By Walter Jobs, Technology writer and editor · Published 8 October 2026
Short answer: the bottleneck does not disappear, it moves. Your processor prepares roughly the same number of frames per second whatever resolution you play at. Your graphics card prepares far fewer as the pixel count rises. Raise the resolution and the card becomes the slower of the two, so the processor stops being the limit. Nothing about the processor improved.
That distinction matters, because the most common advice on this topic is to raise your resolution to fix a CPU bottleneck. It does not fix anything. It hides the symptom by giving your graphics card more work, and you pay for it in frame rate.
The one chart that explains it
We ran a Ryzen 5 5600 paired with an RTX 4070 through four resolutions in our calculator and plotted what each part can sustain on its own. The dotted line is the processor. The blue line is the card.
The processor holds at 126 fps across every single resolution. Not approximately: exactly. Simulating the game world, running the AI, processing input and issuing draw calls does not get harder because you added pixels.
The card falls from 143 fps to 44 fps over the same range, a drop of 58%, because filling 8.3 million pixels at 4K is four times the work of filling 2.1 million at 1080p.
Where the two lines cross is where your build is balanced. Left of it you are processor limited. Right of it you are card limited. Same two components the whole way along.
The same pair, resolution by resolution
| Resolution | CPU ceiling | GPU ceiling | You get | Limited by |
|---|---|---|---|---|
| 720p | 126 fps | 143 fps | 119 fps | CPU, by 11.9% |
| 1080p | 126 fps | 104 fps | 99 fps | Balanced |
| 1440p | 126 fps | 77 fps | 74 fps | Balanced |
| 4K | 126 fps | 44 fps | 43 fps | GPU, by 30.2% |
Read the middle two columns rather than the last one. The processor column never changes. Every single thing that happens on this table is the graphics column falling.
This is why the answer flips with different hardware
A Core i5-12400F with an RTX 4060 is the opposite story. The card is the weaker part everywhere, so there is no resolution at which the processor becomes the limit:
| Resolution | CPU ceiling | GPU ceiling | Limited by |
|---|---|---|---|
| 720p | 137 fps | 86 fps | Balanced |
| 1080p | 137 fps | 63 fps | GPU, by 8.3% |
| 1440p | 137 fps | 46 fps | GPU, by 32.5% |
| 4K | 137 fps | 18 fps | GPU, by 74.3% |
And a Ryzen 7 7800X3D with an RTX 5090 is processor limited almost everywhere, because the card is so far ahead that only 4K gives it enough work to fall behind:
| Resolution | CPU ceiling | GPU ceiling | Limited by |
|---|---|---|---|
| 720p | 167 fps | 351 fps | CPU, by 52.3% |
| 1080p | 167 fps | 256 fps | CPU, by 34.7% |
| 1440p | 167 fps | 189 fps | CPU, by 11.4% |
| 4K | 167 fps | 108 fps | Balanced |
Notice what that last table means in practice. One of the fastest gaming processors you can buy is the limiting part at 1440p with a 5090. A CPU bottleneck is not a sign that your processor is bad. It is a sign that your card is fast relative to it, which is usually a position people paid to be in.
Run your own pair through the calculator and you will get this table for your exact parts.
So should you raise the resolution to fix it?
No, and this is where most articles on this topic give advice that sounds reasonable and leaves people worse off.
Raising the resolution does not give you more frames. Look at the first table again: going from 1080p to 4K took that build from 99 fps down to 43 fps. The bottleneck label changed from balanced to GPU limited, and the actual experience got substantially worse.
What raising the resolution does is make your card the limiting part, which means you are now using all of the card you paid for. That is a reasonable thing to want if you have a fast card and a high resolution screen. It is not a fix for anything.
| If your goal is | Then | Not |
|---|---|---|
| More frames per second | Lower the resolution or the preset, or upgrade the limiting part | Raise the resolution |
| A better looking game at the same frame rate | Raise the resolution or preset, if you are CPU limited | Buy a faster processor |
| Using all of the card you bought | Play at the resolution the card is built for | Play at 1080p on a 5090 |
Settings move the card far more than the processor
The same asymmetry applies to the graphics preset, which is why dropping settings often helps less than people expect when the processor is the limit. Same pair, same resolution, four presets:
| Preset at 1440p | CPU ceiling | GPU ceiling | You get |
|---|---|---|---|
| Low | 139 fps | 131 fps | 123 fps |
| Medium | 132 fps | 100 fps | 96 fps |
| High | 126 fps | 77 fps | 74 fps |
| Ultra | 122 fps | 62 fps | 60 fps |
From Low to Ultra the processor ceiling drops 12%. The graphics ceiling drops 53%. If you are already processor limited, turning settings down gains you very little, because the part that is holding you back barely cares about settings.
