Bottleneck Calculator

What is a bottleneck calculator?

A bottleneck calculator is a free tool that takes a processor and a graphics card, compares how fast each one is relative to the other, and tells you which of the two is holding the other back and by roughly how much. This bottleneck calculator adds resolution, memory, workload and quality preset to that comparison, because the same pair of parts can be limited by different components at 1080p and at 4K.

It is a planning instrument rather than a measurement. It works from published benchmark data for the parts you select, not from readings taken on your own machine, which is exactly what makes it useful before you have bought anything. You can check a pairing you are considering, compare two upgrade paths against each other, or find out whether the part you were about to replace was ever the constraint.

People use it for gaming, video editing, 3D rendering, streaming and everyday multitasking. In demanding titles such as Cyberpunk 2077, Red Dead Redemption 2, Microsoft Flight Simulator or Escape from Tarkov, an unbalanced pairing shows up as a frame rate well below what the hardware on paper suggests, uneven frame pacing, or one component sitting idle while the other runs flat out. If you want the deeper reading on where the line sits, our guide to what counts as a good bottleneck percentage goes through it properly.

What this calculator tells you

Every PC has one component that runs out of headroom before the rest. In a gaming PC it is almost always the processor or the graphics card, and whichever one it is quietly sets your frame rate while the other waits around. This calculator compares the two, tells you which is the limiting part, and puts a number on how wide the gap is.

The reason that matters is money. The most common upgrade mistake is replacing the part that was never the constraint — buying a faster graphics card when the processor was the thing capping performance, then discovering the new card behaves almost exactly like the old one. Checking first turns an expensive guess into a decision you can justify.

It is also worth saying plainly what this is not. It is an estimate built from published benchmark data, not a measurement taken on your machine. It cannot know how a particular game distributes its work, how fast your memory runs, or what your cooling allows. Read the percentage as a guide to the size and direction of a mismatch, not as a promise about a specific title.

How the calculation works

Every result starts by comparing two numbers that were never designed to sit on the same scale. A processor’s CPU benchmark score and a graphics card’s 3D benchmark score come from different suites measuring different workloads, so the first step has nothing to do with bottlenecks at all. It is arithmetic to make the two comparable.

Step one: adjust both parts for the workload

Before anything is compared, each part’s benchmark score is scaled by how hard your selected workload leans on it. A competitive shooter at low settings asks far more of the processor and far less of the card than a path-traced single-player game at maximum quality, so the same two parts are effectively different hardware depending on what you run. The use case, game genre and quality preset all feed this step, which is why changing them moves the result.

Step two: put both parts on one scale

The adjusted CPU figure is divided by 40,000 and the adjusted graphics figure by 38,000. Those divisors are fixed reference points rather than moving averages, which is why a given part’s normalised score does not drift over time as newer hardware is added above it.

Step three: weight for resolution

Rendering more pixels is overwhelmingly graphics work, while a game’s logic, physics and draw calls stay roughly constant however many pixels end up on screen. At a low resolution the processor has to prepare far more frames per second, which is exactly where a weak CPU shows itself; at a high resolution the card is doing so much more work per frame that the processor stops being the constraint. The calculator models that by multiplying the normalised CPU score by 0.7 at 1080p, 1.0 at 1440p and 1.2 at 4K.

Step four: compare, then rate

Whichever adjusted score is lower is named as the limiting part, because it runs out of headroom first and forces the other component to wait. The percentage is the proportional gap between the two scores, measured against the higher one. The saturated component is reported at 95% utilisation, standing in for running as hard as it can, and the other is shown as a proportion of that. A balance score out of 100 is derived from the same figure, so 3% bottleneck reads as 94 out of 100.

The four rating bands

The bands exist to answer whether a result is worth acting on, not just whether there is a gap at all.

One caveat on the top band: because the thresholds are fixed, a pairing that behaves perfectly well in practice can still land above 30% at 4K, where the resolution weighting pushes the processor’s adjusted score up. Read the named limiting component alongside the number rather than treating the band label as a verdict on the build.

Reading your own utilisation numbers

Once you have a monitoring overlay running, the pair of numbers that matters is CPU load against GPU load. There are only six patterns worth recognising, and each one points at a different conclusion. This is the quickest diagnostic in the whole article, because it needs no benchmarks and no calculator — just two figures read off the screen during normal play.

Two cautions. First, total CPU usage hides single-thread limits: a game pinning two cores of a sixteen-core processor shows about 15% overall while being thoroughly CPU-limited, so always switch the graph to logical processors. Second, high GPU usage is not a problem to be solved. A card at 99% is a card you are getting your money’s worth from.

Three worked examples, with the real numbers

These are not illustrations. Each screenshot below is the actual output of the calculator at the top of this page, taken at the stated resolution with the gaming preset. You can reproduce every one of them in about thirty seconds.

