On-Demand GPUs

GPUs is the Console’s home page. It lists every GPU offer in the marketplace, and launching one takes a single form. Instances are paid from your wallet, so keep it topped up before you deploy.

Browsing the marketplace

Three summary tiles sit above the list:

Tile	Meaning
Available GPUs	Offers currently in stock, out of the total number listed.
GPU Models	Distinct GPU models across the catalogue.
Regions	Regions that currently have stock.

Narrow the list with the search box (GPU model name) and the filters:

Filter	Options
Region	North America, Europe, Asia Pacific and other regions carrying stock.
Model	A specific GPU model, for example H100, A100_80G or L40S.
Count	Number of GPUs per instance, for example 1×, 2×, 4× or 8×.
Vendor	GPU vendor.
Type	Instance type, for example VM.
Sort	The last control changes the order of the list.

Reading an offer card

Each card shows the region, the model (prefixed with the count for multi-GPU offers, such as 2x L40S), and two blocks:

• Compute — GPU count and VRAM per GPU.
• Resources — vCPUs, local disk and system RAM.

Below these are the interconnect (PCIe or SXM), the hourly price and the instance type. Badges such as Best Value and Most Popular highlight notable offers. Select View to open the offer.

Launching an instance

The offer page repeats the hardware specification — GPU, VRAM per GPU, vCPUs, system RAM, local disk and region — above the launch form.

Basic configuration

Field	Notes
Instance name	Optional. If you leave it blank, the Console generates a name such as keen-pulse-461.
Region	Fixed by the offer.
SSH key	One of your registered keys. It is installed on the instance for login.
Operating system	The image to boot, for example ubuntu22.04_cuda12.8.
Tags	Optional labels.
Project	The project the instance’s cost is charged to. New creates one inline.
Enable observability	Installs the Velerion agent to stream metrics, logs and alerts for the instance. See Observability.

Advanced configuration

This section is optional.

• Machine environment variables are available to startup scripts and system processes. They are not passed into Docker containers.
• Choose between a Startup script and a Docker container to run at boot. When you use a startup script, the machine environment variables are prepended to it as export statements.

Attach storage

Optionally attach a storage volume, so that data outlives the instance.

Estimated cost

The cost panel shows the hourly rate with per-day and per-month projections. Change Estimate for to see the total for a specific duration. Select Launch instance to deploy.

Caution

If the offer has sold out since you opened it, the panel reads Unavailable and the page shows “This GPU is currently unavailable.” Go back to the list and choose another offer.

After launch

The instance appears under My Resources. Billing is hourly: the wallet records a charge for each hour, and when you delete the instance the unused part of the current hour is returned as a refund.