Methodology
This page explains how our recommendations and calculators work, and — just as importantly — where the numbers are estimates. We would rather be transparent than imply a precision that does not exist in PC hardware.
Build recommendations
Our five budget-tier builds reflect the 2026 consensus 'meta' from reputable build guides and community discussion, chosen to balance intended workloads, without guaranteeing a well-balanced result. They are starting points, not the only correct answer; prices and availability shift constantly.
Performance index
Each CPU and GPU has a relative performance index from 0 to 100, used to estimate balance and resolution suitability. It is a curated estimate of typical gaming performance based on architecture, tier and known relative standing — not a measured benchmark. Real-world differences vary by game, resolution, settings and drivers, so treat the index as a guide, not a guarantee.
PSU wattage
Power estimates add GPU board-power and CPU allowances to supporting-component allowances, multiply the component sum by 1.25, then round up to a listed capacity class. Estimated load is component sum divided by capacity, before the buffer. Per-GPU pages also apply their stated starting capacity and any separately verified manufacturer minimum as floors. A fixed buffer cannot certify transient response, noise or efficiency. Always check the exact GPU and PSU documentation.
SFF clearance
Published case dimensions are stored with their sources; an unpublished dimension remains unknown. Graphics-card dimensions vary by AIB model, so our 'does it fit' pages use a representative compact version of each card; the interactive checker lets you enter your exact card. Always verify your specific model's length, height and slot width before buying.
Bottleneck verdicts
A bottleneck is dynamic, not a fixed percentage. Our estimator weights CPU and GPU by resolution and game type and reports honest balance bands — balanced, slightly limited, or severely mismatched — rather than a fake single number.
Cost, VRAM and latency tools
The cost calculator uses 2026 price estimates that vary by brand and region. The VRAM and RAM-latency tools use published formulas; results are close estimates intended to guide buying decisions, not exact guarantees.
Updates and corrections
We update data as new hardware launches and as the market shifts. Spotted something wrong? Tell us via the contact page and we will fix it.
Local AI memory estimates
The text-LLM calculator estimates weight storage and a full KV cache from explicit architecture fields. Quantized cache storage includes ggml block scales. Runtime/desktop allowance and percentage margin are editable assumptions; neither is a measured system requirement. Context outside a profile’s range is flagged as unverified. Gemma 2 uses a full-cache upper bound. See the formulas, examples and source scope. The separate ComfyUI guide provides a user-run diagnostic procedure and claims no GPU benchmarks.