XurGive each workload its space
Assign GPUs to desktops, models and containers. Independent workloads load in parallel; a failure in one does not discard the others.
One host.
More possibilities.
A gaming desktop. A local model. Or both at once. Xur gives your GPUs a job, with profiles you control from a browser.
An image-based OS for GPU workload profiles, built on Bazzite.
One four-GPU hostIllustration
GPU 124 GB
Workstation ADesktop and games
GPU 224 GB
Workstation BA separate desktop
GPU 324 GB
Language modelTwo GPUs assigned
GPU 424 GB
Two desktops and a language model, each with its own GPUs.
Keep the hardware.
Change what it does.
A profile describes which workloads to run and where they belong. Save a few for the way you use your machine, then switch from the control panel.
Keep your desktop familiar
Reuse a named workstation across profiles to keep its user and Moonlight pairing. Assign USB devices or a hub, and choose a primary desktop for shared peripherals.
Use your models
Find ready language-model endpoints in one place. Try a conversation or retain a benchmark with the model, running workloads and sampled VRAM.
A few ways to share the machine
These example layouts use four RTX 3090s. Your profiles depend on the GPUs, memory and models you have.
Set up a workstation
Start with a guide
Install and make your first profileGetting startedStream a desktop with MoonlightWorkstationsConnect a client to a local modelModels and APIsChoose a channel and keep a way backUpdatesBuilt in the open
Xur is preparing for its first release. Follow development, inspect the code and report what works on your hardware. Release downloads and their notes live on GitHub.
View releases