Today we're taking an updated look at everyone's favorite CPU, the Ryzen 7 5800X3D, because I need to get my quota in. And it's been a few weeks now since I last did a 5800X3D video, which is quite bad. Actually, the truth is I have been trying to come up with a practical solution to a problem, a problem I have now dedicated over half a dozen videos to, but I've made very little progress. And I am of course talking about the problem of how to test CPUs, or rather how to convince a large portion of our audience how we should be testing CPUs. Not like which socket do they go in, we worked that out a little while ago. Now I mean more the testing methodology side of things. You see, the best, most scientific method is to isolate the component that you're actually testing, so in this case the CPU. And the best way to measure gaming performance is to use the fastest possible GPU and then run it at the lowest practical resolution. This helps to minimize the GPU performance limitations, giving us a clearer look at actual CPU performance. Currently, this would mean testing with an RTX 5090 at 1080p. You could go down as low as 720p if you wanted to, but we've found a few examples where the frame rates actually decrease. And we believe the reason for this is because there isn't enough pixel data to efficiently utilize all the cores of the 5090. But whatever the case, we found that 1080p is the sweet spot.
Today's video is brought to you by Sapphire and their premium Nitro Plus range of AMD Radeon 9000 series graphics cards and X870 motherboards. For enthusiasts seeking a no-compromise showpiece system, Sapphire's Nitro Plus Phantom Link graphics cards allow for cable-free power delivery when paired with Sapphire's Phantom Link X870E Nitro Plus motherboard. This is one of the cleanest high-end AM5 builds you can put together, while also supporting PCIe 5.0, Wi-Fi 7, and a stacked I/O. The Nitro Plus Radeon RX 9070 XT OC Phantom Link GPU supports a large triple-fan cooler and plenty of raw compute for 4K gaming, while providing 16GB of GDDR6 memory and all the latest RDNA 4 features. Sapphire's Phantom Link provides a serious step up in aesthetics without sacrificing performance. So, for more information, please check the links in the video description.
This is without question the right way to test CPU gaming performance. It's the best way. It shows you the true performance of the CPU right now, and it has proven in almost all cases to indicate future performance very accurately. And as a quick example of this, if we go back to the year 2022 and compare three Ryzen CPUs running games that were available at the time, such as Watch Dogs Legion, Shadow of the Tomb Raider, and Horizon Zero Dawn, this is how they compared at native 1080p using an RTX 3090 Ti, which was the flagship at the time. We see that the 5800X3D is much faster than the 3950X, 33% faster in that example. But had I tested at 4K with balanced upscaling enabled, so not even native 4K, we would have found that all three CPUs delivered the same gaming performance. And this is really important because at the time some tech enthusiasts claimed that the 3950X, which at the time was heavily discounted but still more expensive than the 5800X3D, but they claimed that the 16-core CPU was a better choice for gamers as its additional cores meant that it would provide higher frame rates and smoother game performance in future more demanding titles. And this was despite the low-resolution data suggesting otherwise.
Now, fast forward just a few years to 2024. Using the RTX 4090 in games such as Starfield, Space Marine 2, and Star Wars Jedi Survivor, all more CPU demanding games than the ones we tested with previously, we see that at 1080p, the 5800X3D's margin over the 3950X has actually extended out to almost 50%. Not only that, but we now have a faster flagship GPU along with more demanding games that require more of your CPU. And that means we're now seeing a big increase for the 5800X3D even at 4K with balanced upscaling. So, the takeaway here being that had we focused on GPU-limited CPU testing in 2022, or even shown it, which can mislead people because they think it doesn't make a difference and could lead you to believe that the better bet would be the 3950X, as again some suggested it would eventually come out on top thanks to its many more cores, had all of that happened, you would have purchased a much worse gaming CPU. But anyway, moving on from that, the biggest and frankly strongest argument we see for GPU-limited CPU benchmarking is that people want to know what the CPU upgrade will offer them their specific hardware configuration.
Now, with all due respect, that's very much asking me a question like, how long is a piece of string? Because you're putting it on me to decide just how GPU-limited your CPU results should be, and then claiming that would be somehow useful, more useful than 1080p CPU-limited results, which isn't true and frankly just doesn't make sense. But here we are. The problem is soon we'll be reviewing next-generation products like AMD Zen 6 and Intel's Nova Lake CPUs, and we're going to run into a wave of the same old comments about how useless 1080p testing is for the people that are making those comments. But the good news is I have come up with a solution, or you know, at least I think I have. Round those people up. No, no, I joke, of course. Instead, I've come up with a blue bar graph solution, the best kind of solution. But before I show you the best kind of solution that I've come up with, let's go over what the problem actually is. And for those of you who are sick of this, I was going to promise it'd be quick. I'm not sure how quick it'll be. I'll attempt to go over it as quickly as I can.
