For years now, we've invested a good chunk of time and energy trying to educate gamers on how and why CPU testing is done the way it is. We know those of you who already understand are probably sick and tired of hearing about it, and believe us, we get it. We've covered this topic from a few different angles before, yet despite all that, very little progress seems to have been made.
Now, for those of you thinking we just need to get over this because it's only a vocal minority that doesn't understand the basics of CPU testing, and that most of you do get it, we're here to disappoint you with a somewhat shocking reality.
You see, whenever we post CPU benchmarking content, which we've been doing quite a lot lately, the comments section here and on YouTube ends up flooded with people demanding 1440p and even 4K testing. Some are simply curious why those resolutions weren't included, while others claim the content is misleading or dumb because it's not "real world."
But it's just the comments section, right? After all, less than 1% of people who read or watch actually comment, making them, by definition, a vocal minority. Well, no. As much as we'd like to believe that's the case, a recent poll we ran on the Hardware Unboxed channel suggests otherwise:
The poll was simple: do you want to see 1080p testing with medium and ultra presets, as we're currently doing, or would you rather see two resolutions, 1080p and 1440p, with a single preset? Shockingly, the majority of people, 56%, voted for the inclusion of 1440p data, which is a tad disappointing for a few reasons.
First, the 1440p data is useless. It tells you nothing that can't already be learned from the 1080p data. This is simply a fact. There's nothing useful about 1440p CPU testing. It's not "real world." It doesn't tell you how your system might perform, it provides no additional insight, and it only serves to potentially mislead you.
The 1440p data is useless...
The poll results were also disappointing because the data we get from testing two presets per game is often insightful. Testing games like Cyberpunk and Spider-Man 2 with and without ray tracing has produced some really interesting results, especially when it comes to CPU performance.
The push for higher-resolution CPU testing is also a bit baffling at a time when most gamers are now using some form of upscaling. Love it or hate it, that's simply the reality. To prove this, we polled the same audience, asking whether they use upscaling at 1440p, and 64% of you said you do.
The reason this matters is because if you're gaming at 1440p with DLSS or FSR enabled, and you're using the highest quality upscaling mode, "Quality," the base render resolution is actually below 1080p. You end up at 960p. The base resolution falls to 835p with "Balanced" mode and then just 720p using "Performance."
So if you're gaming at 1440p with some form of upscaling enabled, while also believing CPU benchmarking should be done at 1440p rather than 1080p because 1080p isn't "real world," then you're actually gaming at a resolution below 1080p. Moreover, modern upscaling techniques work so well that "Performance" mode at 4K is now considered acceptable, and that has a base render resolution of 1080p.
Anyway, we can already hear many of you asking: why not just test at 1440p with DLSS Quality enabled? So that's exactly what we're going to do, comparing that data against native 1080p and 1440p.
Of course, as usual for all of this testing, we're using an RTX 5090. That's a separate issue we'll tackle toward the end of this benchmarking explainer, but for now we want to focus on this fixation gamers have with resolution when it comes to CPU testing. For this comparison, we're using the Ryzen 7 9800X3D, 5800X3D, and 3800X, so let's get into the data.
Test System Specs
| CPU Motherboard Memory |
AMD Ryzen 9800X3D | MSI MPG X870E Carbon WiFi [BIOS 7E49v1A7E] G.Skill Trident Z5 RGB 32GB DDR5-6000 CL30-38-38-96 |
| AMD Ryzen 5800X3D AMD Ryzen 3800X |
MSI MPG X570S Carbon MAX WiFi [BIOS 7D52v1D1] G.Skill Trident Z Neo RGB 32GB DDR4-3600 CL14-15-15-35 |
|
| Graphics Card | Gigabyte AORUS GeForce RTX 5090 Stealth Ice 32G | |
| Power Supply | Kolink Regulator Gold ATX 3.0 1200W | |
| Storage | TeamGroup T-Force Cardea A440 M.2 PCIe Gen4 NVMe SSD 4TB | |
| Operating System | Windows 11 25H2 | |
| Display Driver | GeForce Game Ready Driver 591.74 WHQL | |
Gaming Benchmarks: Battlefield 6
Ultra
Battlefield 6 using the "Overkill" preset is a good place to start because, at first glance, it appears to prove the point people requesting higher resolution testing think they're making. If we look at the 1080p data we normally provide, the 9800X3D, which is still GPU-limited here, is just 8% faster than the 5800X3D, but a massive 106% faster than the 3800X. The 5800X3D is also 90% faster than the 3800X. Again, a huge increase.
