Idle GPU Cost Calculator
A GPU left running idle bills exactly as much as one being used: this prices that out against live per-GPU $/hr rates, for the durations a box actually gets forgotten for.
What idle time actually costs
| GPU (live rate) | Left running overnight | Forgotten over a weekend | Forgotten for a week | Forgotten for a month |
|---|---|---|---|---|
V100 $0.088/hr | $1 | $5 | $15 | $63 |
RTX 3070 $0.088/hr | $1 | $5 | $15 | $63 |
RTX 4070 Super $0.121/hr | $1 | $7 | $20 | $87 |
RTX A2000 $0.132/hr | $1 | $8 | $22 | $95 |
Live per-GPU rate for each model, times hours idle: the same rate you'd pay for that GPU in active use. Idle time is not discounted; a forgotten box bills at its full rate.
Try your own numbers
Auto-stop can remotely stop a box on this provider: this cost is what you'd avoid by turning it on, not what happens on its own. It's opt-in and off unless you switch it on.
What actually stops this
Aquanode has an auto-stop feature that captures the pod's environment and volume, then releases the box after it's been idle past a threshold you set, but it is opt-in, never automatic. It is off until you turn it on: tick it in the console deploy sheet, set the idle policy through the API or the aq CLI, or switch it on later from the pod's Idle tab.
It works the same way on every provider Aquanode supports: auto-stop captures the pod then terminates the box, a create-and-terminate call every provider adapter implements, not a provider-specific stop API. There's no per-provider list to check before relying on it.
Starting a stopped box works the same way too: it rents a fresh box, here or on a different provider, and restores the environment and volume onto it, so it isn't gated by which provider you stopped on either. Recovering after a provider-side failure (spot reclaim, capacity loss) is only ever as good as the volume's last background save tick, which runs on its own, not a schedule you set.
