Pay for results,not hours.
Coming soonzgpu bills per delivered output: per finished job, or per second of output. You never pay for machines booting, models downloading, or GPUs sitting idle.
No paying for dead time.
Rent a GPU by the hour and the meter runs while it boots, downloads, sits idle and retries. Here, only finished work is billed.
Hourly rentalbilled for every minute
zgpubilled per delivered output
How jobs run.
You package the work as a container. The network finds a card that fits, runs it, and hands back the result.
- You send
- a container, a command, requirements
- Requirements
- VRAM, card models, cached files
- Pricing
- per job, or per second of output
- A machine drops out
- the job runs again elsewhere
- An owner starts a game
- the job moves, never marked failed
- Every attempt
- recorded with the job
{
"image": "registry.example.com/your-job:latest",
"cmd": ["render", "--in", "/input", "--out", "/output"],
"requirements": {
"min_vram_mb": 12288,
"gpu_models": ["GeForce RTX"]
},
"timeout_s": 900,
"max_attempts": 3,
"price": { "unit": "output_second" }
}Be first in line.
Customer jobs open after the network has proven itself on real work. Join the list and we'll reach out when it's ready.