CPU.NewsRelative score rel(x)=x/max×100 within the same form-factor pool; desktop CPU gaming 40% + single 25% + multi 25% + efficiency 10%; GPU raster 55% + RT 25% + efficiency 20%; PSUs tiered, not benchmarked.
This is a fully static site: all data lives in public/data/*.json and ships with each release — no runtime scraping, no price APIs. Each metric is defined as follows:
| Metric | Definition | Unit |
|---|---|---|
| CPU single / multi | Cinebench 2024 single / multi at stock power limits | pts |
| CPU gaming | 1080p high settings with a flagship GPU, multi-game average fps relative to the pool leader | relative |
| GPU raster / RT | 1440p (desktop) or 1080p (laptop) multi-game average fps, RT off / on separately | relative |
| iGPU | 1080p low settings multi-game average fps | relative |
| Memory | AIDA64 read bandwidth and latency with XMP/EXPO on | GB/s · ns |
| Storage | Vendor sequential and 4K random read IOPS; post-cache write is the steady state after the SLC cache is exhausted | MB/s · IOPS |
| CPU cooler | Noise-normalized thermal results from public reviews (Gamers Nexus / HWCooling / TechPowerUp) converted to in-pool relative scores; max noise and rated TDP from vendor specs | relative · dBA · W |
| Case | Vendor specs (dimensions, GPU / cooler clearance, radiator mounts, bundled fans); thermal and noise are in-pool relative estimates from airflow layout, intake area, bundled fans and dampening, cross-checked with public reviews (GamersNexus / TechPowerUp) | relative · mm · L |
| Monitor | Vendor specs (size, resolution, panel, refresh rate, brightness, gamut, ports); real-world response is estimated from panel type and refresh rate using typical review results (RTINGS / Monitors Unboxed), not the rated 1 ms | Hz · ms · nits · % |
| Motherboard | Vendor specs (VRM phases and per-phase amps, memory ceiling, M.2 / PCIe / LAN) plus public teardowns and VRM thermal tests, combined into tiers A–D | tier |
| Power | TDP for desktop CPUs, TBP for GPUs; typical cTDP / TGP for laptops | W |
Every product has a form field. Rankings, relative scores, recommendations and rivals are computed only within "same category + same form factor". Two RTX 5070 Laptop entries exist because their TGPs differ; the ladder and compare pages warn when form factors are mixed. Motherboards are pooled by socket (AM5 / LGA1851 / AM4 / LGA1700); boards for different sockets are not comparable.
rel(x) = round(x / max(pool) × 1000) / 10 // higher is better
rel(x) = round(min(pool) / x × 1000) / 10 // lower is better (latency)Example: the desktop CPU pool's highest Cinebench 2024 multi-core is 26,000 (Core Ultra 9 285K). The Ryzen 7 9800X3D scores 16,200, so its multi-core relative score = 16200 / 26000 × 100 = 62.3. Missing values are excluded from max and never shown.
Overall is only the default sort. Each component is normalized to 0–100 within the pool, weighted, then normalized again, so the pool leader is always 100.
| Pool | Weights | Why |
|---|---|---|
| Desktop CPU | Gaming 40 · single 25 · multi 25 · efficiency 10 | Gaming is the main reason to buy a desktop CPU; efficiency matters less on desktop |
| Laptop CPU | Single 25 · multi 30 · efficiency 30 · iGPU 15 | Battery and iGPU define the laptop experience; without an iGPU score the 15% goes to single + multi |
| GPU | Raster 55 · RT 25 · efficiency 20 | Most games are still raster-bound; cards without RT hardware redistribute the 25% |
| Memory | Bandwidth 50 · latency 50 | Games want latency, productivity wants bandwidth — split evenly |
| NVMe | 4K random 40 · post-cache 30 · seq. read 20 · endurance 10 | Daily feel comes from 4K and post-cache; sequential only shows in big copies |
| CPU cooler | Cooling 70 · quiet 30 | Air and AIO are separate pools; cooling decides which CPU it can handle, noise decides daily comfort |
| Case | Thermals 55 · quiet 20 · build 25 | Full / mid / mATX / ITX are separate pools; a case mostly decides system temperatures, then noise; build tier A–D counts as 100 / 85 / 70 / 55 (materials, finish, cable room, expansion) |
| Monitor | Motion 45 · image 55 | 1080p / 1440p / 4K+ / ultrawide are separate pools; motion = 1000 ÷ (1000/Hz + real response in ms), image = contrast 35 + HDR brightness 25 + DCI-P3 gamut 20 + pixel density 20; measured color accuracy and uniformity are not included |
| Motherboard | No score: tier A→D, then VRM phases × amps, then memory ceiling | Boards do not change fps; they differ in power headroom, cooling and expansion — see the tier criteria below |
| PSU | No score: tier A→D, then wattage | PSUs differ in electrical quality, not "performance"; 80Plus certifies efficiency only |
PSUs and motherboards are not benchmarked; they get tiers A–D instead. A tier answers "can this board / PSU reliably carry the CPU and GPU it is meant for", not "which is faster". Within a tier we sort by a measurable figure (wattage for PSUs, VRM capacity for boards), so order inside a tier is a reference, not a verdict.
