Every number on this site is a model, not a measurement. We do not own the hardware we write about and we have never benchmarked your exact combination of parts. This page explains exactly how the figures are produced, where the underlying data comes from, and what they cannot tell you. If anything here is unclear or wrong, tell us and we will correct it.
The idea behind the model
A game needs two things to happen for every frame. Your processor works out what should be on screen, and your graphics card draws it. Those two jobs run at different speeds, and whichever finishes slower sets your frame rate.
The important part, and the part most calculators get wrong, is that these two speeds respond differently to resolution. A processor prepares roughly the same number of frames per second whether you play at 1080p or 4K, because the work of simulating the game does not change with pixel count. A graphics card is the opposite: at 4K it has to fill four times as many pixels as at 1080p, so its ceiling falls sharply.
So instead of applying a fixed CPU and GPU weighting per resolution, which is what the other calculators in this niche do, we estimate each part’s frame ceiling separately and compare them. The limit then moves from the processor to the card on its own as resolution rises, because that is what physically happens.
The formula
For a given resolution and graphics preset:
- CPU frame ceiling = processor index × 1.80 × preset factor × memory factor
- GPU frame ceiling = card index × resolution factor × preset factor × VRAM factor
The processor index is a relative gaming performance figure where the Ryzen 7 9800X3D sits at 100. The card index is relative raster performance at 1440p where the RTX 4090 sits at 100. The constant 1.80 calibrates the scale so that an index of 100 lands at roughly 180 frames per second on the High preset in a modern AAA title.
The resolution factors are 2.60 at 720p, 1.90 at 1080p, 1.40 at 1440p, 1.15 at ultrawide 1440p, 0.80 at 4K and 0.35 at 8K. Preset factors move the card a lot and the processor almost not at all, which matches how settings actually behave: Low is 1.70 for the card but only 1.10 for the processor.
Your frame rate is the lower of the two ceilings, reduced slightly when the two are close, because two parts working flat out at the same time contend for the same frame budget.
What we call a bottleneck, and what we do not
This is where our figures differ most from other calculators. A graphics card being the limiting part is the normal, correct state of a well built gaming PC. Calling that a fault, as most calculators do, produces an alarming percentage for a build that needs nothing changed.
We compare the ratio of the two ceilings against a healthy band. Below a ratio of 1.25 the processor cannot keep the card fed and graphics money is being wasted, so we report a CPU bottleneck. Above a ratio of 2.00 the card is so far behind that processor money is being wasted, so we report a GPU bottleneck. Between the two, the build is correct and we say so rather than inventing a problem.
We also tell you whether the gap matters. A 30% imbalance at 170 frames per second is something you will never perceive. The same 30% at 45 frames per second ruins the game. A percentage with no frame rate attached is not useful information.
Memory, VRAM and the warnings
System memory is applied to the processor ceiling: 8 GB costs around 18% in modern titles, 12 GB around 7%, and 16 GB in dual channel is treated as the baseline. Beyond 16 GB there is a marginal gain for gaming only.
Video memory is applied to the card ceiling. If the card has less VRAM than the resolution and preset want, the ceiling is reduced proportionally and a warning appears, because running out of video memory shows up as frame time spikes rather than a lower average, and an average alone would hide it.
How the PSU figures are produced
Peak processor draw is modelled at 130% of rated TDP, because modern chips boost well above their nominal figure. Graphics card draw is taken at its rated TDP, because for current cards that rating already is total board power. We then add 50 W for the motherboard, 5 W per memory stick, 8 W per SSD, 12 W per hard drive, 3 W per fan and 15 W for a liquid cooler pump.
The recommendation sizes that total at roughly 75% load, which leaves efficiency headroom and room for one graphics upgrade. The minimum figure sizes it at 90% load. Both are usually lower than the number a card or board maker quotes, because those assume a cheap supply whose rating is overstated. On a reputable 80 Plus unit our figure is the one that applies.
Cards rated at 250 W or more that were released from 2020 onward draw brief transient spikes far above their rating, and an older ATX 2.x supply can trip its own protection at a wattage that looks safe on paper. Where that applies we say so and recommend an ATX 3.0 or 3.1 unit rather than simply inflating the number.
Where the data comes from
The performance indexes are built from published benchmark data rather than our own testing. The sources we work from are:
- TechPowerUp GPU Database, for card specifications and relative performance
- PassMark, for processor single thread and multi thread figures
- UL 3DMark, for synthetic graphics scores
- Manufacturer specification sheets from AMD, Intel and NVIDIA, for core counts, clocks, TDP and memory
Where published sources disagree, we take the more conservative figure. Every index in our database is checked against at least two of these sources before a page using it is published.
What these numbers cannot tell you
- They are an average across typical modern games. A strategy title leans far harder on the processor than an open world does, and the same PC can be CPU limited in one and GPU limited in the next.
- They do not model your specific memory timings, your cooling, your driver version or your background software, all of which move real results.
- They are not a benchmark. If you want to know what your PC does, measure it with an overlay while you play. Use this site to work out which part to replace first.
- Game specific and title specific figures are estimates of the same kind, not captured runs.
Corrections
If a figure here looks wrong against a review you trust, we want to hear about it. Send us the source and we will check it and publish a correction. Our hardware database is a single file we update deliberately rather than a feed we leave to drift, and every page records when its data was last reviewed.