But those margins only tell part of the story, and it's not even the part you should care most about, especially if you believe resolution has any real relevance here. As important as the margins are, the real sticking point is "frame rate."
Battlefield 6 is a competitive online shooter. We'd personally consider it more of a casual shooter, but the objective is still to eliminate enemy players and win the match, so by its very nature it's competitive. More casual players may value immersion and visuals over frame rate, meaning the 102 fps we see from the 3800X at 1440p could be considered acceptable.
Why Frame Rate Matters More Than Resolution
But not all gamers are the same, and plenty of Battlefield 6 players are focused purely on winning and gaining every competitive advantage possible. These players lower quality settings for two key reasons, both highly desirable for competitive play: 1) it increases frame rate and reduces latency, giving you a better chance of landing your shot first. 2) it improves visibility by making enemy players easier to spot through foliage, explosions, and darker areas.
So circling back, the margins are only part of the story because ultimately there's a subset of gamers targeting a specific frame rate. Personally, we're not playing a game like Battlefield 6 at less than about 140 fps, and ideally we'd want around 200 fps in any competitive shooter.
This isn't an elitist take, either. You don't need an ultra high-end GPU for this level of performance, you just need to reduce the quality settings. But you do need a CPU capable of delivering those frame rates, and that's ultimately where the bottleneck lies in many competitive shooters.
Why 1440p Testing Can Be Misleading
The Ryzen 7 3800X is going to struggle to push much beyond 100 fps regardless of quality settings. While CPU load can be reduced a little, you're not going to maintain 140 fps or more with this processor. But if you looked at the 1440p Overkill results in a vacuum, you could easily be misled into thinking the 3800X is just as fast as the 5800X3D or even the 9800X3D.
The point is that the 1080p CPU-limited results tell both casual and competitive gamers exactly what they need to know. Only need 60 fps? The 3800X is fine. Need 100 fps? The 3800X still gets the job done. But what about 120 fps? At that point the 3800X is no longer sufficient, and you'll need to start looking at faster processors like the 5800X3D.
For those gaming at 1440p with some form of upscaling enabled, performance quickly moves back toward 1080p territory. For example, the 5800X3D is now 53% faster than the 3800X with DLSS Quality enabled. We're still seeing a GPU bottleneck at 158 fps here, so if that's the level of performance you're targeting, then the 9800X3D isn't going to offer much, if any, improvement over the 5800X3D. But that was already true at 1080p using the Overkill preset.
So the addition of 1440p testing with upscaling didn't really tell us anything new, while native 1440p simply showed that even the RTX 5090 limits performance to roughly 103 fps. That also means it's safe to assume slower GPUs will deliver even lower frame rates.
This data becomes largely useless because if you already know you're happy gaming at 60 fps, then the fact that the 9800X3D can push beyond 200 fps is irrelevant. But if you want more than 120 fps, the 1440p GPU-limited data doesn't tell you which CPU can actually deliver that performance level or which one offers the best value while doing so.
Medium
Once we switch to the medium preset, everything changes without really changing. By that we mean the resolution scaling compared to what we saw with the Overkill preset has changed, but the results, margins, and conclusions from the earlier 1080p testing remain effectively identical.
Regardless of resolution, whether that's 1080p, native 1440p, or 1440p with upscaling, the 9800X3D is still roughly 32% faster than the 5800X3D, while the 5800X3D remains about 47% faster than the 3800X.
