The fleet behind Overdo.
Dedicated GPU nodes on a control plane that keeps them warm for your workloads. Rent capacity by the second — or run your whole inference stack on it.
What's in the fleet
| GPU | Memory | Best for | On-demand | Reserved |
|---|---|---|---|---|
| RTX 5090 | 32 GB GDDR7 | Image, video & inference pipelines | $0.99 | $0.89 |
| RTX PRO 6000 | 96 GB GDDR7 | Large-model inference & fine-tuning | $1.79 | $1.49 |
| Multi-GPU node | up to 8× cards | Training runs & batch rendering | Custom | Custom |
per GPU / hour · billed per second of actual run time
A control plane, not a waitlist.
You never manage instances, you submit jobs. Capacity is staged ahead of your workloads — jobs boot on a hot GPU and the result streams back.
- Warm pool — no capacity lottery, no cold starts
- Jobs API — submit → poll → fetch artifacts
- Per-second billing — you pay for run time, not reservation time
#f31a image done 4.2s #f31b video running 61% #f31c site queued — #f31d voice done 1.1s #f31e image done 3.8s
Warm capacity
Your jobs land on a booted GPU, not a cold one. Capacity stays staged for your workloads — first render in seconds, not minutes.
Burst, then zero
Spin a node up for a batch render or training run, scale it back to zero when it's done. Billed per second of GPU time — never for idle.
Jobs API
Submit generation and rendering jobs programmatically — the same queue that serves the Overdo workspace. Poll, fetch artifacts, meter per job.