ID documents?

    No. We run no KYC: no ID documents, no selfie and no identity verification.

    How do I get into the server?

    We send the server’s address and root access to the email on your order. Add your SSH public key to the order to log in with it; Getting started covers the first login, and SSH access shows how to create a key.

    What does the Software step install?

    Only what you pick, at no extra cost: Ubuntu, the GPU driver with CUDA or ROCm and Docker, then optionally a model server (vLLM, SGLang or Ollama with Open WebUI) with the model of your choice, ComfyUI, JupyterLab or training tools. You are root: change or remove anything afterwards.

    Who is responsible for model licences?

    You are. We download the official weights from Hugging Face or the Ollama library; each model keeps its own licence, shown next to its name. Models whose licence restricts commercial use are marked.

    Which coin and network should I use?

    Any of the options in the Payment step. The next page shows the exact amount in that coin, the address and the network. USDT exists on several blockchains: send on the network named on the invoice only, as a payment sent on another network may not be recoverable.

    What if the quote expires, or I send too little?

    The crypto amount is valid for a limited time, shown on the order page. If it runs out, get a new quote there rather than sending the old amount; your server price does not change. If less than the invoice amount arrives, the order page shows what is missing.

    Can I cancel an order?

    Yes, as long as no payment has been sent: open the order page and cancel it. Once a server is delivered, payments are non-refundable except where the law requires otherwise, and if we cannot deliver a server you ordered, we refund the payment in crypto or credit it, as you choose. See Refunds.

    Customer area

    Sign in to CryptGPU

    Your orders, invoices and servers in one place.

    New to CryptGPU?

    Accounts are created at checkout: choose your server and software, then create your account in the Account step of your first order.

    Deploy a server