Gamers Are Wrong About 1440p vs 1080p CPU Benchmarking
- Thread starter Steve
- Start date
Oops! Many sites do indeed test at multiple resolutions, and, obviously while no one can test all possible combinations, the most sensible approach is to test at the dominant resolution today.
You may be too young to remember that the "standard" of 1080p testing developed when a substantial minority of users ran at 720p and 640p. But 1080p was used because it was the most commonly used resolution at the time. If "removing the bottleneck" was the sole consideration, we'd still be testing at VGA resolutions. Much less GPU load! No more bottlenecks!
You're still pushing the myth of "real cpu performance" against all evidence to the contrary?
Last edited:
So you don't understand gaming testing. Got it.
CPU testing tells you ONE thing: Max FPS possible regardless of GPU.
GPU testing tells you how the game will play if you aren't hitting the CPU Max.
If you want to see "the full picture" you have to look at BOTH tests (or you are objectively wrong).
There is no math or assumptions. If your current CPU can do 275 FPS at 1080p (your example), then it can do 275 FPS at 4K Ultra with Path Tracing - IF you GPU is strong enough to do that too. (Spoiler it won't be but the future RTX 7090 may). If a CPU can do 75 FPS at 1080p, it can only do 75 FPS at 1440p or 4K even if you have a 5090 or 6090 or 7090.
See it's simple. But you have to look at two graphs instead of one.
poshflamingos
Posts: 340 +672
Some do, and it shows how pointless it is. Look at techpowerup's 4K tests for example, 80% of the tested CPU just end up tied at the top due to being GPU limited. How is that useful for anyone?
No, the most sensible approach is to test in a way that produces the most useful data for everyone, regardless of what their individual specs are. And that way is using the best available GPU at low resolutions, to minimize GPU bottlenecks. Anything other than that produces data that isn't useful.
Yes, because it's not a myth, and the evidence supports me.
Look at these 4K tests done with the 2080 Ti in 2019. Look at how the 3700X and the 2700X compare in them. They're tied, right? So you would conclude "the 3800X offers no advantage in 4K", right?
Well, then, look at this other 4K test done in 2024 with the 4090. How do the 3700X and 2700X compare now? What do you think happened? Do you think the 3700X grew magically faster over time and became 12% faster five years later? Or do you think the 3700X was always 12% faster, only the 2080 Ti just wasn't fast enough to show it because the 4K tests were GPU-limited?
Except you didn't need to wait 5 years to discover this. If instead of looking at the 4K tests in 2019 you looked at the 720p tests, you would have seen the 3700X is 11% faster than the 2700X in there already. That's the real performance difference between those CPUs. You could see it in 2019 by testing at low resolutions, but once better GPUs came out years later, they exposed that exact same performance difference on the 4K tests as well.
That is exactly why testing at low resolutions is the only correct way to test CPUs. It exposes the real CPU performance right now, instead of having to wait years for better GPUs to show that that same performance difference also existed at 4K but was being hidden by GPU bottlenecks.
Yes, I agree with the test.
To show the problem of weak CPU gaming performance.
How is the CPU tested in the game? Game benchmark (2min test).
Real gaming on a weak CPU (Ryzen 5600X, 3800X), if it stutters (game bug) 1080p RX6700XT, it will stutter 2160p RTX5090
CPU testers cannot simulate:
Video, author's system reddit- RTX 3080, Ryzen 9 5900X, 32GB RAM 3600MHz.
Ryzen X3D solves these problems.
To show the problem of weak CPU gaming performance.
How is the CPU tested in the game? Game benchmark (2min test).
Real gaming on a weak CPU (Ryzen 5600X, 3800X), if it stutters (game bug) 1080p RX6700XT, it will stutter 2160p RTX5090
CPU testers cannot simulate:
Video, author's system reddit- RTX 3080, Ryzen 9 5900X, 32GB RAM 3600MHz.
Ryzen X3D solves these problems.
MasterMace
Posts: 410 +424
I'd be more curious about seeing if Explicit Multi-Adapter actually works in games to open the graphics bottlenecks in these higher resolutions
It seems pretty obvious - it's pointless testing a component when the bottleneck is elsewhere. It becomes basically meaningless. In a similar analogy you wouldn't test tyres by putting them on a Model-T ford because the car is incapable of stressing them in any way.
Yes, I think you have gotten to the heart of it.
Because it answers the question directly: at 4k, it makes no difference.
