Interactive hardware compass

Which AI PC fits your local LLM?

Four inputs produce a useful direction. The finder estimates model weights, context cache and runtime headroom, then explains why a memory class fits.

Your profile

Calculate your needs

All results are guidance for quantized inference, not a performance guarantee.

1 Which model size do you want to run locally?

7B–8Bfast entry point 12B–14Bmore quality 20B–32Bdemanding 65B–70Blarge model 100B+professional class

2 · Desired context lengthLong context substantially increases KV cache. 3 · Main workloadMore components require additional memory. 4 · Highest priorityThis determines the recommended hardware path. Approximate system budgetBudget does not change memory needs, only the realistic route.

Your guidance

16 GB of fast memory as a starting point

Estimated need7–9 GBModel + context + headroom

Recommended class16 GB VRAMdiscrete GPU

Suitable directionMidrange GPU PCwith at least 32 GB system RAM

Why this recommendation?

Compare suitable systemsVerify the exact memory and GPU configuration.

View systems

What the calculation includes

Conservative headroom

typical Q4 weight size for the model class
allowance for longer KV cache
runtime and selected workflow reserve
rounding up to a realistic memory class

What no calculator knows for sure

Model and software still matter

architecture and exact quantization
Flash Attention and backend support
multimodal encoders and concurrent users
how much unified memory the system exposes

Next step

Understand the memory values behind your result

Our VRAM guide explains weights, KV cache, offloading and unified memory without unnecessary jargon.

Read the VRAM guide