Your workload,
powered by the
gaming grid.

Peer-to-peer batch & async compute tapping the untapped. Gamers have the best hardware on the planet; we connect their idle cycles to your pipeline.

cou-cli | bash
➜ ~ cou job submit --preset diffusion
0 active nodes
0 % uptime
0 avg GPU TFLOPS
0 pre-built presets

Built for the way teams actually work.

We don't sell raw compute and wish you luck. Every preset ships with the scaffolding your workload needs, ready to run.

peer-to-peer mesh

Jobs distribute across a dynamic graph of contributor nodes. No central SPOFs. Resilient by default.

[▶]
pre-built presets

ML inference, video transcoding, data pipelines, diffusion: submit a job in seconds, not hours.

gamer-grade silicon

RTX 4090s. RX 7900s. The high-end cards enthusiasts buy and then leave idle 20 hours a day.

async & batch native

Queue jobs, fan out across nodes, collect results. Fire and forget or await. Your call.

{:}
open job format

Declarative YAML job specs. Version-control your compute pipeline alongside your code.

community first

Contributors earn fair compensation. A healthier alternative to mining: idle compute with purpose.

From request to result.

Four steps. No cluster management. No DevOps PhD required.

01
pick your preset

Choose from a growing library of pre-wired workload templates, or describe your own.

02
submit your job

Push data and config via CLI or API. The scheduler handles placement automatically.

03
nodes do the work

Contributor machines pull shards, process in parallel, and stream results back.

04
collect your output

Results land in your designated sink: S3-compatible, webhook, or raw file.

job.yaml
output.json
1# clusterof.us job spec
2preset: diffusion/sdxl
3version: "1.0"
4
5input:
6  prompts: ./prompts.txt
7  batch_size: 128
8
9resources:
10  gpu_min_vram: "12gb"
11  nodes: 8
12
13output:
14  sink: s3://my-bucket/results
15  format: png
invite only

Tell us what you're building.

clusterof.us is currently invite-only. Tell us what you want to run, then submit with Google. We take name and email from that account so we know the request is yours.

> We prioritize:
  + ML inference & training jobs
  + Media processing pipelines
  + Scientific / simulation workloads
  + Novel distributed use cases
◉ connected 4,200 nodes online
main UTF-8 00:00:00