Example 1: a budget processor with a flagship card

An Intel Core i3-12100F is a four-core, eight-thread Alder Lake part scoring 13,953 in our processor index. A GeForce RTX 4090 scores 38,042 on the graphics side. Pairing them is the textbook mistake, and the calculator says so without hedging.

Read the GPU load column. At 1080p that card is doing about a quarter of the work it is capable of, because the processor cannot prepare frames fast enough to keep it busy. Note also that the figure falls as resolution rises: at 4K the card finally has enough to do that the gap narrows. That is the right direction, and it is why buying a 4090 to pair with an i3 makes least sense at exactly the resolution most people buy it for.

Example 2: a flagship processor with an entry-level card

The reverse case. An AMD Ryzen 7 9800X3D scores 39,919; a GeForce GTX 1650 scores 7,867.

Here the figure gets worse as resolution climbs, which is the opposite of example one and exactly what should happen. Every extra pixel lands on the card, so a weak GPU falls further behind the higher you push it. The 3D V-Cache on that processor is doing nothing for you at all in this build.

Example 3: a genuinely balanced build

The same Ryzen 7 9800X3D with the RTX 4090 — 39,919 against 38,042, which is about as close as two current flagships get.

At 1440p the two parts are matched almost perfectly. At 4K the card becomes the limit, which is the state you want in a gaming PC, because resolution and quality settings then give you something to trade. The 1080p figure is the weakest of the three: 28% overstates the problem, since a 9800X3D at 1080p is one of the fastest gaming processors made. It is a fair illustration of why the limiting component matters more than the number.

Worth noting what changed between examples two and three: the processor stayed the same and only the card moved, and the result went from 78% to 3% at 1440p. That is the whole argument for checking a pairing before buying either half of it.

What is a PC bottleneck?

A PC bottleneck is the point where one component reaches its working limit before the others do, so the rest of the system ends up waiting on it. It is not a fault and not a sign of bad parts. Every computer has a limiting component under every workload; the only question worth asking is whether the gap is wide enough to be worth money.

Consider a PC with a strong graphics card and an older processor. In a competitive shooter at 1080p the card may be capable of 240 frames per second while the processor can only prepare 140. The card waits, and the system sits near 140. That is a CPU bottleneck. Move the same machine to 4K with ray tracing on, and the card is suddenly doing four times the pixel work while the processor’s job barely changes — now the card runs out first and it is a GPU bottleneck. One pairing, two different limiting components, decided entirely by the settings.

That is why no pairing has a single fixed bottleneck percentage, and why any tool quoting one number without asking your resolution is skipping the variable that decides the answer.

Upscaling and frame generation complicate it further. DLSS, FSR and XeSS render at a lower internal resolution and scale up, which reduces the graphics workload and can lift a GPU-limited result. Frame generation raises the number on screen without changing how fast the processor can prepare the underlying frames, so a higher displayed figure does not mean a CPU limit has gone away.

Signs your PC is being held back

• One component sits near full load while the other stays well below it.
• Frame rate barely moves after an upgrade that should have been a large step.
• The average frame rate looks fine but the game still feels uneven to play.
• Frame rate collapses in crowded cities, large battles or simulation-heavy scenes.
• Dropping the resolution produces little or no improvement.
• Raising a single graphics setting costs far more frames than it should.
• Textures load in late, stay blurry, or reload as you turn around.
• Performance falls apart as soon as a browser or recording software is open.
• Clock speeds drop after a few minutes of play and recover when you stop.
• The gap between your average frame rate and your 1% lows is unusually wide.

What causes a bottleneck

• A large performance gap between the processor and the graphics card.
• A high refresh-rate target, which multiplies the work the processor has to do.
• High resolutions and ray tracing, which multiply the work the card has to do.
• Not enough VRAM to hold the textures a game wants resident.
• Too little system memory, pushing the machine into paging to storage.
• Single-channel or slow memory, halving the available bandwidth.
• Weak cooling, which makes a component reduce its own clocks to stay safe.
• Laptop power limits, which can make identically named parts perform very differently.
• Background software taking CPU time the game needed.
• A slow or nearly full drive delaying asset streaming.
• Poor game optimisation, which no hardware choice fully fixes.
• Wrong settings: XMP or EXPO left off, a stale driver, or a power plan capping clocks.

The seven types of bottleneck

A processor-versus-graphics-card mismatch is the bottleneck most people mean, and it is the one the calculator above scores. It is not the only way a PC ends up slower than its parts suggest. Six other limits sit outside that comparison entirely, and none of them will move the percentage — so it is worth knowing how each one announces itself.