| Board tier | VRM | Cooling / build | BIOS & expansion | Suited CPUs |
|---|---|---|---|---|
| A | ≥ 16 phases ≥ 90 A, or 14-phase 110 A-class SPS | Large finned heatsinks with heat pipes, 8+ layer PCB, backplate | Flashback, POST code / debug LEDs, ≥ 4 M.2 (incl. Gen5), 2.5GbE+, Wi-Fi 7, USB4 / Thunderbolt | Any CPU, including 16-core and unlocked-power overclocking |
| B | 12–16 phases at 60–80 A with thermal headroom | Solid heatsinks over all stages, 6-layer PCB | Usually Flashback, 2–4 M.2, 2.5GbE, Wi-Fi 6E / 7 | Any CPU at stock power, including 9950X / 285K |
| C | 8–12 phases at 50–60 A, or doubled DrMOS | Small heatsinks or partial coverage | 1–2 M.2, often no Flashback, 1GbE / 2.5GbE | ≤ 8-core or 65–105 W CPUs; 12+ cores will thermal throttle |
| D | 4–6 phases ≤ 50 A, no doubling | No VRM heatsink or decorative only | 1 M.2; budget H610 / A520 / A620 | Only 65 W six-core and below; high-power CPUs throttle or shut down |
| PSU tier | Electrical quality | Typical platforms | Note |
|---|---|---|---|
| A | Ripple < 30 mV, strong transient response, full protections, Japanese primary caps, 10-year warranty | Seasonic Prime / Vertex / Focus GX, Corsair RMx / HX / AX / SF, Super Flower Leadex, be quiet! Dark Power / Straight Power | For flagship GPUs and overclocking |
| B | Ripple < 50 mV, complete protections, 5–7-year warranty | MSI MPG / MAG, Thermaltake GF3, Cooler Master MWE Gold V2, ASUS TUF, be quiet! Pure Power 12 M, Great Wall X, DeepCool PN | The mainstream pick |
| C | Ripple 50–80 mV, basic protections, 3–5-year warranty | Corsair CX, Gigabyte UD, Antec NeoECO, mainstream Segotep / Huntkey / Golden Field / SAMA | Fine for entry and iGPU builds; avoid 300 W+ GPUs |
| D | Out-of-spec ripple or missing protections, inflated ratings, old double-forward / group-regulated designs | White-label, no-name, decade-old entry units | Not recommended with a discrete GPU |
Entries from 2007–2023 are marked "est.". Older hardware has no Cinebench 2024 or modern RT data, so we take its ratio to an "anchor" product in public reviews (e.g. the i7-2600K is ~60% of the i7-8700K in Cinebench R15) and multiply by the anchor's score in the current pool. This is fine for positioning, not for precise comparison: software, drivers and games differ across eras, so expect ±10–15% error.
Calibration: fill real review data into data-src/calibration.csv and run npm run calibrate; the "est." flag is cleared for those entries and meta.version increments.
Architecture-level facts (core microarchitecture, process, chip design, L1, ISA, SMT, shaders per unit, RT / AI unit generation) are derived per generation; product-level facts (die layout and area, transistors, L2, clocks, PPT / PL2, memory controller, PCIe, SM / CU, TMU / ROP, memory speed, launch price) are entered per product from public specs. Transistor density and compute per watt are derived.