OS

One GPU. Up to eight tenants.

PodVirt is the GPU virtualisation layer in Podstack OS that splits a physical NVIDIA GPU into isolated shares from 12.5 to 100 percent.

PodVirt is the GPU virtualisation layer in Podstack OS. It splits a physical NVIDIA GPU into isolated shares from 12.5 to 100 percent, so a small job pays for a small slice and the operator sells the whole card several times over.

See it in QuickPods
At a glance

PodVirt at a glance

Shares
12.5% to 100%
Tenants per GPU
up to 8
Used by
QuickPods
Licensing
with Podstack OS
What it does

What PodVirt does

Fractional allocation

Shares of an eighth, a quarter, a half or the whole GPU, scheduled by the platform so utilisation stays high.

Isolation

Each share runs in its own isolated environment with its own storage; tenants do not see each other.

Fleet-aware scheduling

The scheduler places shares across the fleet, so a new request lands on the GPU with room rather than waiting on a full one.

Proprietary, purpose-built

PodVirt is Podstack's own layer, built for GPU sharing rather than adapted from general-purpose virtualisation.

What you run and what Podstack runs

You run the facility. Podstack ships the software.

Podstack OS is installed inside your data center and operated by your team. Podstack maintains the software and, if you opt in, brings demand.

You run

  • The GPUs, servers, network and power, in your facility.
  • Your brand on the storefront, the console and the invoices.
  • Your prices, your customers and your customer relationships.
  • Day-to-day operations of the racks and the tenants on them.

Podstack runs

  • The operating system: provisioning, pooling, metering, storefront and console.
  • Updates, security patches and support for the stack.
  • The same software Podstack runs its own cloud on, so behaviour matches.
  • Optional demand through Podstack Grid, at a floor price you set.
How it works

How PodVirt works

  1. 1

    Enable on GPU nodes

    PodVirt installs on DC Suite nodes and advertises fractional capacity to the platform.

  2. 2

    Tenants pick a share

    A tenant launching a pod chooses how much of a GPU they need.

  3. 3

    Meter per share

    Each share is metered per GPU-minute, so the operator bills several tenants for one card.

How it bills

How PodVirt bills

included with Podstack OS

PodVirt

PodVirt ships with DC Suite and is covered by the per-GPU annual license. It can also be licensed on its own; ask for a proposal.

Works with

PodVirt works with

FAQ

Questions about PodVirt

What is PodVirt?

PodVirt is the GPU virtualisation layer in Podstack OS that splits one physical NVIDIA GPU into isolated shares from 12.5 to 100 percent.

How many tenants can share one GPU?

Up to eight, at the smallest share of 12.5 percent each, with larger shares reducing the count accordingly.

Are shares isolated from each other?

Yes, each share runs in its own isolated environment with separate storage, and tenants cannot see one another.

Which Podstack products use PodVirt?

QuickPods on Podstack Cloud run on PodVirt, and any DC Suite fleet can enable it for its own tenants.

Is PodVirt licensed separately?

PodVirt is included with the Podstack OS per-GPU license and can also be licensed on its own.

Is PodVirt based on open-source virtualisation?

No, PodVirt is Podstack's own layer, purpose-built for GPU sharing.

Start with PodVirt

PodVirt is the GPU virtualisation layer in Podstack OS that splits a physical NVIDIA GPU into isolated shares from 12.5 to 100 percent.

See it in QuickPods