1. CPU bottleneck

The processor cannot prepare game instructions as fast as the card can render them. It handles game logic, physics, character behaviour, draw calls and everything running in the background, and when those take too long the card receives fewer frames to work on. A CPU bottleneck does not require total CPU usage to hit 100%: many games lean hard on one or two threads, so individual cores can be pinned while the overall figure looks relaxed.

Signs: GPU usage sits below 90% while one or more cores stay pinned; lowering the resolution changes almost nothing; frame rate drops in crowded areas and large battles; average frame rate looks acceptable but 1% lows are poor; a faster card produced almost no gain.

2. GPU bottleneck

The card takes longer than the processor to finish each frame. It handles textures, lighting, shadows, shaders, reflections and ray tracing, and when that workload exceeds what the card can process — or what its memory can hold — the processor waits. This is the normal state for a gaming PC and the one you generally want, because it means you are getting everything the processor has to give and you still have settings to trade.

Signs: GPU usage pinned at 95–100% with the processor relaxed; a large frame-rate drop when you raise resolution or quality; an immediate gain when you lower either; ray tracing causing unstable frame times; VRAM at its limit with textures loading late.

3. RAM bottleneck

Memory rarely limits a system through raw capacity alone. The more common causes are a single-channel configuration running at half the available bandwidth, or an XMP or EXPO profile left disabled in the BIOS so the kit runs at the motherboard default rather than its rated speed. The symptom is not a lower average frame rate but a widening gap between that average and your 1% lows, usually when a game streams in a new area.

Signs: memory usage close to full during ordinary play; stutter on entering new areas; slow switching between applications; disk activity rising because the page file is in use; brief freezes with normal temperatures; an immediate improvement from closing background programs.

4. Storage bottleneck

A drive limits loading time, asset streaming and general responsiveness — not frame rate in a scene that is already loaded. Moving a game from a hard drive to an SSD shortens load screens and reduces texture pop-in as you move through a level. It does not raise average frames per second, because the drive is no longer in the path once assets are in memory. Replacing a SATA SSD with an NVMe one changes less than people expect for gaming specifically.

Signs: long loading screens; textures and objects appearing late as you move; brief freezes when a game streams in new terrain; disk usage at 100% with low transfer speed; large file copies slowing sharply partway through.

5. Power supply bottleneck

An undersized power supply does not gradually reduce performance. It holds up fine until a sudden spike in graphics card draw trips the protection circuit, and the machine restarts or the screen goes black under load while lighter settings stay stable. Modern cards draw brief transient peaks well above their rated figure, which is why a unit sized only to the sum of the sticker ratings can still fail. A higher-wattage unit does not force extra power into the hardware; each part draws what it needs.

Signs: shutdowns or restarts under load but not at idle; the screen going black when GPU load spikes; crashes that stop when you lower the card’s power limit; a new card causing instability the old one never did; the unit needing its rear switch cycled before it will start again.

6. Thermal bottleneck

Thermal throttling is the one limit that will not reproduce in a short test. The system starts a session at full boost clocks, then drops them once a component reaches its temperature or power limit, so frame rate falls after five or ten minutes of play and recovers once things cool. If performance is fine at first and degrades predictably, the problem is cooling rather than the pairing. High temperature alone is not the issue; the issue is heat forcing clocks down.

Signs: clock speeds falling as temperatures climb; frame rate dropping after several minutes and recovering after a break; frame-time spikes late in a session; better performance with the side panel off or in a cooler room; benchmark scores declining across repeated runs.

7. Network bottleneck

A network limit does not reduce your local frame rate, which is why it gets misdiagnosed so often. It affects online play, cloud gaming, voice chat and downloads. What matters is not just bandwidth but latency, jitter and packet loss: a 100 Mbps connection over weak Wi-Fi from another room can feel far worse than a slower wired one.

Signs: high ping while frame rate and component loads look fine; players appearing to teleport or snap backwards; shots registering late; voice chat breaking up; everything degrading the moment another device starts a download.

Compatibility limits are a separate question again: a processor and graphics card can be perfectly balanced and still not work together in the case and board you own. The PC builder checks socket, memory generation and power draw before you buy, and AM4 versus AM5 covers the upgrade-path question for AMD builds specifically.

Resolution decides which part is the limit

Bottlenecks are not fixed properties of a pair of parts. The resolution you play at changes the balance between them, which is why selecting it honestly matters more than any other input on this page.

Two practical consequences follow. First, a processor bottleneck you measured at 1080p may simply vanish if you move to a 1440p monitor, which occasionally makes a monitor the cheaper upgrade than a CPU. Second, if your result says you are GPU-limited at 4K, upscaling is the lever to reach for before a new card: rendering at a lower internal resolution and scaling up targets exactly the thing that is constraining you.

