How CPU recommendations work

Where does the data come from?

Benchmark scores, specifications, estimated USD prices, and first-seen dates come from PassMark CPU Benchmarks. Find Me a CPU collects and normalizes this data; it does not run these benchmark tests. Each CPU page links to the original record and shows when we retrieved it.

Prices are source estimates, not live retailer offers. Other currencies use USD conversion rates from Frankfurter. Local taxes, stock, shipping, and retailer prices can differ. Release / first seen means the first appearance on PassMark charts, not a confirmed launch date.

Which CPUs qualify?

Candidates must include the Desktop class, exclude the Server class, and have a first-seen year, positive price, and positive single-core and multicore scores. Names containing Xeon, EPYC, Threadripper, Opteron, Celeron, Pentium, or Atom are excluded. The finder targets consumer desktop builds rather than every processor in the benchmark database.

The CPU-only budget is a hard ceiling in USD. An exact budget overrides a preset. For new builds, budgets up to $150 accept first-seen years from 2022; up to $250 from 2023; higher or unlimited budgets from 2024. These cutoffs use the current year, 2026, minus 4, 2.5, or 2 years, rounded down to a whole year. Upgrades use an eight-year window.

When an upgrade socket is known, candidates must match its normalized socket name. When a current CPU is known, each available baseline benchmark must improve strictly, and known core and thread counts must be maintained or increased. Missing baseline fields cannot be checked. Socket matching alone does not verify chipset or BIOS support.

How is workload performance scored?

Each benchmark is divided by the highest corresponding score in the eligible group. This makes scores relative to the available candidates. For new builds, AMD and Intel are scored in separate groups, so their internal scores are not directly comparable.

The office TDP proxy is the smaller of 1 and 65 divided by typical TDP, or 0.5 if TDP is missing or zero. With two workloads, the primary contributes 72% and the secondary 28%. The gaming weights and X3D adjustment are heuristics, not measured game performance or a guaranteed frame-rate advantage.

How do price and priority affect the score?

• Value is workload score divided by estimated USD price, then divided by the highest value in the group.
• Affordability is 1 minus price divided by the budget, capped at a ratio of 1. With no budget ceiling, the highest eligible price is the denominator.
• Power is 1 minus typical TDP divided by the highest TDP, capped at a ratio of 1. Missing TDP contributes 65 W when finding that maximum and uses the maximum when scoring that CPU. If the maximum is zero, the power score is 0.5.

The final score is 55% workload score plus 45% priority score. Typical TDP is a specification used as a proxy; we do not measure electricity consumption.

Worked scoring example

These three synthetic CPUs illustrate mixed use, a $250 budget, and best-value priority in 2026. They are not real products or buying recommendations. Every row has a 2026 first-seen year and Desktop class. The table calls the same scoring function as the finder, before shortlist selection. Results are rounded to three decimals.

Scroll horizontally to see every score.

For Example B, both normalized benchmarks equal 1. Its workload score is therefore 1. Its power and affordability scores are 0 because it has the highest TDP and costs the full budget.

Why is the shortlist different from the score order?

The shortlist aims to offer different tradeoffs. It reserves the highest workload score as a performance option. For best match, it first seeks another CPU with at least 78% of that option's final score and 70% of its workload score, falling back to another candidate when necessary. It then seeks a cheaper alternative and the strongest remaining workload option, without duplicating CPUs.

If no cheaper CPU is available, the fallback may cost more than the best match. The labels describe selection roles, not guaranteed price or performance differences. Fewer than three eligible candidates retain score order. New builds show results only when both AMD and Intel have three selections, for six total; otherwise the finder asks you to adjust your answers. Upgrades return up to three results.

The database's default value sort is separate from the finder: it divides six times the single-core score plus the multicore score by estimated USD price. A position in that table is not a personalized recommendation.