AI PC builds that choose VRAM before speed

For local models the graphics card's memory decides what runs at all, so an AI build ranks cards by VRAM first and only then by speed.

More graphics memory always wins

Cards are ranked by memory first and speed second, so a slower card with more memory beats a faster one with less. A model that spills out of graphics memory runs many times slower, so the engine never trades memory away for speed, at any budget. Nvidia only: CUDA is what the tooling expects.

A system-memory target before a processor upgrade

Model weights stage through system memory on the way to the card, so an AI build reaches a memory target scaled to the card's own memory before it spends on a same-platform processor tier. The one exception is a climb over the core floor, which comes first. The target rises as the card's memory rises.

The processor feeds the card, it does not race it

Inference is bound by memory bandwidth, not by processor cores, so the processor is chosen to keep the card fed and to handle data loading. It is still held to the same pairing rung as every other use case, so a low-core chip is never placed behind a card it would starve; beyond that, multi-core throughput breaks ties between otherwise equal builds. Two variants are built, Intel + Nvidia and AMD + Nvidia.