How GamerKit's calculators work: our methodology, and where they can be wrong
We built GamerKit's calculators to give a quick, honest answer to common PC questions. They are models, not benchmark databases. A model is only useful if you know what it assumes, so this page documents how each estimate is produced, how we calibrate it, and the situations where it will be off. If you find a result that disagrees with your real hardware, please tell us: viewer reports have already changed the model more than once.
What kind of tool this is
GamerKit does not run games or store thousands of measured benchmark results. Instead each calculator uses a simple, explainable model built from published specifications and widely reported relative performance, and then applies it to the parts you choose. That makes the answers instant and consistent, and it lets us show the reasoning. The cost is precision: expect a directional answer, typically within a band of roughly 15 to 25 percent of what a careful benchmark measures, not an exact frame count.
The FPS checker
Every game has an engine-demand tier from 1 (very light, such as an esports shooter) to 10 (very heavy, such as a path-traced showcase). The tier sets the expected frame rate on a reference graphics card, at 1080p on High settings with no ray tracing or upscaling. The reference card is performance index 60, which is roughly an RTX 4070.
| Demand tier | Reference FPS (index 60, 1080p High) |
|---|---|
| 1 | 300 |
| 2 | 240 |
| 3 | 180 |
| 4 | 140 |
| 5 | 110 |
| 6 | 90 |
| 7 | 70 |
| 8 | 55 |
| 9 | 45 |
| 10 | 35 |
Your graphics card is placed on the same index. The RTX 4090 anchors the scale at 100, the RTX 4070 is 60, the RTX 3060 is 38 and the GTX 1650 is 16. The estimate is then the reference FPS multiplied by your card's index divided by 60, adjusted for resolution and quality:
- Resolution divides the result: 720p ×0.70, 1080p ×1.00, 1440p ×1.28, 4K ×1.62 (these are compressed pixel-count ratios, because frame rate does not fall in step with pixels).
- Quality multiplies it: Low ×1.55, Medium ×1.22, High ×1.00, Ultra or ray-tracing ×0.72.
For games that are limited by the processor as well as the graphics card, such as simulation, survival and large multiplayer titles, the estimate is also capped by a CPU ceiling: the reference FPS multiplied by your CPU's index divided by 65, times 1.15, scaled slightly by the quality setting. The lower of the GPU result and the CPU ceiling is shown. This is why a weak processor can hold back a strong card in these games.
Results are labelled by what they mean in play: under 30 FPS is rough, 30 to 59 is playable, 60 to 99 is good, 100 to 164 is great and 165 and above is overkill. Some games also have hard requirements the maths cannot model. For example, Gears of War: E-Day requires hardware ray tracing, so older GTX cards are shown as unable to launch rather than given a frame rate.
How we calibrate it
We anchor each game's tier to published requirements and widely reported results. If a publisher says an RTX 3060 Ti is recommended for 60 FPS at 1080p, the tier is set so the model agrees. We also re-check games when users report that the numbers look wrong. In October 2026 we recalibrated several titles after viewer feedback: Cyberpunk 2077, Call of Duty: Warzone, Baldur's Gate 3, Alan Wake 2, Black Myth: Wukong and Planet Zoo 2 were all moved, and Fortnite is now treated as CPU-sensitive, which affects older processors.
In these tools, High means High without ray tracing or upscaling. Ray tracing is covered by the Ultra option, which carries a larger penalty.
The PSU calculator
The PSU calculator adds up typical power draw for the parts you pick: the processor and graphics card (increased by 15 percent if you tick overclocking), 5 W per memory stick, 7 W per NVMe drive, 5 W per SATA drive, 2 W per fan, 15 W for an AIO pump, and a fixed 55 W for the motherboard and 10 W for miscellaneous items. That total is the expected peak load.
The recommended size is the smallest standard PSU where that load sits at about 55 percent, which is the efficient part of the curve and leaves room for short graphics-card power spikes, which can briefly reach about twice the rated draw. The minimum shown is the smallest size where the load is at most 85 percent. High-end recent Nvidia cards are flagged as wanting a native 12V-2x6 (12VHPWR) cable.
The bottleneck calculator
Each processor and graphics card has a relative performance index. The tool compares them at your resolution, giving the processor extra effective headroom as resolution rises, because higher resolutions put more of the work on the GPU. If the processor's effective index is below the card's, the gap (scaled by 0.9 and capped at 60 percent) is reported as a CPU bottleneck. If the card is the limit, which is the normal and desirable state in gaming, the gap is dampened by 0.4. The percentage is then labelled balanced, minor, moderate or severe.
What the models do not do
- They do not model video memory. A card with too little VRAM for a game's highest textures can perform worse than the index suggests.
- They do not know about driver versions, game patches, background software or thermals.
- They do not model upscaling (DLSS, FSR, XeSS) or frame generation. Using them usually raises FPS.
- Performance indexes are approximate and refreshed as new hardware appears. Prices shown in some tools are indicative UK street prices and change often.
- Games vary by scene. A town and a boss fight in the same game can differ by a large margin.
Reporting a problem
If a result looks wrong for your hardware, send us the parts, resolution, settings and the frame rate you actually see through the contact page. We use these reports to recalibrate games and add missing parts. Several processors and graphics cards in the tools exist because viewers asked for them.
