By Walter Jobs, Technology writer and editor · Published 8 October 2026
Short answer: being CPU bound is not a property of a game. It is a function of the frame rate you are chasing. At 1440p and 60 fps almost nothing is processor limited. At 1080p and 360 fps almost everything is, and competitive titles are where people chase those numbers.
Every list of “CPU intensive games” you will find puts Counter-Strike and Cities: Skylines in the same column. They do belong in the same column, but for opposite reasons, and the hardware that fixes one does not fix the other. That distinction is the whole article.
The frame budget is the entire explanation
Every frame your PC shows you is the end of a chain: the processor works out what should be on screen and issues the draw calls, then the graphics card draws it. Both have to finish inside the same window.
At 60 fps the budget is 16.7 milliseconds, which is comfortable. At 240 fps it is 4.2 ms. At 360 fps it is 2.8 ms.
Now the asymmetry. Graphics work spreads across thousands of shader cores and shrinks when you lower settings. The processor’s work is largely serial, runs mostly on one thread, and does not shrink when you lower settings. Dropping to Low takes the graphics card from 8 ms to 3 ms and takes the processor from 5 ms to 4.8 ms.
So the competitive player’s standard move, lowering every setting to chase frames, removes the graphics card from the equation and leaves them staring directly at their processor. That is why esports titles are the most CPU bound games there are, despite being the least demanding to look at.
Two completely different kinds of CPU bound
| Esports titles | Simulation and strategy | |
|---|---|---|
| Examples | CS2, Valorant, Overwatch 2, Apex Legends, League of Legends, Fortnite | Microsoft Flight Simulator, Cities: Skylines 2, Factorio, Stellaris, Crusader Kings 3, Total War |
| Why it is CPU bound | The frame budget is tiny because the target is 240 to 360 fps | The simulation itself is enormous, even at 60 fps |
| The limiting work | Draw call submission and the render thread | Game logic: thousands of agents, pathfinding, economy ticks |
| What helps | High clock speed and large cache | Large cache and fast memory |
| Does lowering settings help? | Barely. The processor is the limit already | No. Settings do not touch the simulation |
| Does a faster GPU help? | No | No |
| Realistic expectation | Hundreds of fps, limited by the processor | 60 fps or less in the late game, whatever you buy |
The practical consequence: a player struggling to hold 240 fps in CS2 and a player whose Cities: Skylines 2 city slows to a crawl have the same diagnosis and completely different expectations. The first can reach their target with a faster chip. The second cannot, because late game simulation load grows without limit.
Why the engine matters more than the genre
No article on this subject names the engines, which is strange, because the engine decides how the work is threaded.
- Source 2 (CS2, Deadlock) is built for very high frame rates and leans on single thread performance and cache.
- Unreal Engine 4 and 5 (Fortnite, Valorant’s foundations, countless others) have a dominant render thread. UE5 in particular is known for traversal stutter, which is a processor side problem that no graphics card fixes.
- Clausewitz (Stellaris, Crusader Kings) runs a simulation that is close to purely single threaded, which is why late game slowdown is so severe and so resistant to hardware.
- Factorio’s custom engine is the clearest case in gaming of a title limited by memory latency and cache size rather than clock speed.
- DirectX 11 versus DirectX 12 and Vulkan matters too. DX11 submits draw calls largely on one thread, so older titles hit a processor wall sooner than newer ones.
Why large cache changed this
Processors with stacked cache, such as AMD’s X3D parts, win gaming benchmarks by margins their clock speeds do not explain. The reason is that game simulation data is accessed in unpredictable patterns, and a cache miss costs the processor a trip to main memory worth hundreds of cycles.
A much larger cache means more of the working set stays close to the core. In games that are limited by this, which is most CPU bound games, it is worth more than extra cores or extra clock speed. In graphics limited games it is worth nothing at all, which is why the same chip can look transformative in one benchmark and irrelevant in the next.
What your processor can actually sustain
We ran several processors against an RTX 5090 at 1080p on the Low preset, which removes the graphics card from the equation almost entirely and leaves the processor ceiling exposed:
| Processor | Frame ceiling | Clears 60Hz | Clears 144Hz | Clears 240Hz |
|---|---|---|---|---|
| Ryzen 5 1600 (2017) | 83 fps | Yes | No | No |
| Core i5-12400F | 151 fps | Yes | Yes | No |
| Ryzen 5 7600 | 158 fps | Yes | Yes | No |
| Core i7-14700K | 182 fps | Yes | Yes | No |
| Ryzen 7 9800X3D | 198 fps | Yes | Yes | No |
An important caveat about these numbers. Our model is calibrated against modern AAA titles, not esports titles. A 9800X3D in CS2 will run far above 198 fps, because CS2 asks for a fraction of the simulation work a modern open world does. Read the table as a ranking and as a demonstration of the principle, not as a CS2 prediction. We would rather say that than quietly present a figure that does not apply.