The thing nobody mentions: upscaling moves you back
DLSS, FSR and XeSS render the frame at a lower internal resolution and scale it up. DLSS Performance at 4K renders at roughly 1080p.
That means upscaling gives your graphics card an easier job while leaving your processor’s job untouched. In the terms of the chart above, it slides you back to the left. If you were comfortably card limited at native 4K, turning on aggressive upscaling can put you back into a processor limit, and the frame rate will not rise as much as the resolution drop suggests it should.
Frame generation is a separate case again: it adds work on the processor side to produce each generated frame, so a system that is already processor limited sees less benefit than one that is not.
Upscaling and frame generation change the answer again
Two features that did not exist when most advice on this topic was written now move the limit on their own, and almost nothing ranking for this question mentions either.
Upscaling slides you back toward the processor
DLSS, FSR and XeSS render each frame at a lower internal resolution and scale it up. DLSS Performance at 4K renders at roughly 1080p internally, which means your card is doing 1080p work while your monitor shows a 4K image.
Your processor’s job does not change at all. So upscaling gives the graphics card an easier task and leaves the processor exactly where it was, which on the chart above slides you back to the left. If you were comfortably card limited at native 4K, aggressive upscaling can put you back into a processor limit and the frame rate will not climb as far as the resolution drop suggests.
| Setting at 4K | Internal render | What the GPU does | What the CPU does |
|---|---|---|---|
| Native | 3840 x 2160 | Full 4K work | Unchanged |
| DLSS Quality | 2560 x 1440 | 1440p work | Unchanged |
| DLSS Balanced | 2227 x 1253 | Close to 1253p work | Unchanged |
| DLSS Performance | 1920 x 1080 | 1080p work | Unchanged |
Read the last column. It is the same every row, which is the whole point.
Frame generation adds processor cost
Frame generation produces intermediate frames from the ones your PC rendered. It raises the number on your counter without reducing the work of producing the real frames, and it adds a small amount of processor side cost to schedule and present each generated frame.
A system that is already processor limited therefore benefits least from it, which is the opposite of what people expect. If your frame rate barely moves when you enable it, that is a strong signal your limit was never the card.
Why DirectX 12 did not fix this, and sometimes made it worse
DirectX 11 submits most draw calls on a single thread. That is the historical reason processors hit a wall in older titles: one core doing all the submission work while the rest idle.
DirectX 12 and Vulkan let a game spread submission across several threads, which should have removed the problem. In practice it moved responsibility from the driver to the game developer, and the results vary enormously by studio. Some DX12 titles scale beautifully across cores. Others are worse than their DX11 path because the engine never got the threading work it needed.
This is why two games that look equally demanding can behave completely differently on the same hardware, and why a benchmark average across several titles tells you less than it appears to.
- If a game offers both APIs, try each. The faster one is genuinely game specific.
- Unreal Engine 5 traversal stutter is a processor side problem. It shows as hitches when you move into new areas, and no graphics card fixes it.
- Shader compilation stutter is also processor work, usually in the first minutes of a session. Letting a game finish compiling before you play is the fix.
Memory is often the real cause of a CPU limit
A processor limit is frequently a memory limit wearing a disguise. Game data is accessed in unpredictable patterns, and every trip to main memory that misses cache costs hundreds of cycles the processor spends waiting rather than working.
