We run a bottleneck calculator, so this may be an odd article to publish. But a tool is only useful if you know what it is measuring, and most of the arguments about bottleneck calculators come from people expecting a kind of answer no calculator can produce.
The methodology page covers the formula. This covers the limits.
What the number actually is
We model the maximum average frame rate each part can produce on its own at your chosen resolution, from published benchmark averages across recent AAA titles. Then we compare the two and report how much of the faster part’s headroom goes unused.
So the number is: how much of one component’s average-case potential is unused when paired with the other, across a basket of games, on a reference test system.
That is a genuinely useful thing to know. It is not the same as “how many frames will I get in the game I play tonight”, and the gap between those two is where every complaint about calculators lives.
The four things a bottleneck calculator cannot know
1. Your specific game
A basket average smooths over enormous variation. The same pair of parts can be 5% CPU limited in a corridor shooter and 40% CPU limited in a late game strategy save or a large multiplayer map. Simulation, strategy, MMOs and anything heavily modded sit far outside the average; well optimised single player titles sit on the other side of it.
If you only play one game, look up a benchmark of that game rather than trusting any basket average, ours included.
2. Frame consistency
We model averages, because published data is dominated by averages. Averages say nothing about stutter, and stutter is what people actually feel. A modest CPU deficit on paper can be a very visible one in practice. That is the subject of why 1% lows matter more than average FPS.
3. Your machine’s condition
The model assumes a healthy system. It cannot see that your memory is running at 4800 instead of 6000 because XMP was never switched on, that you have one stick instead of two, that your case is choking the card thermally, that Windows is on the balanced power plan, or that a frame limiter is quietly pinning you at 60.
Any one of these is worth more than the difference between two adjacent processors, and all of them are free to fix. The checklist is in how to tell if your GPU is being held back.
4. VRAM and settings
We model performance at stock settings. If you turn on ray tracing, run textures beyond what the card’s memory holds, or use upscaling, the card’s side of the equation changes and the balance moves with it. A card that runs out of video memory does not degrade gently, and no percentage captures that — see running out of VRAM is not the same as a bottleneck.
Where the number is genuinely reliable
Three questions, all of which it answers well:
- Shortlisting a purchase. “Will my current processor waste an RTX 5080?” is exactly the question you cannot test on hardware you do not own. It is also the question behind how to split your budget between CPU and GPU.
- Comparing two options against the same partner. The relative ordering is far more robust than the absolute figure.
- Understanding the effect of resolution. Switching between 1080p, 1440p and 4K on the same pair shows a real mechanism, not an artefact. See why resolution changes the answer.
How to read a percentage sensibly
Treat it as a band, not a decimal.
| Result | What it means | Action |
|---|---|---|
| Under 5% | Balanced | None |
| 5 to 15% | Minor | None needed |
| 15 to 30% | Noticeable | Worth investigating |
| Above 30% | Significant | Worth acting on |
The difference between an 11% and a 13% result is noise, and anyone presenting bottleneck figures to two decimal places is selling certainty that does not exist.
Also note which direction the limit runs. A GPU limit is the healthy state for a gaming PC and generally needs no action. A CPU limit is the one worth investigating.
Be sceptical of any calculator that
- Ignores resolution. The single biggest input into the answer.
- Returns one number for all workloads. Games are not one workload.
- Tells you a bottleneck is a defect to be eliminated. In every PC something is the limit; the goal is choosing which.
- Will not show you its method. Ours is written down, including where it is weak.
Use it for the right job
An estimate is for hardware you do not own. An overlay is for hardware you do. Use the bottleneck calculator to shortlist, then measure on your own machine before spending anything.
FAQs
Are bottleneck calculators accurate?
They are accurate at the job they are built for: estimating how two components match up on average, across a basket of games, at a given resolution. They are not accurate as a frame rate prediction for one specific game on your specific machine, and no calculator can be.
What is a good bottleneck percentage?
Anything under 15% is fine. 15 to 30% is noticeable but not urgent. Above 30% is worth acting on. Treat the figure as a band rather than an exact value — small differences are noise.
Is a 10% bottleneck bad?
No. Ten percent sits in the minor band and is normal in a well matched system. It is not a fault and does not need fixing.
Why do different bottleneck calculators give different results?
Because they use different benchmark data, different baskets of games and different formulas. Some ignore resolution entirely, which produces wildly different answers. The method matters more than the number, which is why ours is published.
Should I trust a bottleneck calculator over a real benchmark?
No. A benchmark of your game on comparable hardware is always better evidence. Calculators are for hardware you cannot test right now.
Can a bottleneck calculator detect a VRAM problem?
No. Bottleneck models compare processing speed. Running out of video memory is a capacity limit that shows up as stutter rather than a lower average frame rate, and no percentage will reveal it.
My calculator result looks bad but my games run fine. What should I do?
Keep playing. The percentage describes a reference system across a basket of titles, not your actual experience. Measured performance always beats an estimate.