Run your own pair through the calculator for a figure based on your actual parts.
Games people most often ask about
| Game | Mostly limited by | Why |
|---|---|---|
| Counter-Strike 2 | CPU | Very high frame targets, low graphics load |
| Valorant | CPU | Deliberately light graphically so it runs on anything; the ceiling is the processor |
| Fortnite (Performance mode) | CPU | Performance mode strips graphics work to chase frames |
| Minecraft (Java) | CPU | Chunk generation and a largely single threaded engine |
| Microsoft Flight Simulator | CPU | Enormous simulation and streaming workload |
| Cities: Skylines 2 | CPU | Agent simulation grows with city size, without limit |
| Cyberpunk 2077 with ray tracing | GPU | Heavy graphics load dominates at any sensible resolution |
| Most AAA at 1440p or 4K | GPU | Pixel load outweighs simulation load |
Notice that the right hand column is doing the work. It is not the genre, it is the ratio between simulation load and pixel load.
Why cache beats cores, with the mechanism
Processors with large stacked cache win gaming benchmarks by margins their clock speeds and core counts do not explain. The reason is specific and worth understanding, because it tells you what to buy.
Game simulation data is accessed in unpredictable patterns. The processor asks for a piece of data, and if it is not in cache it has to fetch it from main memory, which takes hundreds of cycles during which that core does nothing useful. In a workload that must finish inside a 4 millisecond budget, a stall like that is enormous.
A much larger cache means more of the working set stays close to the core, so those stalls happen less often. In games that are limited by this, which is most processor limited games, it is worth more than extra cores or extra clock speed.
| Workload | What decides performance | So buy |
|---|---|---|
| Competitive shooters at high refresh | Cache size and single thread speed | A large cache part |
| Simulation and strategy | Cache size and memory latency | A large cache part, and fast memory |
| AAA at 1440p or 4K | The graphics card, mostly | Spend on the card instead |
| Rendering, compiling, encoding | Core count and multithread throughput | A high core count part |
| Gaming while streaming on the CPU | Core count, and then cache | A high core count part |
That table is why one chip can look transformative in one benchmark and irrelevant in the next, and why you should never take a single benchmark average as a verdict on a processor.
Each processor page in our CPU database shows both a gaming index and a multithread index for exactly this reason: they are different rankings and conflating them is how people buy the wrong chip.
A frame is not one job: the render thread problem
Describing the processor’s work as one task hides why more cores so rarely help.
- The game thread runs the simulation: physics, AI, input, game logic.
- The render thread turns the result into draw calls for the graphics card.
- Worker threads handle asset streaming, audio, decompression and so on.
The game thread and the render thread are largely serial and largely cannot be split. Worker threads parallelise well but are not usually what runs out of time. So a game can saturate two cores while six idle, and adding a ninth and tenth core changes nothing at all.
This is also why total processor usage in Task Manager is such a poor diagnostic. Two cores pinned on an eight core chip reads as roughly 32% while the game stutters.
Which engine your game runs on, and what it means
| Engine | Games | Processor behaviour |
|---|---|---|
| Source 2 | CS2, Deadlock | Built for very high frame rates. Leans on single thread speed and cache |
| Unreal Engine 4 | Fortnite, Valorant’s foundations, many others | Dominant render thread. Processor limited at high frame targets |
| Unreal Engine 5 | Recent AAA releases | Known for traversal stutter, a processor side problem no card fixes |
| Clausewitz | Stellaris, Crusader Kings 3 | Simulation close to single threaded. Late game slowdown is structural |
| Factorio custom engine | Factorio | The clearest case in gaming of a memory latency and cache limit |
| RE Engine | Resident Evil, Monster Hunter | Generally well threaded, less processor limited than peers |
Engine matters more than genre. Two shooters on different engines can behave completely differently on identical hardware, which is why game specific advice beats general advice every time.
The refresh rate you are targeting decides everything
The single most useful reframing on this topic: stop asking whether a game is CPU bound and start asking what frame rate you are chasing.
| Your target | Typical processor demand | Who is limited |
|---|---|---|
| 1440p at 60 | Low | The graphics card, almost always |
| 1440p at 144 | Moderate | Usually the card, sometimes the processor |
| 1080p at 144 | Moderate to high | Often the processor |
| 1080p at 240 | High | The processor, in most titles |
| 1080p at 360 | Very high | The processor, in essentially every title |
The same PC appears perfectly balanced in the first row and badly processor limited in the last. Nothing about the hardware changed. Only the target did.