There's two question being asked: Steven is describing how to answer the question "how does CPU performance compare to each other", and the methodology proposed is the correct one to answer that question.
The question many gamers want the answer to is "how will this upgrade affect _my_ frame rate today", and while a publication cannot test at every possible combination, a chart that shows a tiny difference at a specific resolution answers that question directly, no thinking or inference required, which is why they find those valuable, even if it provides no information to answering the relative CPU performance question.
To Endymio's point, the price per performance metric only matters when the performance is based on what you actually are getting, which depends on which part of the system is bottlenecking (aka, the resolution one is playing at). To Steven's point, 1080p testing answers how CPUs behave relative to each other with minimum (possible) bottlenecking.
And that leads to the third question that sits in between these two: "at what point does bottlenecking switch from the CPU to the GPU, and by how much?".
Both testing strategies are correct, but they answer different questions. The question one wants answered informs how they answer the poll, but only multiple resolutions (or whatever setting meaningfully shifts the bottleneck, as the article says, it doesn't have to be the resolution) answers the third question.
Glad you asked! Because it tells you that, if you're interested solely in 4K gaming performance, you're wasting money by purchasing an overpowered CPU. See how simple this all is?
kiniku
Posts: 14 +15
While I understand the logic of using a flagship GPU like the RTX 5090 at 1080p to 'remove the bottleneck,' it creates a data set that is fundamentally misleading for 99% of the audience.
By testing in a 'GPU-bound vacuum' that costs $4,500 to enter, you are artificially inflating the performance gaps between architectures (like AMD’s X3D vs. Intel’s Core Ultra).
Here is why this methodology fails the 'Real World' test:
Performance Compression: If you were to run these same CPUs with an RTX 5070 or even a 5080—the cards people actually buy—those 20% performance gaps frequently shrink to 2-3%. Recommending a CPU based on a 1080p/5090 chart is like recommending racing tires for a commuter car because they performed better on a track with no speed limits.
The Resolution Reality: Almost nobody buying a Core Ultra 250K or a 9850X3D is playing at 1080p. On a high-fidelity display (1440p or 4K), the GPU is the worker, and the CPU’s 'theoretical ceiling' becomes irrelevant.
The Bottom Line: those charts don't show which CPU is 'best'; they show which CPU is best if money is no object for the GPU. For the rest of us, these 'victories' are invisible in actual gameplay.
By testing in a 'GPU-bound vacuum' that costs $4,500 to enter, you are artificially inflating the performance gaps between architectures (like AMD’s X3D vs. Intel’s Core Ultra).
Here is why this methodology fails the 'Real World' test:
Performance Compression: If you were to run these same CPUs with an RTX 5070 or even a 5080—the cards people actually buy—those 20% performance gaps frequently shrink to 2-3%. Recommending a CPU based on a 1080p/5090 chart is like recommending racing tires for a commuter car because they performed better on a track with no speed limits.
The Resolution Reality: Almost nobody buying a Core Ultra 250K or a 9850X3D is playing at 1080p. On a high-fidelity display (1440p or 4K), the GPU is the worker, and the CPU’s 'theoretical ceiling' becomes irrelevant.
The Bottom Line: those charts don't show which CPU is 'best'; they show which CPU is best if money is no object for the GPU. For the rest of us, these 'victories' are invisible in actual gameplay.
fadingfool
Posts: 461 +617
I voted for two resolutions not for the the 1440p but because I am not interested in "Ultra". It is often a performance hurting setting with little visible benefit.
Mark Fuller
Posts: 805 +188
Data transfer rates are not that much affected by screen resolution. If anything, it requires more power the higher the resolution, putting an additional burden on the processor. Looks like to me that's why 1080p was usually faster in your tests with the 9800X3D and 5800X3D processors, and as least as fast as the 3800X. People need to do a lot more research into what they are doing, like you do, before they invest. Bigger numbers don't necessarily mean increased performance. If you invest to upgrade a screen from 1080p to 1440p, you increase the power requirement, given the same cpu and gpu, meaning slower data transfer rates.
Last edited:
Chadeuk
Posts: 6 +2
Both sides are right. It's just that Steven is saying he's trying to show the performance difference in general. While the other side wants to see how it will affect them in a setting closer to their current situation specifically. But saying it's pointless Steve, isn't correct, unless you're simply not interested in what your viewers want to see.
