← All tools
ARTIFICIAL INTELLIGENCE · INFRASTRUCTURE

Local AI Hardware Calculator

Estimate whether your GPU memory can accommodate a language model at your intended context length and concurrency. Separate weights, KV cache and runtime headroom.

🔒 Browser-only. No hardware scan or AI inference. A model fitting in VRAM does not guarantee acceptable speed.
Model memory
Hardware available
Local vs cloud cost illustration
Cloud AI vs local AI guide →AI API Cost Calculator →

Calculation method

Weight memory = parameter count × bits per weight. KV cache ≈ 2 × layers × KV heads × head dimension × context tokens × concurrent sequences × bytes per element. This excludes quantisation metadata and temporary buffers except for the overhead reserve. VRAM values are decimal GB. Real model architecture, runtime, tensor parallelism and GPU memory availability can materially change the result. This tool deliberately does not invent tokens-per-second estimates.