This is why our calculator asks for your resolution and preset before it reports anything, and why any bottleneck figure quoted without them is not a number you can act on.
If you are CPU bound in a competitive title
- Confirm it first. Drop to 720p. If your frame rate barely moves, the processor is the limit. If it jumps, it was the card.
- Enable XMP or EXPO. Memory running at its default speed instead of its rated speed is extremely common and costs real frames in exactly these titles.
- Check per core usage, not total. One core pinned while the average reads 35% is the normal picture here.
- Stop lowering graphics settings. You have already removed the graphics card from the equation; going lower costs image quality and buys nothing.
- Turn off overlays and background applications. At a 3 ms frame budget, the overhead of a browser or a capture overlay is a measurable share of your budget.
- Consider cache rather than cores. If you do upgrade, a large cache part beats a higher core count one for this workload, every time.
One thing that makes it worse, not better
Upscaling. DLSS, FSR and XeSS reduce the graphics card’s work by rendering at a lower internal resolution. If you are already processor limited, that does nothing for your frame rate, because the card was never the problem. People turn on upscaling, see no improvement, and conclude the feature is broken. It is working perfectly; it is solving a problem they do not have.
The frame rate ceilings quoted for each title are modelled from published processor benchmark data and from the frame time budgets above, not captured in a match. The model behind them is documented on how we calculate this.
Frequently asked questions
Are esports games CPU or GPU intensive?
CPU intensive, because of the frame rates people target rather than the visuals. Competitive titles are deliberately light graphically, and players lower settings further to chase frames, which leaves the processor as the only limit.
Is Valorant CPU or GPU intensive?
CPU intensive at the frame rates competitive players want. It is designed to run on modest hardware, so the graphics card is rarely the limit, and above roughly 200 fps your processor decides your frame rate.
What games are the most CPU bottlenecked?
Two groups, for opposite reasons. Competitive shooters and MOBAs, because the frame budget at 240 fps and above is tiny. And simulation and strategy games such as Flight Simulator, Cities: Skylines 2, Factorio and Stellaris, because the simulation itself is enormous even at 60 fps.
Why does lowering settings not help my frame rate?
Because settings control graphics work and your processor is the limit. Lowering them shortens the card’s half of the frame budget, which was never the part running out of time.
Will a better GPU help in CS2 or Valorant?
At high frame rates, no. If you are already processor limited, a faster card renders the frames it is given more quickly and then waits longer. The frame rate does not change.
Does more cores help in CPU bound games?
Rarely, past six or eight. Most game engines cannot use many cores effectively. Single thread performance and cache size matter far more, which is why processors with large stacked cache win gaming benchmarks against chips with twice the cores.
Will DLSS or frame generation raise my frame rate in CS2 or Valorant?
Almost certainly not, and competitive titles are the clearest case. These games already run at hundreds of frames per second on modest cards, which means the graphics card is nowhere near the limit, so removing graphics work with upscaling removes work that was not costing you anything. Frame generation is worse than useless here: it adds processor side cost and inserts latency, which is the one thing a competitive player should not trade away for a higher number on the counter.
Is a CPU bottleneck in esports games bad?
It means you have reached what your processor can deliver. Whether that matters depends on your monitor: if you have a 144Hz panel and your processor sustains 180 fps, the limit costs you nothing. On a 360Hz panel the same limit is the thing holding you back.
Why does a CPU with large cache beat one with more cores in games?
Because game data is accessed unpredictably, and a cache miss leaves the core waiting hundreds of cycles for main memory. A larger cache means fewer of those stalls. In a workload with a four millisecond frame budget that matters far more than extra cores the engine cannot use.
How many CPU cores do I need for gaming?
Six to eight is the sweet spot for almost everyone. Most engines have a game thread and a render thread that are largely serial, so cores beyond that sit idle during gameplay. Spend the difference on cache, memory speed or the graphics card.
Does the game engine affect whether I am CPU bound?
Substantially, and more than the genre does. Source 2 is built for very high frame rates, Unreal Engine has a dominant render thread, and Clausewitz runs a near single threaded simulation. Two shooters on different engines can behave completely differently on identical hardware.
Check your own frame ceiling
Browse every processor we track to compare gaming index against multithread index, which are different rankings.
Enter your processor and graphics card to see which part sets your frame rate, at the resolution you actually play at.