MODEL FIT · RULES-BASED ESTIMATOR
Will it run on my GPU?
An honest first-pass estimate of whether a model fits your hardware. We show you the arithmetic, not a black-box "yes." Always confirm against a real benchmark before buying.
ESTIMATE
Fits comfortably
Estimated VRAM footprint: 18.97 GB of 24 GB
THE MATH (transparent)
- →Q4-K-M ≈ 0.55 bytes/parameter (industry convention; actual artifact may differ).
- →KV cache + runtime estimated at ~2.47 GB for 8,192-token context.
- →Comfortable fit requires ≥15% VRAM headroom (3.6 GB).
- →Estimate does NOT account for: layer counts, attention implementation, batch size, multi-GPU splitting, or OS VRAM. Real measurements override this.