Estimate whether a local AI model is likely to fit in a mini PC, AI PC, Mac or desktop memory budget. Use the exact model-file size when you have it; parameters alone are not enough to establish memory use.
Local AI memory estimate
Will this model likely fit in memory?
Enter model-file and runtime memory estimates. This calculator runs in your browser; Kingy does not receive the values you enter.
- Estimated requirement
- Estimated usable memory
- Upper estimate / usable
- Confidence
Formula and decision boundaries
required = model weights + KV cache + backend overhead + runtime workspace + concurrency reserve
usable = installed memory − OS reserve − shared graphics reserve
- Likely fits: upper requirement is at most 85% of usable memory.
- Borderline: upper requirement exceeds 85%, but the lower estimate is no more than usable memory.
- Does not fit: lower requirement exceeds usable memory.
What this estimate can and cannot tell you
- It estimates memory fit from an explicit range; it does not predict useful speed or output quality.
- KV cache changes with architecture, precision, context, batch size and concurrent sessions.
- Shared-memory systems, GPU offload, multimodal components and backend workspaces can materially change the requirement.
- A model that loads can still be too slow, thermally constrained or incompatible with the selected runtime.
Kingy has not yet published a matching measured mini-PC run. Until the fixed three-story hardware protocol is complete, treat every result as a planning estimate. Browse the current AI mini-PC buying guide.