A bottleneck calculator knows two things about you: the names of two components. Everything it tells you is derived from those two names and whatever model sits behind the site.
That is a real constraint, and it is worth being concrete about what it rules in and out — including for the calculator on this site.
What a calculator can legitimately tell you
Which component runs out of capacity first, and roughly where. This follows from the specifications and a model of how they translate into frames. It is the core of the job and it is genuinely useful, particularly for parts you do not own yet.
How that changes with resolution. Because the two ceilings move differently — the processor’s barely at all, the card’s steeply — a calculator can show you the crossover point. This is the single most useful thing it does, and the one a forum answer usually gets wrong by not naming a resolution.
A comparison between options. “Would a 4070 or a 4070 Super suit this chip better” is a relative question, and relative questions survive modelling error well. If a model is 15% optimistic, it is usually 15% optimistic for both options, and the comparison still holds.
What no calculator can know
Your memory configuration. Dual channel or single, running at its rated speed or at the fallback, tight timings or loose. This moves processor-limited performance by a double-digit percentage and no calculator asks.
Your cooling and your case. A chip that throttles after ten minutes behaves nothing like the same chip with adequate airflow.
Your specific games and settings. The spread between a competitive shooter and a recent open-world title on identical hardware is several-fold. A single number represents a hypothetical average title that nobody plays.
Background load, drive speed, driver version, operating system. All real, none visible.
Variance between board partners. Two cards with the same chip can differ by several percent on clocks and power limits.
The gap between a modelled figure and your machine is therefore not small, and anyone presenting one as a prediction is overselling.
How to judge whether a calculator is trustworthy
Four questions, in order of how much they tell you:
Does it ask for a resolution? If not, it cannot be answering correctly, because the answer genuinely differs by resolution. This alone eliminates a lot of them.
Does it publish its method? A site that shows you the arithmetic is making a checkable claim. A site that produces a percentage with no explanation is asking for trust it has not earned.
Does it ever tell you not to buy anything? A calculator that always finds a problem is a sales funnel. Some pairings are fine, and a tool that cannot say so is not measuring anything.
Does it distinguish the average from the experience? Running short of video memory, inconsistent frame times and stutter are not visible in an average frame rate, and a tool that only reports an average will miss the thing that actually ruins a game.
How to read this one
Our calculator reports two ceilings rather than one verdict: what your processor can sustain, and what your card can render, at each resolution. The lower one is your frame rate. We lead with that because it is the honest shape of the answer.
Specifically:
The frame rate figures are modelled, not measured. They come from published specifications — shader counts, clocks, memory bandwidth, cache, core configuration — run through a coefficient model described in full on the methodology page. We do not republish anyone else’s benchmark scores, and we have not run our own. Treat the numbers as well-reasoned estimates with a real margin.
The percentage is the least useful output. We publish one because people look for it, but “how much of the combined potential goes unused” is not a decision you can act on. The frame rate, which side limits, and what a specific change would buy are the parts worth reading.
“Limited by” is the field to look at. It tells you which component to think about, and it is the most robust output of the model — getting the ordering right is far easier than getting absolute frame rates right.
The comparisons are more reliable than the absolutes. Use it to decide between two options rather than to predict a number.
Where we know the model is weakest
Stating this plainly because a methodology page nobody reads is not transparency:
- 4K figures are the least reliable. They sit at the end of the longest extrapolation, and a bug found in September had the video-memory penalty firing far too aggressively at 4K, which made several cards look much worse than they are. It is fixed. It is also the kind of error that is hard to catch without measured data to check against.
- Cache benefit is a single coefficient applied to every processor, when in reality it ranges from near-zero to very large depending on the game.
- Nothing is validated against measured benchmarks. The model is anchored to known product equivalences — pairs of parts widely agreed to perform similarly — which constrains it but does not verify it. This is the largest open weakness on the site and we are not going to pretend otherwise.
The thing that beats every calculator
If you already own the hardware, measuring it takes five minutes and is strictly better than any model, including this one. Open a game, watch GPU usage, drop the resolution and see whether the frame rate moves. The diagnosis guide walks through it.
Calculators are for hardware you do not own yet. For hardware in front of you, your own machine is the better instrument.