Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.ornn.com/llms.txt

Use this file to discover all available pages before exploring further.

When you have an accepted reservation, Ornn lets you choose how your GPU hardware is materialized and accessible to you. VM gives you a managed virtual machine backed by the Ornn base image or an approved custom image. Bare Metal gives you direct SSH access to the host with no virtualization layer between you and the hardware. You set this choice per reservation at /portfolio/access.

Comparing VM and Bare Metal

VMBare Metal
Use caseMost ML training and inference workloadsWorkloads that need direct NVIDIA driver access or maximum throughput
Performance profileNear-bare-metal; slight virtualization overheadNo overhead — direct physical host access
Image optionsOrnn base image (Ubuntu + CUDA + PyTorch) or an approved custom imageNo image — you manage the software environment directly
Recommended forMost Ornn Cloud customersAdvanced users who need full hardware control
VM is the default and recommended mode. If you are unsure which mode to use, start with VM.

How to set your access mode

1

Go to /portfolio/access

Navigate to /portfolio/access. If you have multiple reservations, use the Portfolio reservation dropdown to select the one you want to configure.
2

Select VM or Bare Metal

Click either the VM or Bare Metal card. Your selection is saved immediately. The page confirms the saved mode and updates the URL to reflect your choice.
3

Continue to your reservation

After saving, click Continue to VM management or Continue to Bare Metal management to go to the reservation detail page, where SSH connection details appear once provisioning completes.
You must have at least one accepted reservation before you can set an access mode. If your portfolio is empty, visit /reserve to find and reserve capacity.

Learn more

VM access

Connect via a managed VM using the Ornn base image or an approved custom image.

Bare Metal access

SSH directly into the GPU host for maximum performance with no virtualization overhead.