So in this example, 1440p is no more "real-world" than 1080p. More importantly, once again, it's all about frame rate. If we artificially limited performance to 110 fps by introducing a GPU bottleneck, that doesn't suddenly mean the 5800X3D and 9800X3D are pointless or fail to provide meaningful real-world gains over the 3800X. It simply means you've chosen 110 fps as the point where performance becomes capped.
That would be a pretty silly thing to do, but let's move on.
Marvel Rivals
Ultra
Next up we have Marvel Rivals, and here we see that when looking at the original 1080p data using the "Ultra" preset, the 9800X3D is 23% faster than the 5800X3D and 90% faster than the 3800X. This configuration is also introducing some GPU bottlenecking for the 9800X3D.
Now if we switch to 1440p while enabling upscaling, the margin actually increases to 29%. It's important to note that DLSS isn't a completely free performance booster. There is a small amount of overhead, and that's what we're seeing here. Because the 5800X3D and 3800X are already maxed out, enabling DLSS can slightly reduce performance, though this won't always happen, often resulting in performance remaining largely unchanged.
But if you have loads of CPU headroom available, as is the case with the 9800X3D, then if the GPU can render more frames with DLSS enabled, it will. That's exactly what we're seeing here, with a small 2.5% performance boost over native 1080p.
Now if we compare native 1080p to native 1440p, we do see the 9800X3D run into a significant GPU bottleneck at 1440p, whereas this wasn't the case for the 5800X3D or 3800X. These 1440p results saw the 9800X3D lead the 5800X3D by a mere 6%, while the margin over the 3800X dropped from 90% at 1080p to 58% at 1440p.
But now we're capped at 133 fps, as that's as fast as the RTX 5090 can go at 1440p using the Ultra settings, regardless of the CPU being used. If you want more than 133 fps on average, you'll need to reduce either the resolution or quality settings. As you can see, enabling DLSS Quality upscaling increases performance to 164 fps on average.
Of course, if you're happy with just 60 fps, then the CPU largely doesn't matter in the context of this testing, and even the 3800X could be considered overkill.
Medium
But, if you're willing to reduce visual quality settings in order to boost frame rates, which is what many Marvel Rivals players end up doing, then we see some very significant changes in our testing.
First, the RTX 5090 jumps from a frame rate cap of 133 fps at native 1440p to 210 fps, an almost 60% increase simply by dropping from Ultra to Medium settings.
Even at native 1440p, the 9800X3D is still 27% faster than the 5800X3D and 88% faster than the 3800X. But again, those margins are only part of the story. If you want more than 140 fps in Marvel Rivals, it doesn't really matter that the 9800X3D is 88% faster than the 3800X. At that point, it might as well be 400% faster because the 3800X still isn't reaching your minimum desired target, given that it's limiting average performance to 112 fps.
As for the 9800X3D, if we want to talk about real-world differences, at 1080p it was 40% faster than the 5800X3D, but at 1440p with DLSS Quality enabled, that margin actually increases to 52% because the 9800X3D does a better job handling the DLSS overhead.
Whether or not you need more than 250 fps on average in Marvel Rivals is up to you to decide, but at least this data allows you to make that decision while understanding the true performance of all the CPUs tested.
Baldur's Gate 3
A Useful Example of Consistent CPU Scaling
Ultra
Next up we have Baldur's Gate 3, and this absolutely isn't a game that needs to be played at extremely high frame rates. Anything that keeps the 1% lows above 60 fps is highly enjoyable in our opinion, though of course opinions will vary.
So this data is less about performance in the here and now, and more about using Baldur's Gate 3 as a benchmark tool to gauge not only the raw gaming performance of each CPU, but also how they might compare in future games.
Using the Ultra preset, we see very consistent scaling regardless of resolution. The 3800X does become a severe bottleneck, though, limiting the average frame rate to 79 fps in our test, with 1% lows of 55 fps. Still, as we said, this should be sufficient performance for most people playing Baldur's Gate 3.
