We tested 5 bottleneck calculators: what the test measures
The test asks one question: when a bottleneck calculator gives a number for a CPU and GPU pairing, how well does that number match what actually happens in games? The comparison point is real benchmark results — published, measured frame rates for the same pairing at the same resolution — not another calculator's opinion. A calculator is judged against measurements, not against other calculators.
The test does not ask which calculator is best in general, because calculators do not all measure the same thing. Some report a single percentage, some report a range, some report a verdict such as balanced or bottlenecked, and some change their answer by game or resolution while others do not. The test records what each calculator actually says, in its own terms, before any comparison is attempted.
It also does not measure anything that has not been measured yet. At the time of writing, the full measured results are still being collected: the pairs have been chosen, the method has been fixed, and the benchmarking is in progress. This article publishes the method so that the results, when they arrive, can be checked — not the other way round.
The 20 pairs chosen, and why these twenty
The twenty pairs were chosen to cover the situations where calculators disagree most, not the situations where they agree. If the test only used balanced, modern pairings, every calculator would look right; the interesting cases are the mismatches, the old processors with new cards, and the pairs that change behaviour with resolution. The pairs are grouped below, four in each group, so that each group tests a different kind of claim.
| Test group | Pairs in the group | What the group tests |
|---|---|---|
| Old CPU, strong GPU | Ryzen 5 3600 + RTX 4090; Ryzen 5 3600 + RTX 3070; Core i5-12400F + RTX 4090; Ryzen 5 5600 + RTX 4090 | Whether calculators catch a CPU limit at 1080p |
| Balanced modern pairs | Ryzen 7 9800X3D + RTX 5070; Ryzen 5 7600X + RTX 5070; Core i7-14700K + RTX 5070; Ryzen 7 7800X3D + RX 7800 XT | Whether calculators agree when the pair is balanced |
| Strong CPU, modest GPU | Ryzen 7 9800X3D + RTX 4060; Core i7-14700K + RTX 3060; Ryzen 7 7800X3D + RTX 4060; Ryzen 5 7600X + RTX 3060 | Whether calculators catch a GPU limit |
| Budget and older pairs | Ryzen 5 5600 + RTX 3060; Ryzen 5 3600 + RTX 2070; Ryzen 7 5700X + RTX 3070; Core i5-12400F + RTX 4060 | Whether calculators stay accurate on older hardware |
| Resolution changers | Ryzen 5 5600 + RTX 5070; Ryzen 7 5700X + RTX 5070; Core i5-12400F + RTX 5070; Ryzen 5 7600X + RTX 4090 | Whether the answer changes correctly with resolution |
Why different definitions of bottleneck percentage make comparison hard
The hardest problem in this test is not the benchmarking — it is that the five calculators do not define bottleneck percentage the same way. One calculator's percentage may mean the share of graphics performance lost; another's may mean the share of CPU time spent waiting; a third may report a score that is not a percentage of anything at all, but a label on a scale of its own. Comparing those numbers directly would be comparing different things.
The test handles this by recording each calculator's output in its own terms first, and only then translating it into a common question: does this calculator say the pair is CPU-limited, GPU-limited or balanced, at this resolution? That yes-or-no translation is cruder than the original numbers, but it is the only comparison that is fair across definitions — and it is the comparison that matters to a builder, who needs to know which part to upgrade.
The measurement method, step by step
The measurement method is fixed in advance, and it is deliberately boring. For each pair and each resolution, the same set of games is run at the same settings, and the frame rate and the CPU and GPU usage are recorded. The games are chosen to cover the genres where the limit differs — a competitive shooter, an open-world title, a simulation, and a demanding AAA game — because a pair that is balanced in one genre can be limited in another. See how esports and AAA games differ for why that matters.
The calculators are queried with the same pair and resolution, on the same dates, and their outputs are saved as they appeared — screenshots and saved text, not memory. If a calculator changes its answer during the test period, the change is recorded, not hidden. That is the only way the comparison stays honest.
What we can say so far — and what we cannot
What we can say so far is this: the method is fixed, the pairs are chosen, and the definitions problem is real — the five calculators do not define their headline number the same way, so any ranking that compares their percentages directly would be misleading. That finding alone is useful, because it tells builders to treat any single percentage with caution, including ours. Read how to read estimates at each resolution and what calculators can and cannot tell you for the background.
What we cannot say yet is which calculator is most accurate, which pair each calculator gets wrong, or by how much — because the full measured results are still being collected and have not been analysed. This article claims no ranking, no verdict and no numbers for any calculator. When the results are ready, they will be published with the raw data, including the pairs, the settings and the repeated measurements, so that the conclusion can be checked.
Until then, the honest advice is the same as it always was: use a calculator as a starting point, check it against published benchmarks at your resolution, and test your own machine in your own games. See how we built our dataset for where our own numbers come from.
Limitation: this article is a methodology explainer, not a results article. No measured test results exist yet, so no rankings, verdicts or accuracy numbers are claimed here; the full measured results are being collected and will be published with their raw data when they are ready.
Frequently asked questions
Have you finished testing the 5 calculators?
No. The method is fixed and the 20 pairs are chosen, but the full measured results are still being collected. This article explains the method and what the test will show; it claims no rankings, verdicts or accuracy numbers.
Why test 20 pairs and not more?
Twenty pairs cover the cases where calculators disagree most — old CPUs with strong GPUs, balanced modern pairs, strong CPUs with modest GPUs, older budget pairs, and pairs whose answer changes with resolution — while keeping the test small enough to measure properly at three resolutions.
Why is comparing bottleneck percentages hard?
Because the calculators do not define the percentage the same way. One may mean lost graphics performance, another may mean CPU waiting time, and a third may be a score on its own scale. Comparing those numbers directly would compare different things.
Which games will you use for the benchmarks?
A small set covering the genres where the limit differs — a competitive shooter, an open-world title, a simulation and a demanding AAA game — run at the same settings for every pair and resolution, with repeats recorded. The exact list will be published with the results.
When will the results be published?
When the measurements are complete and analysed, with the raw data — pairs, settings, driver versions and repeated measurements — published alongside them. Until then, no result is claimed, and this article will be updated to link to the results when they exist.