Register your GPU.
Register your GPU and commit it to a coin. Paid jobs open with the contributor pilot; until then jobs run on managed GPUs.
Your GPUs. Your workspace.
Register machines and run the connection test. Community GPUs can’t earn yet. The current pilot supports Linux machines with NVIDIA GPUs.
What is your rig good for?
For ex-mining rigs and single cards alike. Run the worker’s local check, paste what it prints, and see which work fits and what blocks the rig today.
Run the local check on the rig
Download the worker to the Linux machine, review it, then run:
python3 compute-worker.py diagnose
It reads
Download compute-worker.pynvidia-smi, makes no network request and needs no credential.Read in this browser only. Nothing you paste is uploaded or saved.
How verdicts are decided
The worker applies one fixed table per group of identical GPUs. Memory is per GPU, because it does not pool across cards. These are rules of thumb, not qualification: an operator still reviews each machine for a specific workload.
| Use | Good | Limited | Not suitable |
|---|---|---|---|
| AI inference | 24 GB+ | 12–23 GB: smaller or quantized models | Under 12 GB |
| AI training | 24 GB+ on x8 or wider links | 12–23 GB, or narrower links | Under 12 GB |
| 3D rendering | 8 GB+; risers are fine | 4–7 GB | Under 4 GB |
| Scientific and engineering | Compute capability 7.0+ and 8 GB+ | Anything else | — |