Even so, at native 1440p the 5800X3D was 71% faster, while the 9800X3D was an incredible 162% faster. Those figures increase only slightly at 1080p, where the 5800X3D is now 85% faster and the 9800X3D 186% faster.
All three test configurations maintained nearly identical margins between the 5800X3D and 9800X3D, so in reality it wouldn't really matter whether we tested at 1080p or 1440p in this example, as the margins remain largely the same.
Medium
We see much the same story with the medium settings. The 9800X3D was 50% faster than the 5800X3D when testing at 1080p, then 43% faster at native 1440p and 48% faster at 1440p with upscaling enabled.
The gap to the 3800X grows even larger, and now even at 1440p the 9800X3D was 165% faster, though again, you probably don't need 228 fps in Baldur's Gate 3.
Cyberpunk 2077: Phantom Liberty
Ray Tracing Shifts the Bottleneck Toward the GPU
Ultra
Next up we have Cyberpunk 2077: Phantom Liberty, and these results are quite interesting. First, the 9800X3D delivered higher frame rates when using ray tracing at 1440p with upscaling enabled compared to native 1080p. This happens because these results are heavily GPU limited, while 1440p DLSS Quality actually renders at a lower internal resolution than native 1080p.
However, at native 1440p performance drops to just 94 fps, and here the 9800X3D is only 3% faster than the 5800X3D, though it still manages to be 24% faster than the 3800X.
That said, once upscaling is enabled, the margin over the 3800X jumps to 88%, while the 9800X3D also becomes 28% faster than the 5800X3D. More importantly, enabling DLSS increases average performance from just 112 fps to a much smoother high refresh rate experience at 143 fps.
It's also worth noting that while native 1440p heavily restricts the 9800X3D to sub-100 fps performance, enabling DLSS, which most gamers apparently do, actually increases the CPU margin beyond what we saw in the original 1080p testing. The lead over the 5800X3D grows from just 13% to 28%.
Medium
Now if you don't care about ray tracing and would rather maximize frame rates, the medium quality preset still looks good while performance goes through the roof.
We also find a scenario where 1080p and 1440p deliver effectively identical results with the RTX 5090 because all three test configurations are CPU-limited, which is exactly what you want to see when evaluating CPU gaming performance.
As a result, the data here remains unchanged from our original testing, so let's move on.
Space Marine 2
A Heavily CPU-Bound Modern Game
Ultra
Moving on to Space Marine 2, we have a very CPU-demanding game, so it wasn't surprising to find that including 1440p results produces numbers very similar to what we see at 1080p, even using the Ultra quality preset.
This is a great example of a game that is primarily limited by CPU performance, even at high resolutions with maxed-out quality settings.
Medium
That being the case, switching to the less demanding medium preset shows more of the same. Once again, whether we test at 1080p or 1440p makes little difference here.
Even so, why risk unnecessarily introducing GPU bottlenecks into CPU testing simply by increasing the resolution?
Assetto Corsa Competizione
High Frame Rates Expose CPU Limitations Clearly
Ultra
Although Assetto Corsa Competizione is a racing simulator, it also represents the kind of performance scaling you'll see in unrelated games such as Apex Legends or Fortnite, as it's a lightly threaded title that performs extremely well at high frame rates.
Even using the Epic settings, we're looking at comfortably more than 200 fps with the 9800X3D, making it at least 50% faster than the 5800X3D.
Medium
Given that the Epic preset already revealed nearly identical performance metrics at both 1440p and 1080p, it's entirely unsurprising that the subsequent testing using the Medium quality preset produces the same universally CPU-limited data.
This clearly indicates that across a range of graphical settings, the bottleneck preventing higher frame rates isn't the GPU, but rather the CPU itself. The system is simply hitting a wall where the processor can no longer feed data to the GPU fast enough to achieve higher performance.
