How my tools work

All hardware information lives in a MongoDB database I maintain by hand. There's a separate collection for each component class — CPUs, GPUs, motherboards, RAM, PSUs, storage, coolers, cases, and games — and each record stores the manufacturer-published spec sheet (cores, clocks, cache, TDP, VRAM, memory bandwidth, socket, release year, MSRP, and so on) alongside a few derived fields that the calculators use.

The derived fields — most importantly each part's performanceScore — come from publicly available benchmarks that I cross-reference across multiple reviewers (TechPowerUp, Tom's Hardware, Hardware Unboxed, Gamers Nexus, PassMark) and normalise to a fixed reference point. I don't run my own test bench. Anything that looks like raw FPS in the tools is computed from those normalised scores plus per-game baseline data — never hand-typed per card.

When a page asks for hardware data it goes through one endpoint, /api/hardware?type=<cpu|gpu|game|...>, which returns the relevant collection. Server-rendered pages cache the response for an hour.

The FPS Calculator uses a single, locked baseline: an RTX 3060 paired with a Ryzen 5 5600X playing each game at 1080p High. For every game in the database I store the average FPS that combo produces, and the calculator scales from there.

The pipeline is: take the game's 1080p-High baseline, multiply by a preset factor (Low ≈ 1.54×, Medium ≈ 1.22×, High = 1.0×, Ultra ≈ 0.80×), then by a resolution factor that depends on how demanding the game is (a demanding title at 4K runs around 0.35× of its 1080p number; an esports title closer to 0.45×). The GPU you picked then scales the result by its performanceScore relative to the RTX 3060, and the CPU adjusts the result based on how CPU-bound that game is — esports titles weight the CPU around 50%, ultra-demanding games around 15%. Ray tracing applies a 0.55× penalty; DLSS adds roughly 1.40×, FSR around 1.30×. A game-specific FPS cap (where one exists) trims the final number.

The output is an estimate, calibrated to the same fixed baseline. It's closer to average benchmark FPS than to any one reviewer's exact number — useful for planning a build, not a substitute for benchmark videos of your specific configuration.

The PSU Calculator builds a power budget from the components you select, then sizes the PSU with enough headroom to handle transient spikes and aging.

The breakdown adds: CPU sustained power (TDP from the spec sheet), GPU sustained power (TDP), 0.5 W per GB of RAM, at least 7 W per storage drive (whichever's higher between that and the per-drive estimate), 3 W per case fan, and a fixed 35 W motherboard baseline with a small allowance for USB devices. If you flag overclocking, that whole subtotal gets a +10% bump; if you flag “plan for an upgrade,” another +30%.

From there, peak load is 1.2× the subtotal to model bursts, and the recommended capacity is another 1.3× on top of peak — roughly 25–30% real-world headroom. The final number is rounded up to the next standard PSU tier (450, 550, 650, 750, 850, 1000, 1200, or 1500 W). That rounding policy is deliberate: it's safer to slightly over-spec than to land between tiers, where pricing rarely rewards you anyway.

GPU Compare shows VRAM and memory type, base/boost clocks, memory bandwidth and bus width, shader-core count (CUDA / stream processors), RT and tensor cores, TDP and recommended PSU, architecture, release year, and launch MSRP. The headline performance numbers are two normalised scores out of 100: an aggregate raster performanceScore and a separate rayTracingScore. Both are scaled so an RTX 3060 sits at the calibration point used by the FPS engine.

CPU Compare shows cores and threads, base/boost clocks, L2/L3 cache, TDP and peak power, socket, generation, release year, and MSRP. Each chip carries a gamingScore and a multiThreadScore (both 0–100), normalised to a Ryzen 5 5600X gaming baseline.

When the tool prints a verdict, it does so by averaging the calculated FPS across a basket of games at 1440p, comparing the percentage gap, and weighing that against the TDP delta and FPS-per-dollar (using launch MSRP as the price input). I avoid weighted “total scores” that hide the math — what you see in the table is what the comparison was actually based on.

This is a solo project, so updates aren't on a fixed cron. There's no auto-refresh process pulling new prices in nightly. What there is: a set of seed scripts I run manually when I add new hardware or revise existing scores, plus a calibration pass on game baselines when a major game patch or driver release noticeably moves performance.

In practice, that means new GPUs and CPUs land within a few weeks of launch — sooner if I catch the launch coverage early — and existing entries get reviewed when I'm already touching that area of the database. MSRP is recorded at launch and isn't updated to track street prices, because street prices move daily and I'd rather show one honest reference than a stale “current” number. If something looks out of date or wrong, tell me and I'll fix it faster than any schedule would.

These tools produce estimates, not measurements of your machine. A few specific things to keep in mind:

• FPS numbers are aggregates calibrated to a fixed RTX 3060 + R5 5600X baseline. Driver versions, RAM speed and timings, storage, thermal headroom, background processes, and in-game settings beyond the preset all shift real results — typically by single-digit percentages, occasionally more.
• The PSU calculator includes 25–30% real-world headroom on top of peak draw, but it doesn't model extreme overclocking, mining-style sustained loads, or non-standard accessories. If you're pushing hard on those fronts, treat the recommended wattage as a floor, not a ceiling.
• CPU bottleneck detection in the FPS engine only fires at 1080p with high-end GPUs. At 1440p and 4K, I treat the workload as GPU-bound by default — which matches how real benchmarks behave but does mean CPU choice looks less impactful at higher resolutions than it sometimes is.
• Pricing is launch MSRP. Street prices for older parts can be 10–40% lower or, during shortages, much higher. Always cross-check before buying.
• Comparison numbers are based on reference-clock specs. AIB cards with factory overclocks, undervolted CPUs, and laptop variants all behave differently than the desktop reference they're labelled as.

For tested numbers on a specific configuration, cross-reference with TechPowerUp, Hardware Unboxed, or Gamers Nexus. These tools are a starting point for planning; reviews are the ground truth for hardware-tested numbers.