As an example, here's the 14-game average data from my Day One 9800X3D review published two years ago. Now, as you can see, at 1080p using an RTX 4090, the 9800X3D was on average 39% faster than the 5800X3D. Not a massive margin given the circumstances, and that's probably why so many of you have stuck with the 5800X3D. But even so, this data was problematic for some, as it made them feel as though the 5800X3D was now inadequate, even though it is rubbing shoulders with the likes of the Ryzen 7 7700X while averaging over 140 FPS. Others have claimed that this sort of data is misleading, as most won't see an almost 40% increase when upgrading from a 5800X3D to a 9800X3D, as their games will still mostly be GPU-limited, which is true for a lot of situations. The thing is though, you should only upgrade your CPU when you need to. Crazy concept, I know. And you shouldn't do it simply because you're starting to generate feelings of inadequacy on behalf of your CPU. Probably don't do that. The 5800X3D, it doesn't care. It's an inanimate object. I promise your 5800X3D does not care how much faster the 9800X3D is. I promise.
So ideally, you should only upgrade your CPU when you need to. And once that point in time becomes clear, you should then research which CPU will A) deliver the kind of performance that you're after and or B) fit neatly within your budget. And that's exactly what these reviews are designed to do, provide you with all of that relevant information. In fact, if you are looking for a solution that suits your budget while maximizing what you can get for your dollar, our 9800X3D review, and all of our reviews for that matter, do answer that question. For example, here's our cost per frame graph. And if you are value-conscious, I have to imagine the 9800X3D for the $480 US MSRP, it's starting to look a lot less attractive. And that's why so many people opted for the 7600X, for example, or its various variants in favor of the more expensive X3D models. In the case of the 7600X, you're getting over 70% of the performance for well under half the price. Or as this graph neatly illustrates, the 9800X3D came in at a 67% premium.
Now, circling back to the 'I just want to know how it will perform for my specific configuration' question. This is traditionally how you should go about addressing that. Though I warn you, it does require some additional steps, a bit of extra research. But again, don't worry as I am working to eliminate those steps, and that's the point of this video, and I'll get to it shortly. But what you should normally do is look at the 9800X3D data, ideally CPU-limited 9800X3D data, in the games that you're interested in. Let's go with Starfield for example. And again, we'll pull the Day One review data here. The 9800X3D was able to throughput 140 FPS at 1080p using the ultra preset, while the 5800X3D was limited to 93 FPS. So then, what the hell does it do at 4K using my RTX 4080, you might ask? And the answer is not much better than a cinematic slideshow, as we're looking at about 40 FPS. So given that, you can safely assume that the 5800X3D, which can throughput over 90 FPS in this particular scenario, isn't going to be the primary performance-limiting component. That would be the RTX 4080.
But then you do have to ask yourself, do I like 40 FPS? Am I going to play Starfield at 40 FPS? If the answer is no, which I hope so, and instead you'd like to better utilize, maybe you have a 144Hz monitor, then you will need to lower the visuals and/or enable upscaling, at which point the 5800X3D could become the primary bottleneck once again. But the point is you can correlate the performance you see in the GPU content to that of the CPU content to roughly work out where things stand. Though, I'd argue this isn't really the right way to go about making a CPU purchase anyway, but it does help to address the 'I just want to know how it would perform for my specific configuration' question. So, given all of that noise, what is my solution? Well, unfortunately, my solution is to quadruple my workload to provide you all of that information in a single video. Not because I think you necessarily need it, but because I've succumbed to the fact that some people think they need it. I'm joking, sort of. But in all seriousness, I am interested to hear what you guys think about this approach. So, let's go check it out.
First, I'm going to use the Marvel Rivals data to explain what I'm attempting to do here. The first graph should look pretty normal. It's a simple head-to-head comparing the 5800X3D and 9800X3D using the medium and ultra presets, both of which have of course been tested at 1080p using the RTX 5090. Using the medium preset, the 9800X3D smokes the 5800X3D, delivering 47% greater performance, though a less impressive 30% increase to 1% lows. Then with the more GPU-limited ultra preset, the average frame rate margin is reduced to 34%, and that margin is also maintained for the 1% lows.