Spider-Man 2
Ray Tracing Still Leaves the Game Heavily CPU-Limited
Ultra
Moving on to Spider-Man 2 using the Ultimate ray tracing preset, we previously found when testing at 1080p that the 9800X3D was on average 44% faster than the 5800X3D, which is obviously a significant performance uplift. We also saw that the 5800X3D was 29% faster than the 3800X, another sizeable improvement.
Now when we introduce 1440p results, either at the native resolution or with upscaling enabled, the margin between the 5800X3D and 3800X remains completely unchanged, as the data was already heavily CPU-limited to begin with. We know the RTX 5090 can render at least 104 fps on average, as demonstrated by the 9800X3D results.
This is important to note because readers often claim that testing older or lower-end CPUs with an RTX 5090 is unreasonable, unrealistic, or something along those lines. That's an argument based more on feeling than logic. It doesn't matter whether you use an RTX 5090 or an RTX 5070 Ti. If the GPU can render more than 70 fps in this test, you're going to end up completely CPU limited, exactly as we see here.
For the 3800X configuration to become GPU limited, you would need to step down to something like the RTX 5060 Ti, which is still a very capable GPU. But once we go down that path, we're also missing the point because now we're talking about sub-60 fps PC gaming. In that situation, we'd always look to reduce quality settings for a performance boost.
Medium
But even if we significantly reduce the visual quality settings, the 3800X still struggles to push beyond 90 fps for the 1% lows, while averaging around 120 fps overall. That's still very good performance, but again, if you want more performance, a processor like the 5800X3D is going to be more than 50% faster, and that's while using what most people would consider "real world" quality settings.
Horizon Zero Dawn Remastered
A Balanced CPU and GPU Workload
Ultra
Horizon Zero Dawn Remastered is demanding on both the GPU and CPU. As a result, it doesn't really matter whether we test at 1080p or 1440p because the results end up effectively identical. That's exactly what we see when testing using the "Very High" preset, which is the highest quality preset available in the game.
Medium
Given what we saw with the Very High quality preset, it comes as no surprise to find that the 1080p and 1440p results remain virtually identical using the medium quality preset as well.
Crimson Desert
Another Example of CPU Throughput Limitations
Ultra
Finally, we have Crimson Desert, and this is yet another example where 1440p testing doesn't meaningfully differ from 1080p testing.
We learn that the 3800X can't deliver much more than 85 fps on average using the Cinematic preset, and that limitation obviously doesn't disappear simply because the resolution is increased.
Medium
Even when using the medium preset, the 3800X gains very little additional performance, improving by up to just 5%. To be fair, though, we saw similarly small gains for the other two CPUs, strongly suggesting that all of this data is entirely CPU-limited.
14 Game Average
Ultra Settings
So if we were to test at 1440p rather than 1080p, how would that have changed our findings and recommendations? Looking at these three CPUs, not much would have changed.
At 1080p, the 9800X3D is 33% faster than the 5800X3D, while at native 1440p it's 27% faster. That difference doesn't really change the overall conclusion. But had we tested at 1440p with Quality upscaling enabled, using the settings most 1440p gamers actually use, the 9800X3D ends up 36% faster.
The margin over the 3800X does change somewhat, but even then the 5800X3D is still 44% faster at 1440p compared to 62% faster at 1080p. That's a notable difference, but as we've alluded to many times now, if you're targeting more than 80 fps, it doesn't really matter whether the 5800X3D is 44% or 62% faster.
What matters is that it's substantially faster and potentially capable of delivering the level of performance you're actually looking for.
Medium Settings
However, that was using maxed-out quality presets. If we switch to medium settings in search of higher frame rates, the CPU becomes much more important.
At that point, it barely matters whether we test at 1080p or 1440p because the data remains largely CPU-limited, meaning the performance margins stay effectively unchanged.
What We Learned
So there you have it. Despite so many people requesting or even demanding 1440p CPU testing, it adds surprisingly little to the discussion while risking muddying the data in ways that can ultimately mislead readers.