You are spot on, it'd be more useful to test with quality presets than same preset but different resolutions. Since some games do just give the GPU more work at higher presets, but some do things that increase CPU load. Naturally for a CPU benchmark (as said in the article and comments), you're no longer testing the CPU if the game becomes GPU-bound.
I'll tell you an example I had of this -- I am now running my GTX1650 in a Coffee Lake (i7-8700) but had it in an Ivy Bridge (i5-3470.) The only thing that'd get that GPU up to 99-100% load on there was GravityMark. It was CPU (and possibly PCIe speed) limited on any other game (well ones that didn't hit 60FPS and cap there). So I ran CP2077 on there, on low, and got whatever FPS (like 28FPS or so as I recall.) Ran it on medium. 28FPS. Ran it on high. 27FPS. This was at 1280x720. It just drove the GPU load up from like 30% on low to about 80% or so on high. (This was before the 2.0 patch came out, which made the game more demanding.) In this case presumably putting it on high didn't increase CPU load (or it did, but on other cores than the one feeding the GPU.) I had a few other games like that too, might as well run high settings since it was having little to no effect on frame rate (just driving up the GPU usage from under 50% to some more reasonable number.)
But, whatever, looking at the 1080 results will still be informative even if the 1440 results aren't so much.
I'll tell you an example I had of this -- I am now running my GTX1650 in a Coffee Lake (i7-8700) but had it in an Ivy Bridge (i5-3470.) The only thing that'd get that GPU up to 99-100% load on there was GravityMark. It was CPU (and possibly PCIe speed) limited on any other game (well ones that didn't hit 60FPS and cap there). So I ran CP2077 on there, on low, and got whatever FPS (like 28FPS or so as I recall.) Ran it on medium. 28FPS. Ran it on high. 27FPS. This was at 1280x720. It just drove the GPU load up from like 30% on low to about 80% or so on high. (This was before the 2.0 patch came out, which made the game more demanding.) In this case presumably putting it on high didn't increase CPU load (or it did, but on other cores than the one feeding the GPU.) I had a few other games like that too, might as well run high settings since it was having little to no effect on frame rate (just driving up the GPU usage from under 50% to some more reasonable number.)
But, whatever, looking at the 1080 results will still be informative even if the 1440 results aren't so much.
poshflamingos
Posts: 340 +672
This isn't just about overpowered CPUs. You may be deciding between a Ryzen 5600 vs a 7600, or between a 12400F vs a 250K Plus. None of those are expensive/overpowered CPUs. Testing at 4K can mislead you into thinking there's no difference between them, but testing at lower resolutions shows the 7600/250K are around 30% faster than the 5600/12400F, and those 30% will eventually show up even at 4K when you upgrade your GPU later, and/or when more CPU-demanding games come out.
Besides, you don't need the reviews to test 1440p or 4K to know this. The 1080p test already tells you what the uncapped performance of a CPU is. If the 1080p test shows the CPU can run Game X at 120 FPS, but your GPU runs Game X at 80 FPS at whatever resolution you use, you already know that CPU won't give you more performance because you'll be GPU-limited, you'll just get 80 FPS regardless. You don't need the reviewer to test 1440p/4K to spoonfeed you this information, you can already figure it out yourself by looking at GPU tests.
That's the point in a nutshell. Benchmarking isn't about numbers, it's about comparing numbers. Fact is 1440p is "real world"... kind of. It's as common a default resolution as 1080p if not more common. That's IMHO where this "real world" issue/idea comes from. But it simple means your comparing different numbers and the GPU comes more in to play on the effects on those numbers so they're not as consistent from CPU to CPU.
I remember when CPU testing used lower resolutions by default. 720p or even as low as 800x600 or 1024x768. The basic idea being to make the CPU the bottleneck instead of the GPU, which would happen with a higher resolution, and why testing at 1440p in the end is a waste of time. The problem is just like with bottlenecking, too many users are obsessed with numbers, especially in comparison to their system/s.
More importantly it's like arguing with anyone who has a poor understanding of the issue at hand like anti-vaxxers, or flat earthers, and have decided it's the hill they'll die on. The more we explain why 1440p is a waste of time, the more they'll dig their heels in because that's their systems default setting. Eventually a few might see reason, but humans are a stubborn bunch, and if it was me I wouldn't waste my time trying to talk sense to the senseless...
poshflamingos
Posts: 340 +672
Except that's not true.