Now, that data was recorded using the RTX 5090. And here's how the 5090 compares with the RTX 5080, 9070 XT, and 9060 XT using the 9800X3D. As you can see, the 5090, even at 1080p, is a good bit faster than most of the GPUs tested. But as we increase the resolution, we see frame rates come down. But simultaneously, as the results become more GPU-limited, the 5090 is able to extend its lead over the lower-end models. Then at native 4K, we're under 100 FPS using the ultra preset, even with the 5090. And now the data is heavily GPU-limited, at least when testing with the 9800X3D. But now you have CPU-limited CPU testing along with some comparable GPU data. This is obviously a lot of testing to get to this point, but it's also messy combining all of this data, and it creates just a lot of graphs to go over. So the solution that we've come up with is this: the original CPU graph but with the GPU data points overlaid while maintaining the same graph scale.
Now it's important to note that I am comparing the 1% lows for the GPU data as I think this makes the most sense for CPU-related content. And just to be clear, this means the GPU data shown here, so the GPU data points are based on the 1% lows seen on the previous graphs, or more specifically in this example, the 1080p GPU data. What this tells you is that under these conditions, the 5800X3D should only limit the performance of the RTX 5090 when using the ultra settings at 1080p, and then the 5090 and 5080 with the medium settings. Jumping up to 1440p in this specific title will only see the 5800X3D limit the performance of the 5090, and it shouldn't limit the performance of the 5080 or 9070 XT, for example. It's of course not an exact science as the GPUs were tested on the 9800X3D, but overall performance where we're not CPU-limited should be about the same with the 5800X3D. Then at native 4K, as you'd expect, neither of these CPUs will limit the performance of any gaming GPU currently on the market, or in this case the RTX 5090, which is kind of off the market, but coming back. We'll have to see how that plays out.
Now, let's look at a more CPU-intensive title such as Space Marine 2. Here, the 9800X3D delivered 45% greater performance, seen when looking at the average frame rate, or a massive 51% faster when comparing the 1% lows. If you're wondering how and why the medium and ultra results are the same, it's because we are heavily CPU-limited, even with the 9800X3D. What's crazy about this is that the 5800X3D is heavily limiting the performance of not just the RTX 5090, but also the 5080 and 9070 XT. In fact, when using the medium settings, the 5800X3D is even limiting the 9060 XT, and by some margin. Using the ultra settings though, the 5800X3D and 9060 XT are pretty much the perfect pairing. For those of you playing at 1440p, the 5800X3D will still heavily limit the performance of the 9070 XT, 5080, and 5090, regardless of the settings used. But if you're running a 9060 XT, then the GPU will be the weakest link.
Something else I will note here is that we often see comments from people claiming that we shouldn't test with extreme high-end GPUs like the RTX 5090, suggesting that the RTX 5080, for example, would result in far more realistic and real-world results. What these commenters don't seem to understand is that for most of these tests, the RTX 5090 will end up delivering the exact same results at 1440p as we are heavily CPU-limited at 1080p. And Space Marine 2 is the perfect example of this because here at 1440p, it doesn't matter if we tested using the RTX 5080 or even the 9070 XT with the medium settings, the results would have been much the same. Now, the 5800X3D might have performed a bit better with the 9070 XT due to Nvidia's CPU overhead issue, but that's a different problem and we won't complicate the issue further with that discussion here, and I have dedicated videos covering that topic if you're interested.
Even at 4K, the 5800X3D will limit the performance of the 5090 when using the ultra settings, and then the 5080 and 5090 when using the medium settings. However, if you have a 9070 XT or 9060 XT, then the 5800X3D will work just fine. Here we have another example with Assassin's Creed Black Flag Resynced. And this one really highlights the importance of 1% lows, especially for CPU testing. Looking at the average frame rate, we see that the 9800X3D is a mere 6% faster using the medium settings, or just 3% faster with the ultra high settings. But if we focus on the 1% lows, we see that the AM5 processor is 40% faster using medium, or 26% faster using ultra high. And this means at 1080p, the 5800X3D is heavily limiting the performance of the RTX 5090, 5080, and 9070 XT, at least when using the medium settings. The ultra high settings though only saw the 5800X3D limiting the performance of the RTX 5090.
Playing at native 1440p, we see the 5800X3D is really only limiting the performance of the RTX 5090 with the medium settings, while there's a reduction of a few FPS for the RTX 5080. So, at native 1440p, the 5800X3D should do just fine with most graphics cards. Then we have the 4K data, and no surprises here. If you want to game at lower frame rates, the CPU matters far less, especially when getting down to just 60 FPS, like what we see with the ultra high preset.
