Methodology
Performance indexes
Each CPU carries a gaming index (0–100) and a multi-core index. Each GPU carries a 1440p index; the 1080p and 4K indexes are derived from it (1080p compresses the spread between cards because the CPU takes over; 4K widens it). The indexes are relative estimates that we assemble from published independent reviews and adjust by hand; they are not our own lab measurements, and they will drift as drivers and games change.
The bottleneck formula
For each resolution we estimate the gaming index a CPU needs to keep a given card fully busy: needed = a + b × GPU index, with a, b = 45, 0.62 at 1080p, 40, 0.60 at 1440p and 30, 0.50 at 4K. If the CPU's index is below needed, the CPU is the limit and the severity is 1 − CPU / needed. If it is above, the GPU is the limit and the severity is 1 − needed / CPU. The severity is the shortfall of the limiting part's index relative to what the pair needs (not how much faster that part would have to be, and not a measured FPS loss). When the CPU is the limit: under 10% balanced, 10–25% mild, 25–40% moderate, over 40% severe. When the GPU is the limit — the normal state for a gaming PC — we use a wider band: up to 20% balanced, 20–40% GPU-bound, over 40% the CPU is overkill for the card.
The constants were tuned so that pairings reviewers describe as well matched (for example a mid-range six-core with a mid-range card at 1440p, or a flagship X3D chip with a flagship card) land in the balanced band.
Power supply
Sustained load = the graphics card's measured average gaming power (from TechPowerUp — the card's own review, or its 2023 re-test chart for eleven older cards; linked on every PSU page) + CPU power (the manufacturer's TDP; for Intel 12th gen and later its maximum turbo power, which is above the base power — games usually draw less) + 75 W for motherboard, memory, storage and fans. A size (450, 550, 650, 750, 850, 1000, 1200, 1600 W) fits when the sustained load is at most 80% of the label and the card's measured 20 ms spike plus the rest of the system stays under the label. For the three cards with no 20 ms spike data (RTX 2070 Super, GTX 1650, GTX 1080 Ti) we require 70% instead. Safe = fits and meets the higher of the card maker's recommended PSU and TechPowerUp's minimum. Good unit = fits and meets TechPowerUp's minimum but is below that higher figure; only offered when spikes were measured.
Sources and check dates
Every figure on this site comes from one of the sources below. A date is the day the figures were last compared with that source, not the day the site was built. Where we have no record of such a comparison we show no date.
- Measured graphics-card power (gaming average, spikes, minimum PSU): TechPowerUp reviews. Last checked 2026-09-26.
- Board power and maker-recommended PSU: NVIDIA, AMD and Intel product pages. Last checked 2026-09-26.
- CPU core counts, clocks and sockets: Intel and AMD published specifications. No check date on record.
- CPU power figures: Intel ARK and AMD product pages (42 of 48 processors compared; those pairing pages name the figure and link its page, the other six name no source). Last checked 2026-10-06.
- Performance indexes and bottleneck scores: PCPairs estimates assembled from published reviews, not measurements. No check date on record.
Each graphics card's PSU requirements page links the exact TechPowerUp review and the exact NVIDIA, AMD or Intel page it was taken from; the power-supply paragraph of every pairing page links the TechPowerUp review. A processor's pairing page names which power figure it uses (TDP or maximum turbo power) and links the Intel ARK or AMD page it was compared with; the six processors not matched to such a page say so.
Limits
Real frame rates depend on the game engine, settings, memory speed, background software and drivers. Treat the score as a planning aid, not a promise.