Worked example 1: a single FPS estimate
Take a Ryzen 5 5600 with an RTX 3060 playing Cyberpunk 2077 at 1080p on High. Cyberpunk is a demand tier 5 game, so the reference frame rate is 110 FPS on the RTX 4070-class card (index 60). The RTX 3060 sits at index 38, so the estimate is 110 × 38 ÷ 60 = about 70 FPS. High has a quality multiplier of 1.0 and 1080p a resolution multiplier of 1.0, so nothing else changes the result. Cyberpunk is not flagged as processor-limited, so there is no CPU cap.
| Setting changed | Estimate | How the number moves |
|---|---|---|
| 1080p, High (baseline) | 70 FPS | 110 × 38 ÷ 60 |
| 1080p, Medium | 85 FPS | × 1.22 |
| 1080p, Ultra with ray tracing | 50 FPS | × 0.72 |
| 720p, High | 100 FPS | ÷ 0.70 |
Worked example 2: when the processor is the limit
Now put the same Ryzen 5 5600 (index 58) with a much stronger card, an RTX 4090 (index 100), in Call of Duty: Warzone at 1080p on High. Warzone is demand tier 4 (reference 140 FPS) and is flagged as processor-limited. The graphics-card result would be 140 × 100 ÷ 60 = about 233 FPS. The processor ceiling is 140 × (58 ÷ 65) × 1.15 = about 144 FPS. The lower number wins, so the estimate is 144 FPS, and you can see that the RTX 4090 is mostly idle. This is the situation the Bottleneck Calculator is designed to flag.
Worked example 3: PSU sizing
A Ryzen 5 7600 with an RTX 4070, two memory sticks, one NVMe drive and four fans. The tool uses 105 W for the processor and 200 W for the graphics card, then adds 10 W for memory, 7 W for the drive, 8 W for fans, 55 W for the motherboard and 10 W for miscellaneous parts. The peak load is 395 W. Dividing by 0.55 gives about 718 W, which rounds up to the next standard size of 750 W, where the load sits at about 53 percent. The minimum (load at 85 percent) is 500 W. The recommendation leaves room for transient spikes, and a Gold-rated unit at that size is efficient at the typical load.
Worked example 4: bottlenecks and resolution
A Ryzen 5 5600 paired with an RTX 4070 at 1080p is reported as well balanced, with a 3 percent CPU gap. A Ryzen 5 3600 with the same card at 1080p is a 18 percent bottleneck, labelled moderate, because the older processor cannot feed the card. Move the same pairing to 4K and the gap falls to about 9 percent, because the graphics card has more work per frame and the processor matters less. This is why advice to upgrade a processor depends on your resolution.
| Pairing | 1080p | 1440p | 4K |
|---|---|---|---|
| Ryzen 5 5600 + RTX 4070 | 3% (balanced) | 8% (minor) | Not an issue |
| Ryzen 5 3600 + RTX 4070 | 18% (moderate) | Lower | 9% |
Change log
- 22 September 2026: quality presets (Low, Medium, High, Ultra) added after viewers pointed out that identical hardware can look 'wrong' when the settings are not stated.
- 3 October 2026: 720p added, and older processors and graphics cards (GTX 900 and 10 series, RX 500 series, Ryzen 1000 to 3000 and older Intel parts) added after viewer requests. A hard ray-tracing gate was added for Gears of War: E-Day.
- 3 October 2026: several games recalibrated (Cyberpunk 2077, Call of Duty: Warzone, Baldur's Gate 3, Alan Wake 2, Black Myth: Wukong, Planet Zoo 2). Fortnite became processor-sensitive.
- 12 and 13 October 2026: Roblox DevEx rates updated to the standard $0.0038, legacy $0.0035 and US 18+ $0.0054 rates in the Robux tools.
How to sanity-check any estimate
- Compare the number with a benchmark video or review of the same game and a similar graphics card at the same resolution and settings.
- Check that the settings match. Ultra and ray tracing can halve the frame rate that High delivers.
- Remember that averages hide the lows. A game averaging 60 FPS can still stutter in busy scenes.
- If your result differs by more than about a quarter, tell us. That is the kind of report that improves the model.
FAQ
How accurate is the GamerKit FPS checker?
It is a directional model. Expect results within roughly 15 to 25 percent of a careful benchmark for popular games, with larger gaps for unusual settings, upscaling or very new titles.
Why does the checker say a game will not launch?
Some games have hard requirements, such as hardware ray tracing or TPM and Secure Boot. The model flags these separately from the frame rate.
Do the calculators use real benchmark data?
They use published specifications and widely reported relative performance to build a model, then calibrate it against known results. They do not store a database of measured benchmarks.
How can I report an incorrect estimate?
Use the contact page with your parts, resolution, settings and the real frame rate you see. Reports are used to recalibrate the model.
Why does the checker give the same FPS at different resolutions?
When a game is processor-limited, the CPU ceiling can be lower than the graphics-card result at every resolution, so changing resolution does not change the shown number until the card becomes the limit.
Does the PSU calculator include a safety margin?
Yes. The recommended size puts your peak load at about 55 percent of the PSU's rating, which keeps it efficient and leaves room for short power spikes.