Select the resolution you actually play at. Checking a pairing at 1080p when you own a 4K monitor will name the wrong component.

Bottlenecks by what you actually use the PC for

Balance is not one target. A machine tuned for competitive shooters and a machine tuned for 4K video work want the money spent in different places, and the same pairing can be well matched for one and badly matched for the other.

This is why a result can look alarming and not matter. A 35% GPU-limited figure on a machine used for 4K single-player gaming is describing precisely the state you want. The same figure on a machine used for 1080p competitive play would be a genuine waste of a processor.

Which upgrade actually pays

Suppose the calculator names your processor as the limit at 1080p and you have a fixed budget. The order below is roughly how much performance each step returns per pound or dollar spent, and the first three are free.

1. Enable XMP or EXPO. Two minutes in BIOS. In a CPU-limited scenario this is frequently the largest single gain available, because memory bandwidth and latency feed the processor directly.
2. Close what you are not using. Browsers with many tabs, launchers, overlays and recording software all take processor time. Measure with and without before assuming hardware is at fault.
3. Check for thermal throttling. Watch clocks over a ten minute session. A throttling processor behaves exactly like a slower one, and cleaning a heatsink costs nothing.
4. Add a second memory stick. If you are running single-channel, this is the cheapest hardware fix there is and roughly doubles available bandwidth.
5. Move to 16GB, or 32GB if you stream or create. Cheap, and it removes the paging stutter that gets misread as a processor problem.
6. Upgrade the processor within your existing socket. Far cheaper than a platform change. Check what your board supports first — AM4 still has strong options, and AM4 versus AM5 covers whether to move at all.
7. Change platform. New processor, board and often memory together. The largest gain and by far the largest bill, so confirm the limit first.

If the result named the graphics card instead, the order inverts: enable upscaling, lower the one or two most expensive settings (ray tracing and shadows, usually), confirm the card is not thermally limited, then buy. And check your power supply can take the new card before you order it — our PSU calculator sizes for transient spikes rather than sticker ratings.

A final sanity check that costs nothing: run your current pairing and your intended pairing through the calculator above and compare the two results. If the number barely moves, you have just saved the price of the part.

Why this is the most accurate bottleneck calculator

Most bottleneck calculators show a percentage and nothing about how it was reached. That is the single biggest difference between this tool and the field: every number here traces back to a formula you can actually see, from the workload adjustment through the 40,000 and 38,000 normalisation divisors and the 0.7/1.0/1.2 resolution weighting to the band thresholds. A tool that will not show its working cannot be checked, and a result you cannot check is not more trustworthy for looking confident.

Several competing calculators treat resolution as a label rather than an input, offering it as a dropdown that barely changes the underlying number. That gets the physics wrong. Rendering pixels is graphics work and game logic is processor work, so the same CPU and GPU pairing can be limited by different parts at different resolutions. A calculator that does not let resolution meaningfully move the result is quietly averaging over the variable that should be driving the answer.

A number of these tools also publish precision they cannot back up, quoting error percentages or accuracy scores with no explanation of what was measured against what. We would rather tell you plainly that this is an estimate built from synthetic benchmark scores, that it cannot see your specific game engine, your memory speed or your cooling, and that the fixed band thresholds can read harshly at 4K.

The utilisation breakdown is another point most calculators skip entirely, returning a percentage and a one-line verdict with nothing showing how the gap plays out between the two parts. Seeing the saturated component alongside the other one’s estimated load is what makes a result actionable rather than just a badge.

Finally, the upgrade guidance here is tied to the actual result rather than generic advice. A calculator that tells everyone with a CPU bottleneck to buy a faster CPU is not adding anything past the verdict it already gave. This one distinguishes a marginal lead from a decisive one, gives separate guidance for 4K gaming and high refresh-rate 1080p or 1440p play, and says plainly when neither upgrade is worth the money, which happens more often than the upgrade-focused tone of most competing pages would suggest.

One last point of honesty, since it comes up: different bottleneck calculators will give you different percentages for the same two parts. They use different benchmark sources, different normalisation and different weighting, so they are not measuring quite the same thing. Ours tells you what it did and why. If three tools disagree on the number but agree on which component is the limit, that agreement is the useful part.

Conclusion

Every PC has a limiting component. Finding out which one it is costs nothing and takes about thirty seconds, and it is the difference between an upgrade that changes how your games feel and one that changes your bank balance and nothing else.

Use the calculator above to check the pairing you own or the pairing you are considering. Read the named limiting component first, the utilisation split second, and the percentage last. Then, if you already own the parts, confirm it with the Task Manager and resolution tests on your own machine — the two together will tell you more than any calculator can on its own.

And if the answer comes back as excellent balance, take it at face value. The most useful thing a tool like this can tell you is that you do not need to buy anything.