ModelRigSYS: READY
LOCAL AI / MEMORY FIT ENGINE

Know what runs.
Before you buy.

ModelRig estimates whether an AI model and quantization fit a GPU, Apple unified-memory system, mini PC, or workstation. Memory fit first; benchmarks second.

MODEL × QUANTIZATION × MEMORYNO FAKE TOKENS/SECGPU + APPLE + MINI PCLOCAL OR CLOUD FALLBACK
ENTRY POINTS

Match from either side

MODEL FIRST

Pick an AI model

See memory estimates across quantizations and hardware.

Inspect →HARDWARE FIRST

Enter your rig

See which local models fit the memory ceiling.

Inspect →INTERACTIVE

Run Fit Matcher

Combine model, quantization, context, and device.

Inspect →
HOT PATHS

Buying decisions with consequences

24 GB

RTX 3090 vs RTX 4090

Same capacity, different generation and price.

Inspect →32 GB

What fits RTX 5090?

Map the 32 GB ceiling before spending.

Inspect →UNIFIED

Mac Studio for 70B models

Capacity can fit; speed and runtime still vary.

Inspect →
FIT MODELweights + runtime overhead + context reserve ≤ usable memory

Estimates are directional. Architecture, KV cache, runtime, context, batch size, and offload strategy change real usage.