A big part of the problem is that many of these arguments assume all gamers fall into the same category, and that's simply not how PC gaming works. Not everyone plays the same games, uses the same settings or hardware, or prioritizes visuals over frame rate, and vice versa.
That's why showing the true gaming performance of a processor remains the most useful approach. That data isn't just relevant for enthusiasts chasing every last frame, it's equally valuable for people simply trying to understand where products genuinely sit relative to one another. It gives you the complete picture.
Testing CPUs under GPU-limited conditions, on the other hand, rarely tells you anything meaningful about the processor itself. In many cases, it almost feels like some people want GPU-limited CPU testing because it helps justify delaying an upgrade. "Look, the 9800X3D is no faster than the 5800X at 4K when the GPU caps performance at 50 fps." But that's completely missing the point of CPU benchmarking in the first place. The goal will never be to make anyone feel bad about their current hardware configuration or pressure them into upgrading.
If the 5800X can deliver more than 300 fps in Rainbow Six Siege and all you need is a frame rate north of 200 fps, then it simply doesn't matter that the 9800X3D can push beyond 500 fps. Artificially handicapping both processors with a GPU bottleneck doesn't help anyone.
If the Ryzen 7 5800X can already push more than 300 fps in Rainbow Six Siege and all you care about is maintaining a frame rate north of 200 fps, then the fact that the 9800X3D can exceed 500 fps simply doesn't matter to you. You don't need that level of performance, and your current setup is already doing the job. Artificially handicapping both processors with a GPU bottleneck doesn't help anyone.
Showing uncapped CPU performance, at least to the best of our ability, isn't misleading, elitist, or disconnected from the real world. We've been PC gamers for nearly 30 years, and throughout that time the goal has largely remained the same, whether we were playing Quake III Rocket Arena on a 16 MB Voodoo Banshee or modern titles on an RTX 5090: maximize performance, minimize limitations, and understand where the bottlenecks actually are.
Showing the actual uncapped performance – at least to the best of our ability – isn't misleading, elitist, or disconnected from the real world. We've been PC gamers for nearly 30 years now, and during that time we've always chased maximum frame rate performance, whether that was in Quake III Rocket Arena or Fortnite, using everything from a 16 MB Voodoo Banshee to an RTX 5090. The goal has always been the same.
But even if you're primarily a single-player gamer targeting around 60 fps, it's still important to understand how products within your budget truly compare because that allows you to make a more informed purchase.
Some people will inevitably ask: if the results are often similar at 1440p, then why not just test at 1440p? The answer is simple. It's a flawed methodology because it can, and often does, introduce GPU bottlenecks that mask the CPU's true performance, resulting in misleading data.
Resolution only matters to the extent that it impacts GPU load enough to turn the GPU into the primary performance bottleneck.
The only reason we no longer test at 720p is because we eventually reached a point where the games and hardware configurations we were using no longer showed meaningful changes in performance margins between 720p and 1080p under any conditions. In fact, we've seen instances where frame rates regressed at 720p. We believe this may have been caused by there not being enough pixel data to keep all the cores fed, creating inefficiencies within the GPU, or perhaps it's just a driver issue because GPU makers won't optimize for very low resolutions. Whatever the explanation, it became increasingly common.
At the end of the day, the real issue here is the fixation on resolution itself, which is misguided to begin with. Resolution doesn't matter for CPU testing because the CPU doesn't inherently care about rendering resolution. Resolution only matters to the extent that it impacts GPU load enough to turn the GPU into the primary performance bottleneck.
Claiming that 1080p CPU testing is useless simply because you personally game at 1440p or 4K demonstrates a misunderstanding of what CPU benchmarking is actually trying to measure.
We've done our best trying to educate gamers on this topic over the years, and with 56% of the audience still asking for higher-resolution CPU testing, it seems we still have a lot of work ahead of us. Hopefully, though, this latest explainer helped clarify why the methodology exists in the first place.