Two things follow, and both are free or cheap:
| Situation | Effect | Fix |
|---|---|---|
| Memory running at JEDEC default instead of its rated speed | Several percent of frame rate, more in CPU limited scenes | Enable XMP or EXPO in the BIOS |
| A single stick instead of two | Roughly half the memory bandwidth available | Add a second matching stick |
| 8 GB total | The processor ceiling drops by around 18% in modern titles | 16 GB in dual channel |
| Mismatched kits | Often falls back to the slower timings | Buy a matched kit rather than mixing |
Running your memory at its default speed is extremely common, costs nothing to fix, and shows up precisely in the scenes where you are processor limited. Check it before you consider a new chip.
Averages hide what a CPU limit actually feels like
Every table on this page, and almost every benchmark you will read, reports average frame rate. That is the metric that makes a processor limit look harmless at 4K.
The part you feel is frame time consistency, usually reported as 1% lows. A processor limit tends to produce uneven frame delivery rather than a uniformly lower number, which is why a game can read 80 fps on the counter and still feel like it is hitching. Raising the resolution raises the average back up and does not necessarily fix the hitching, because the moments that caused it were the processor struggling with a specific scene, not the card struggling with pixels.
If a game feels rough but the counter looks fine, turn on a frame time graph in your overlay rather than watching the average. Spikes on an otherwise flat line point at the processor.
The frame rates and percentages used across the three resolutions above come from our own calculator, which models them from published benchmark data for each part rather than measuring them on a test bench. The resolution scaling it applies, and how far to trust it, are set out on how we calculate this.
Frequently asked questions
Does a CPU bottleneck go away at 1440p?
Not literally. Your processor pushes the same number of frames at 1440p as it did at 1080p. What changes is that your graphics card now pushes fewer, so the card becomes the slower of the two and the processor stops being the limit. The processor did not get faster and your frame rate did not go up.
Is 1440p more CPU or GPU intensive than 1080p?
More GPU intensive, and identically CPU intensive. 1440p has 78% more pixels than 1080p, all of which is graphics work. The processor’s job, simulating the world and issuing draw calls, is unchanged.
Does 4K eliminate CPU bottlenecks?
It hides them in most games, because at 4K the card is usually so loaded that it limits everything. It does not eliminate them. CPU heavy games such as large strategy titles and simulators can still be processor limited at 4K, and 1% lows can stay poor even when the average looks healthy.
Why is my GPU usage low at 1080p?
Because your processor is the limit. The card finishes each frame and waits for the next one to be prepared, so it never reaches full utilisation. GPU usage sitting well under 100% while the processor is pinned is the clearest sign of a CPU bottleneck.
Should I upgrade my CPU or GPU?
It depends entirely on the resolution you actually play at, which is why a single answer is useless. Run your pair through the calculator at your resolution: if it reports a CPU limit, the processor is the upgrade; if it reports a GPU limit, the card is.
Does a higher resolution monitor need a better CPU?
Usually the opposite. Moving to a higher resolution reduces the demand on your processor relative to your card, so the same chip that struggled at 1080p high refresh is often comfortable at 4K 60.
Does DLSS fix a CPU bottleneck when I move up to 1440p or 4K?
No, and at higher resolutions the reason is easy to see. Upscaling works by rendering at a lower internal resolution, so a 1440p output rendered from 1080p asks the card for roughly half the pixels. That is graphics work removed. Your processor is still issuing the same draw calls and running the same simulation at the same rate, so if it was the limit before upscaling it is the limit after, at the same frame rate. The one thing that does change is that moving genuinely to 1440p or 4K without upscaling raises graphics work, which can hide a processor limit by making the card the slower part again.
Is a CPU bottleneck worse in DirectX 12 games?
It depends entirely on the game. DirectX 12 allows draw call submission across multiple threads, which should help, but it moves that work from the driver to the developer and the quality of the implementation varies enormously. Some DX12 titles are far better than their DX11 path, and some are worse.
Will faster RAM fix a CPU bottleneck?
It can help noticeably, and enabling XMP or EXPO costs nothing. A processor limit is often really a memory latency limit, because game data is accessed unpredictably and a cache miss leaves the processor waiting. Running memory at its default speed instead of its rated speed is very common and shows up in exactly these scenes.
Check your own pair
Browse every processor we track, or every graphics card, to see where your parts sit and what each one pairs with.
Enter your processor and graphics card and you will get this table for your exact hardware, at every resolution, with the frame rate to expect and the one upgrade worth paying for.