Now, before I wrap up the testing, I would like to just point out something that I found interesting with this new data set, which this video only showed a small portion of, but this is more of a feeling-out process to see what you all think of this new format and if it's worth going forward with. But if we go back to the 9800X3D review, we see across the 14 games tested that the 5800X3D averaged 148 FPS, making the 9800X3D 39% faster. In this updated data set, which is also based on 14 games, and I plan to use it for upcoming CPU reviews, the 5800X3D averaged 142 FPS, making the 9800X3D 33% faster. A small reduction in margin there to what was seen previously, but this newer data is using a lot more maxed-out presets, which can lead to more GPU-limited data, whereas the 9800X3D Day One review used slightly dialed-down high presets. Using the medium settings, the margin was 35%, so much closer to the original 39% margin.
But the point of all of this is back when I first reviewed the 9800X3D, I showed across 14 modern games of the time that the 5800X3D was capable of throughputting over 140 FPS using high-quality settings. And today, we found that's still the case. That means if you're typically targeting over 100 FPS in your games, the 5800X3D will deliver that level of performance, assuming your GPU is using the appropriate settings in order to achieve that level of performance. What I'm saying here is if nothing else limits the performance, the 5800X3D will typically be capable of delivering that level of performance regardless of the resolution. And this has remained true for testing that spans two years now.
So, was that useful? Are these GPU data points something you'd like to see more of going forward? I can't promise you that all CPU reviews and benchmark content will feature this data, as it does take multiple days to just gather the GPU side of things, but if it helps people, I'm happy to put in the work and do it. As long as I don't have to update the data constantly, I should be able to use it for some time. So, there is that. And it really is best to think of this data as more of a rough guide given that all the GPU results are recorded using a 9800X3D. But the idea is to show some optimal GPU data there, just as we do for the CPU side.
Now, for those of you who are going to point to the Assassin's Creed Black Flag Resynced 1440p GPU data and say something like, 'See, this is why CPU reviews are misleading,' or, you know, whatever you want to come up with. This is just one game and it does miss the entire point of the testing to begin with. And it's important to note that games featured in CPU reviews are used more as tools to measure the actual CPU gaming performance, more so than they are a guide for how the hardware will perform in those games. A CPU review isn't a system review, and a system review isn't useful as a buying guide for a CPU. Personally, I believe this topic couldn't be more straightforward. Test all the CPUs to see their true unimpeded gaming performance. Then compare that performance in some kind of value analysis, such as a cost per frame graph that tells you which CPUs are the fastest while also letting you know which are the best value. So whichever category you fall into, you can find the fastest or best value product for you.
It's also bad to point to a single game because you can have examples like Assassin's Creed Black Flag Resynced that aren't particularly CPU demanding, or then you can have games like Space Marine 2 that can be massively CPU demanding. Of course, the 5800X3D is still a very capable CPU that isn't necessarily in need of an upgrade, but if you play a lot of Space Marine 2 and you want to keep frame rates north of 100 FPS in those big horde battles, then the 9800X3D will be a significant upgrade. But if you're happy with 90 FPS or less, then a 5800X3D, it's going to work a treat. And you can learn all of that from plain old 1080p benchmarks and a melted graphics card.
Also, just as a side note, I've attempted to make a few jokes throughout this video to poke fun at the topic, and you know, they were meant to be jokes. They might not have been funny, but the intention was for them to be jokes. So, please don't be offended by this. It's a joke. And while I disagree with people who call for GPU-limited CPU testing, that's really as far as it goes. I have seen some people suggest that we despise our viewers when we get frustrated with this sort of thing. And look, that's not entirely true. We only despise some of you.
There you go. I did it again. Okay, I'm not going to try and smooth things over, at least not right now. Anyway, that is going to do it for this video. Again, I am genuinely very interested to see what you guys think about the GPU data added to the CPU testing. What portion of you find it unnecessary and it just clutters the graphs? What portion of you find it relevant and useful? So let me know in the comment section and I'll be sure to read all of that. But other than that, we're done for this video. So, if you want more Hardware goodness, we have the join button or Patreon. Signing up to either one of those things gives you access to exclusive live streams for members only, our exclusive Discord server, surprisingly for members only, behind-the-scenes content and Q&A stuff. So check that out if you're interested. If not, that is perfectly fine. And of course, I would like to thank you for watching this video. I'm your host, Steve. See you again next time.