It shows you that, with the GPU used on the test, there is no difference. But in the future, faster GPUs will exist, games that are more demanding on the CPU will come out, and there will be a difference then. That's why knowing how the CPUs perform when not GPU-limited is more important.
See the links in the comment you replied to. Looking at the 4K tests in 2019 would have made you think there's no difference in performance between the 2700X and 3700X. On the 2024 test, with a faster GPU, the 3700X is 12% faster. It was always 12% faster, you just couldn't see it in 2019 because the 4K tests from 2019 were GPU-limited and hid that difference.
It doesn't matter what the gamers want the answer to, CPU tests are not made to answer that question. Reviewers can only answer that question for you if they do the test with the same GPU you have, but then that test is useless to everyone who has a different GPU than you.
No, it doesn't.
A chart like that would mislead less savvy gamers into thinking those CPUs perform the same, when they don't.
The only correct way to answer that question for them is to provide the proper, non-GPU-bound performance of the CPUs, and have people cross-reference it with GPU tests to know whether they'll be CPU-bound or GPU-bound at whatever resolution they use.
If you show a chart with 80% of the CPUs tied because they're all GPU-limited (like techpowerup's 4K tests do), the correct conclusion is not that "these CPUs perform the same at 4K", it's "this GPU isn't fast enough to show the real difference, but future GPUs will". The CPU performance difference is still there, even at 4K.
So you're basically saying you want reviewers to double/triple their workload and make reviews take much longer to come out (or cut the number of games they can test in half/a third, making for a much less representative pool of games), and put out data that is potentially misleading to the less savvy readers, just so you can have the tiniest of conveniences?
No reviewer will ever be able to answer that for you, because the point where that happens also changes per each game.
Again, just figure it out yourself by looking at proper low resolution CPU tests, plus GPU tests.
The low resolution tests done with a RTX 5090 answer "what is the real performance of those CPU" for everyone.
The high resolution tests done with a RTX 5090 answer "what is the performance of those CPUs at 4K when paired with the 5090", which is only useful for people who also have a 5090. If you don't have a 5090, this test answers jack squat to you.
The low res test done with the 5090 answers „what is the performance of those CPUs at low resultion when paired with a 5090“ while the high res test done with the 5090 answers „ what is the performance of those CPUs at 4K when paired with the 5090“ - both results / performance measurements are equally „real“. Where they substantially differ is relevance for the audience / specific reader. I for myself dont see any relevance in CPU tests done with a 5090 and in 1080p, because I neither waste thousands of bucks on a GPU nor do I game in 1080p since like 8+ years. My „gaming“ is >90% done in GPU limit, so I dont care for tests which (artificially) remove this - if you or others do, fine.
If Id be up to purchase a new car for everyday use, I dont read reviews which test on track with racing fuel and slick tires either.
Do a reality check… the last like 20 years have clearly shown that games (aswell as growing resolutions) mostly demand more GPU power, not CPU power.
An i5 8400F with a GTX 1060 Ti was a typical p/p midrange system from late 2017. Now, nearly 9 years later, we hopefully dont need to discuss that upgrading to a current gen midrange GPU like a 5060 Ti (and keeping the old CPU) would run most current games atleast relatively fine while upgrading to a current gen midrange CPU like a R5 9500F (and keeping the the old GPU) would absolutely not - or do we?
poshflamingos
Posts: 340 +672
No, genius. Testing at low resolutions with the best GPU available gives everyone information about where the ceiling for performance with that CPU is. The purpose of using an expensive GPU at low resolutions is exactly to ensure that the GPU isn't interfering with the measurement at all, you just want to measure how the CPU performs when the game is CPU-limited.
On the other hand, testing CPUs at 4K just produces a lot of useless data, because then you're not measuring the CPU at all. You just get a bunch of tests that are GPU-limited with the CPUs sitting under-utilized.
The purpose of CPU tests is not to emulate how you use your PC. The world doesn't revolve around you.
Car analogies LMAO
This second comment of yours made it abundantly clear that you have no idea how CPU performance actually works in games. But worry not, I will teach you.
No, it's not only the GPU that matters. Games demand performance from both the CPU and the GPU. Depending on the game, both the CPU portion and the GPU portion can be heavy or lightweight. This article right here that you're commenting under has examples of games that are more demanding on the CPU than they are on the GPU (like Baldur's Gate 3), and games that are equally demading on the CPU as they are on the GPU (like Space Marines 2).
