Understand the platform
Availability
Capacity changes. Here's how to plan around it and confirm hardware before you commit.
What the dashboard tells you
Capacity goes up and down. The dashboard shows what you can provision right now — and that's the only count that matters. If a GPU is in the marketing docs but not selectable in the dashboard, it isn't yours to use yet.
Before a deadline-sensitive run
- Confirm the exact hardware and config in the dashboard — not in memory.
- Run a short validation job before you commit a big training or inference run.
- Reach out to ANK when you need a specific GPU, a certain amount of memory, or a specific start time.
- Keep checkpoints somewhere you can restore — because deleting an instance takes them with it.
When asking ANK to confirm capacity, send the model, required memory, expected runtime, and preferred hardware. Full checklist lives on Support.