At any given moment something is the limiting factor. If it were not, you would be looking at a machine whose parts are all idling — which would mean you overpaid for every one of them. A bottleneck is not a defect; it is the shape of a system that has been bought sensibly.

The question worth asking is not "do I have a bottleneck" but "is the limiting part the one I would want to be limited by, at the settings I actually use?"

Resolution moves the bottleneck

This is the single most misunderstood point, and it inverts the answer completely.

At low resolution the graphics card finishes each frame quickly, so the processor's ability to prepare frames becomes the limit — the system is processor-bound. Raise the resolution and each frame costs the card far more work while the processor's job barely changes, so the card becomes the limit — graphics-bound.

The same two components, unchanged, produce opposite conclusions at different resolutions. Any bottleneck figure quoted without a resolution attached is not telling you anything actionable.

A single bottleneck percentage compresses a complicated relationship into one number, and a lot goes missing on the way.

Averages hide the problem you actually feel

Average frame rate is a poor description of how a game feels. A system holding a steady rate feels smooth; one with the same average punctuated by irregular hitches feels broken. Frame-time consistency is what you perceive, and an averaged figure says nothing about it.

Processor limitations tend to show up precisely in this consistency rather than in the average — which is why a build can look balanced on paper and still feel wrong in the moments that matter, like a crowded scene or a sudden camera swing.

Workload matters more than the parts

A simulation-heavy strategy game, a competitive shooter at high refresh, and a cinematic single-player title at 4K place completely different demands on the same hardware. A pairing that is badly matched for one can be well matched for another.

The bottleneck detector takes a specific pairing and reports which side is limiting rather than producing a context-free score.

Once you know which side is limiting, the decision is mostly about cost and consequence rather than benchmarks.

Upgrading the graphics card

Usually the cleanest change: pull one part, fit another, done. The two things to check first are whether your power supply can feed it — including the connector question and the transient behaviour covered in the power supply guide — and whether it physically fits the case.

Upgrading the processor

Rarely just the processor. Depending on how old the platform is, it can drag in the motherboard and the memory with it, which changes the calculation from a component purchase into most of a new machine. Worth knowing before you start rather than halfway through.

The upgrade people skip

Memory speed and capacity are routinely overlooked and are sometimes the actual constraint, particularly on systems that have been carried forward across several years. It is the cheapest thing on this list to check and the easiest to fix.

If you are planning a change, the build simulator lets you test the combination for compatibility before any money is committed.

The bottleneck detector puts both parts on one performance index and divides the gap by the larger of the two. Within 5 % it calls the pair balanced; up to 15 % the limit is minor, up to 35 % moderate, and severe beyond that. Here is what that gives for six real pairings.

• Ryzen 9 9900X + RTX 5070 Ti: both sit at 8,200 — balanced.
• Ryzen 7 9800X3D + RTX 5070 Ti: 9,200 against 8,200, an 11 % minor limit on the graphics card. For gaming that is the side you want limiting, and it leaves room for a faster card later.
• Ryzen 5 5600 + RTX 5060 Ti: 4,900 against 5,600, a 13 % minor processor limit. A sensible budget pairing.
• Ryzen 5 7600 + RTX 4060: 5,900 against 4,300, a 27 % moderate graphics limit. The processor has headroom the card cannot use, which makes the card the obvious first upgrade.
• Ryzen 7 9800X3D + RTX 5090: 9,200 against 12,000, a 23 % moderate processor limit — with the fastest gaming processor in the catalogue. This is the resolution point from chapter one: at 4K, where a 5090 belongs, the card does the heavy lifting and the gap matters far less than the number suggests.
• Ryzen 5 5600 + RTX 5080: 4,900 against 9,800, a 50 % severe processor limit. This is the pairing the tool exists to catch: the money would have gone further split differently.

Turning the result into a decision

A minor result in either direction needs no action. A moderate one is worth acting on only if it is the processor and you play at 1080p with a high-refresh monitor, or if it is the card and you want more detail on screen. A severe result is the one to fix before buying, usually by moving money from the stronger part to the weaker one.