CotrievalAI

Confidential GPU hosting

Your own Confidential GPU, by the hour.

Run open and uncensored models on an NVIDIA GPU made for your booking alone: pictures and video with sound on a Confidential VM, or GLM-5.3-Flash on a private 4-GPU machine. Your time starts when it is ready, and the machine is deleted when it ends.

NVIDIA H100 SXM
glm-5.3-flash-uncensored226–231 tok/s · 4× H100
4× H100 80GB · measured Oct 14
NVIDIA RTX PRO 6000 Blackwell Server Edition
CC status: ONAMD SEV · 97887 MiB
RTX PRO 6000 Blackwell · what a booked machine reported, Oct 16
NVIDIA RTX PRO server with RTX PRO 6000 Blackwell GPUs
ready 6:11 after paymentsealed · deleted at the end
One booking, one machine

Pick it like a ride.

Choose a machine

GLM-5.3-Flash, uncensored. Prices are per hour; nothing is created until you pay.

2 hours US$19.006 hours at most
Choose 4× H100 80GB

Choose the model and the GPU, then the hours. You see the price before you pay, your time starts only when the machine is ready, and unused time comes back automatically.

Open the portal →

Why it takes a few minutes

Most hosted models answer from servers many customers share. Yours is made for you. It takes a few minutes, and nobody else's work has ever run on it.

Preparing your machine

Your time has not started yet.

    NVIDIA RTX PRO 6000 Blackwell 6MIN

    Iowa · Google Cloud1× RTX PRO 6000 · Confidential VM

    A real Confidential GPU booking on Oct 16, played back 30 times faster. The last lines are what the machine reported about itself.

    1

    booking per machine

    6 min

    from payment to a ready Confidential GPU

    0 min

    billed before it is ready

    Models

    Two ways to book, each on a machine of its own.

    Every machine runs on its own NVIDIA GPUs in Google Cloud, is made when you book and is deleted when your time ends.

    NVIDIA RTX PRO 6000 Blackwell Server Edition
    Pictures, video with soundConfidential

    Confidential GPU

    ComfyUI in your browser with six ready-made workflows: Qwen-Image 2.1 for pictures and edits, MiniMax H3 for video with sound, on one RTX PRO 6000 Blackwell inside a Confidential VM.

    Runs on
    1× RTX PRO 6000, 96 GB, Confidential VM
    Protection
    AMD SEV memory encryption, GPU in confidential mode
    Speed
    5.2 s per 1024 px picture once the model is loaded
    Your files
    Only in its encrypted memory
    Ready
    About 6 minutes
    NVIDIA RTX PRO server
    LanguageUncensored

    GLM-5.3-Flash

    An uncensored build for long coding, research and security work, served with vLLM behind an OpenAI-compatible API, with an address and a key that are yours alone.

    Context
    1M tokens
    Runs on
    4× H100 80GB or 4× RTX PRO 6000
    Speed
    226–231 tokens/s on H100, about 145 on RTX PRO 6000
    Protection
    Private machine, not confidential: Google sells confidential GPUs one at a time
    Ready
    10 to 15 minutes

    US$9.50an hour

    Book GLM-5.3-Flash
    • Book 1 to 6 hours. Your time starts when the machine is ready, not when you pay.
    • End early, or lose the machine to a Google Spot interruption, and the unused time is refunded.
    • If no zone has a free GPU or the machine never becomes ready, you get everything back.

    Privacy

    Private, by design.

    Each line is true of the machine you book today. Where a protection is not on, it says so. The details and the known gaps are on the security page.

    Privacy protections on each machine
    ProtectionConfidential GPUGLM-5.3-Flash · 4 GPUs
    Made for your booking alone Yes Yes
    HTTPS ends on the machine, no proxy Yes, its own certificate Yes, its own certificate
    Only you can open it Yes, booking token Yes, your API key
    Nobody logs in to it Yes, SSH closed Yes, SSH closed
    Memory encrypted while it runs Yes, AMD SEV Not on this machine
    GPU in confidential mode Yes Not on this machine
    No outbound internet once ready Yes, sealed Not on this machine
    Machine and disk deleted at the end Yes Yes

    Google offers confidential GPUs only as single-GPU machines, and GLM-5.3-Flash needs four, so its machine runs as a standard virtual machine. The Confidential GPU does not yet offer remote attestation, so you cannot check its confidential state cryptographically from outside; the booking page shows what the machine reports about itself.

    Book your own RTX PRO 6000 Blackwell at CotrievalAI today.

    96 GB of GDDR7 per GPU, on a machine made for your booking alone.

    See the RTX PRO 6000 →

    FAQ

    Questions

    Anything else, write to hello@noahgeneralgroup.com.

    What does "confidential" mean here?

    The Confidential GPU runs in a Google Confidential VM: the processor encrypts its memory with AMD SEV, and the RTX PRO 6000 runs in NVIDIA's confidential-computing mode. HTTPS ends inside the machine, its outbound internet is closed once it is ready, and your uploads and results stay in its encrypted memory. When your booking page shows the machine, it also shows what the machine reports about these settings.

    Why does it take minutes to be ready?

    Because the machine is new. Each booking gets its own machine, and the model is loaded into its memory before it serves you: about 6 minutes for the Confidential GPU, 10 to 15 for GLM-5.3-Flash. You are not billed for that time; your hours start when it is ready.

    How do I use it?

    The Confidential GPU opens ComfyUI in your browser from your booking page. GLM-5.3-Flash gives you an address and a key for an OpenAI-compatible API, so OpenCode, the OpenAI SDKs and curl work with it. The docs have examples.

    What happens if I end early, or the machine is interrupted?

    The unused share of your payment goes back to your card automatically. Machines run on Google Cloud Spot capacity, which Google can take back at short notice; if that happens, the booking ends and you are refunded for the time you did not get.

    Is the GLM-5.3-Flash machine confidential?

    No. Google offers confidential GPUs only as single-GPU machines, and GLM-5.3-Flash needs four. It runs as a private standard machine, created for your booking, reachable only over HTTPS with your key, and deleted with its disk when the booking ends.

    What is not allowed?

    Anything illegal, including sexual content involving minors, intimate images of real people without their consent, and attacks on systems you do not own. The full rules are in the terms.