Private pilot. By invitation.Request access

Node Lodge
Request access

GPU Cloud

NVIDIA GPUs in partner datacenters.

A dedicated VM per session. Book 1 to 24 hours, SSH in from your IP, deleted at the end.

Request access Quickstart
ubuntu@nl-4f2a
# Ubuntu 24.04 · NVIDIA driver 570 · CUDA 12.8$ ssh ubuntu@203.0.113.42$ nvidia-smi --query-gpu=name,memory.total --format=csv,noheaderNVIDIA H100 80GB HBM3, 81559 MiB$ uv pip install vllm --torch-backend=auto
Example session.

The GPUs

Pick the VRAM. We handle the rest.

See rates in the console
  • HopperDatacenter

    H200 141GB SXM

    Largest models in one GPU

    VRAM
    141 GB
    Billing
    Per min
    Access
    SSH
  • HopperDatacenter

    H100 80GB SXM

    Training and fast inference

    VRAM
    80 GB
    Billing
    Per min
    Access
    SSH
  • HopperDatacenter

    H100 80GB PCIe

    Inference and fine-tuning

    VRAM
    80 GB
    Billing
    Per min
    Access
    SSH
  • AmpereDatacenter

    A100 80GB SXM

    Proven training workhorse

    VRAM
    80 GB
    Billing
    Per min
    Access
    SSH
  • BlackwellDatacenter

    RTX PRO 6000

    Large models, rendering

    VRAM
    96 GB
    Billing
    Per min
    Access
    SSH
  • Ada LovelaceDatacenter

    L40

    Inference, rendering, video

    VRAM
    48 GB
    Billing
    Per min
    Access
    SSH
  • AmpereDatacenter

    RTX A6000

    Fine-tuning and rendering

    VRAM
    48 GB
    Billing
    Per min
    Access
    SSH
  • AmpereDatacenter

    RTX A4000

    Small models and batch jobs

    VRAM
    16 GB
    Billing
    Per min
    Access
    SSH

Stock changes by the minute. Live availability and rates in the console.

A session

Catalog to shell. Four steps.

  1. 01

    Pick a machine

    GPU, VRAM, vCPU, RAM, disk, region, rate.

  2. 02

    Book

    1 to 24 hours. The maximum is held from your credit.

  3. 03

    Start

    Paste your public key and your IP. Ready in about 2 minutes.

  4. 04

    Stop

    Pay for the minutes used. The rest returns.

Isolation

One port. One address. One key.

Each session gets its own VM and its own firewall, created with the machine and gone with it.

Security
An isolated session Each session sits behind its own firewall. Inbound, only SSH on port 22 from your IP is allowed; every other source is dropped. Outbound traffic is allowed except SMTP on port 25, which is dropped. At the end, the machine and its disk are destroyed. FIREWALL YOUR SESSION running Your IP in 22/tcp ALLOW Other sources in * DROP ALLOW Internet out * DROP SMTP out 25/tcp AT THE END Machine and disk destroyed
An isolated session Each session sits behind its own firewall. Inbound, only SSH on port 22 from your IP is allowed; every other source is dropped. Outbound traffic is allowed except SMTP on port 25, which is dropped. At the end, the machine and its disk are destroyed. Your IP in 22/tcp Other sources in * ALLOW DROP FIREWALL YOUR SESSION running ALLOW DROP Internet out * SMTP out 25/tcp AT THE END Machine and disk destroyed

The machine

A plain Ubuntu VM. Yours for the session.

Install what you need with sudo and apt or pip. On your side, SSH is all it takes.

Image
Ubuntu 24.04 LTS, NVIDIA driver 570, CUDA 12.8
Access
SSH as ubuntu, port 22, from the IP you give
Outbound
Open over IPv4, except TCP 25
Duration
1 to 24 hours, chosen at booking
Interruptible
Marked in the catalog. The provider can reclaim it.
At the end
Machine and disk deleted. No backups.

Request access.

Invitations go out by email.

Request access Sign in