The way it works is that, for each particular game (with its own set of demands for CPUs and for GPUs), you'll have a maximum framerate your CPU allows and a maximum framerate your GPU allows, and the performance you'll actually see is the lower of the two. If you have a CPU capable of running Game X at 100 FPS, and a GPU capable of running Game X at 50 FPS, you'll see 50 FPS (and in that case you're GPU-limited, with the CPU under-utilized). But if you have a CPU capable of running Game X at 60 FPS and a GPU capable of running it at 90 FPS, you'll see 60 FPS (and in that case, you're CPU-limited, with the GPU sitting under-utilized).
Now suppose you take CPU Y, and you'll test it with Game X. Suppose you first pair it with a RTX 5060, run it and get 60 FPS. Then you pair it with a 5060 Ti, and you get 70 FPS. Then you pair it with a 5070, and you get 80 FPS. Then you pair it with a 5070 Ti, and you get 90 FPS. Then you pair it with a 5080, and you get 90 FPS again. Then you pair it with a 5090, and you get 90 FPS again. Do you understand what happened there? That's right, once you moved to a 5070 Ti, you hit the ceiling of what CPU Y could do in Game X. Beyond that point, GPU upgrades don't give you more performance, you're CPU-limited and those biggers GPUs (5080, 5090) are sitting under-utilized. No matter what you do, CPU Y will never give you more than 90 FPS on Game X.
That is the point of testing with the best available GPU at low resolutions. The goal is to find exactly that, how a CPU performs when CPU-limited, to understand what its performance ceiling is. The best GPU at low resolutions ensures you're not running into GPU bottlenecks that interfere with that measurement.
First of all, "GTX 1060 Ti" doesn't exist.
Second, you would have a disastrous experience in modern games with a 8400F. You'll be CPU-limited well below 60 FPS in today's demanding games with that CPU. There's no point in upgrading a 8400F system to a 5060 Ti, that CPU's ceiling is too low to properly make use of that GPU. You'll run games that the 5060 Ti could be running at 100+ FPS, but end up with 40 FPS anyway because the CPU can't go any higher than that.
Not that upgrading the CPU instead would be much better, you'd end up in the opposite situation, a powerful CPU being under-utilized due to being heavily GPU-limited. Someone with a 8400F + GTX 1060 would have to upgrade both to see meaningful gains.
I say this from experience, I did exactly that. Until two years ago I had an old system with an i7-4770 that was still chugging along. It went through three GPU upgrades (R9 380 > GTX 1060 > RTX 2060 Super). But that last upgrade was a mistake, by the time I got the 2060 Super most games on my PC were CPU-limited. For example, in Elden Ring the i7-4770 + 1060 system would drop to ~40 FPS in some spots in the open world. When I put the 2060 Super in its place, it still dropped to the exact same 40 FPS on those spots, I gained no FPS with it. It was CPU-limited, I could turn the resolution up without losing any FPS (because the GPU was under-utilized), but nothing could make it go above 40 FPS in those spots, because that was the best my CPU could do. Taking a 8400F system and upgrading to a 5060 Ti would be an even worse version of that.
Last edited:
brucek
Posts: 2,353 +3,774
If your budget is unlimited, you only the need the low resolution testing, because it tells you which CPU has the most power. And why buy any other.
But as soon as you have to allocate that budget across components, now you need to know where incremental spending is going to do more for you: CPU or GPU? And that's where you need the second set of resolutions, because it tells you how much realized contribution if any you can expect from that potential power.
If it's so hard to test two resolutions, pick one reviewer partner to split them up with, share the data, and then you can both benefit from the fuller picture that your readers just told you they value.
But as soon as you have to allocate that budget across components, now you need to know where incremental spending is going to do more for you: CPU or GPU? And that's where you need the second set of resolutions, because it tells you how much realized contribution if any you can expect from that potential power.
If it's so hard to test two resolutions, pick one reviewer partner to split them up with, share the data, and then you can both benefit from the fuller picture that your readers just told you they value.
You're putting words in my mouth and you're misconstruing my point by selectively listening to different parts of my post.
Let me be clear: I'm making no argument that the method Steve proposed or that Techspot uses is invalid, nor am I saying they should increase their workload by adding configurations to their testing. The 1080p methodology makes clear which CPUs perform well with respect to each other. What I am saying is that it takes work to go from that methodology to answer questions that the test does not directly answer, questions that myself and others have pointed out, and indeed, there are questions that the 1080p methodology alone cannot answer.
Similar threads
