The Fleet Isn't Just for Games
The same community node fleet that streams cloud gaming sessions also runs general-purpose GPU compute jobs — inference, training, transcoding, and analytics — submitted from a template or a custom container and billed by the minute.
Built for Real Jobs, Not Just Renders
Submit a job against a public template or bring your own container image or script — with GPU backend, minimum VRAM, and a maximum duration you set.
Inference
Run a packaged model against your inputs and get results back — sized to whichever GPU backend the job requests.
Training
Longer-running training jobs go through the same node-matching and artifact pipeline, scheduled against your maximum duration and minimum VRAM.
Transcoding
Media and transcoding workloads run in the same containerized job pipeline, with output artifacts ready to download when the job completes.
Analytics
Batch analytics scripts run on demand, with job logs streamed live so you can watch a run in progress.
Custom Scripts
Bring your own container image or script, with environment variables and input artifacts included in the job spec for a fully custom workload.
Metered by the Minute
Every job is billed per minute against the GPU backend it runs on, with a platform fee built into the rate. Spot jobs are preemptible and priced at half the on-demand rate.
For scripts and analytics jobs that don't need a GPU.
For workloads that target the ROCm backend.
For workloads that target the CUDA backend.
Rates include a 15% platform fee and a one-minute minimum charge. Spot pricing is 50% off the on-demand rate shown above and jobs may be preempted.
How a Compute Job Runs
Submit a Job
Start from a public job template or define your own spec — container image or script, GPU backend, minimum VRAM, and a maximum duration — then submit it.
The Fleet Allocates a Node
The same allocator that matches cloud gaming sessions to nodes finds an operator node with the right GPU backend and capacity, then starts your container.
Watch It Run, Collect the Result
Job logs stream live while it runs. When it finishes, your output artifacts are ready to download and the actual cost settles against the estimate held on submission.
Two Ways to Get Involved
Submit a job as a builder, or list your GPU as an operator and earn from both gaming sessions and compute jobs on the same node.
GPU compute runs on the same operator fleet as cloud gaming. Availability depends on your tournament organizer's configuration.