Today we're taking an updated look at Nvidia's long-known CPU overhead issue by comparing the GeForce RTX 5070 Ti and Radeon RX 9070 XT across a range of games using two different CPUs. The goal is to work out if Nvidia's latest Blackwell architecture still suffers from the overhead issue, and if so, to what degree.
But before we get into it, today's sponsor spot is brought to you by Thermal Grizzly and their Wire View Pro 2, the best way to safeguard your 12-volt high-powered connector from melting. This little device plugs in between your graphics card's power connector and the power supply's cable to measure current, voltage, temperatures, and the power of each individual pin. It has a built-in display with configurable views for per-pin current, total power, and temperature, plus audible and visual alarms if thresholds are exceeded. The CNC-machined aluminium body includes a semi-passive fan for cooling, automatic data logging, and optional automatic system shutdown via the included splitter cable for extra safety. Thermal Grizzly are so confident that the Wire View Pro 2 will keep your graphics card safe that they're backing the product with an extended warranty covering the GPU connector for two years. That is a crazy level of support. So, for more information, please check the link in the video description.
Okay, so the issue that we're looking into here is Nvidia's driver overhead, which requires more CPU processing power when compared to Radeon to translate game engine instructions into commands for your GPU. Now, while all drivers do require CPU time, Nvidia's architecture demands considerably more. Before your graphics card can render a single pixel on your monitor, your CPU has to prepare the scene. The CPU calculates the game logic. It determines player inputs, physics, AI movement, and collision detection. Once it calculates where everything should be, it then sends a command called a draw call to the GPU. Because this pipeline happens sequentially, the CPU must repeat its workload for every frame rendered. And this means at 60fps, your CPU has to calculate game logic and issue draw calls 60 times every second. So, at 240 FPS, it must do this 240 times per second, meaning the CPU is now working four times harder.
Generally speaking, it is at these higher frame rates where the GeForce GPUs can overwhelm your CPU, especially if you're not using the latest and greatest CPU. We've seen examples where much slower Radeon GPUs can outperform what should be a faster GeForce GPU in CPU-limited scenarios. And this is because the Radeon GPU uses less of your CPU, and that means it frees the CPU up to process more frames, resulting in better performance. This can make Radeon GPUs a better choice for those of you with lower-end PCs where system resources are at a premium, but for the most part, for those of you with high-end systems, it's a non-issue. At least without an RTX 5090-like competitor from AMD. As I said, for today's testing, I'll be comparing the GeForce RTX 5070 Ti and Radeon RX 9070 XT, and I'll be doing this with the Ryzen 7 9800X3D and Ryzen 5 5600. So, two very different CPU configurations, which is the idea, as we want to see how these GPUs behave when CPU limited. Okay, let's get into the results.
Starting with Pragmata, using the quality preset plus ray tracing, we see that when using the 9800X3D at 1080p, the 9070 XT was 10% faster than the 5070 Ti, averaging 201 FPS. Switching to the Ryzen 5600 reduced performance by around 6%, but even so, the 9070 XT was still 7% faster than the GeForce GPU. However, that's when looking at the average frame rate. If we focus on the 1% lows, we see that with the slower CPU, the Radeon GPU's performance was reduced by just 5%, while the 5070 Ti suffered a massive 23% performance loss, as the added CPU overhead hurts 1% low performance with the slower CPU. All of that said, increasing the resolution to 1440p masks this overhead issue, as now the results are heavily GPU limited, and as a result, we see very little difference between the 9800X3D and 5600 configurations.
Now, using the minimum quality preset with ray tracing disabled, we find some pretty interesting results using the 9800X3D at 1080p. The 1% low performance of the 5070 Ti and 9070 XT configurations is much the same, though the Radeon GPU produced a 12% greater average frame rate. But when we installed the slower Ryzen 5 5600 processor, the margin grew to 16% in favor of the Radeon GPU, and more crucially, it was also 20% faster for the 1% lows. However, once again, switching to the 1440p resolution negates the CPU overhead issue as the data becomes heavily GPU limited.
Next, we have Fortnite, and this Epic quality data isn't particularly revealing. Interestingly, the 9070 XT performs better when paired with the 9800X3D, but when switching over to the Ryzen 5 5600, it slips behind the 5070 Ti by a small margin. Then at 1440p, we're looking at pretty similar results between the two GPUs, though the 9070 XT does produce slightly stronger 1% lows. However, if we test again, but this time with the performance mode enabled, we see Nvidia's overhead issue in action. Because this data is heavily CPU limited, it doesn't matter which resolution we look at, 1080p or 1440p, it's all much the same. The 9800X3D, for example, was limited to the same level of performance at both resolutions, and here the 5070 Ti and 9070 XT were able to max out the CPU to a similar level. So, no signs of CPU overhead issues here with the 9800X3D. However, the Ryzen 5 5600 does struggle more when paired with the 5070 Ti, allowing the Radeon GPU to deliver 12% greater performance. Not a lot, admittedly, especially when rendering well over 300 FPS, but this is a clear indication of the Nvidia overhead issue.
Moving on to 007 First Light, tested with the ultra preset, we saw that the 9070 XT easily outpaced the RTX 5070 Ti, delivering 14% greater performance with the 9800X3D, and that margin was extended to 22% with the 5600. You'd probably expect that margin to shrink as the results become more CPU limited with the Ryzen 5 processor, but due to Nvidia's CPU overhead, the margins grow, and we even see this margin maintained at 1440p. In fact, here it's even more noticeable. Using the 9800X3D, the 9070 XT was 11% faster. So, with a slower CPU, you'd naturally expect that margin to shrink as the data becomes more CPU limited than GPU limited, but the opposite happens. Now, the 9070 XT is 24% faster when comparing the average frame rate and 28% faster for the 1% lows.
Now, for the low preset data, and in the past, I've noted that testing with low settings can be a poor way to measure CPU performance in games, assuming it's not a competitive shooter that will often be played using low settings, that is. In this example, the low preset reduces CPU load by lowering NPC density, as one example. And as a result, the 5070 Ti fares much better when paired with the Ryzen 5 5600 processor, as the overhead issue is less problematic when reducing overall CPU load, which is what the low preset does.
Next up, we have Marvel Rivals. Starting with the ultra quality data, and for testing this game, I am using the built-in benchmark. The first thing to note is that when using the 9800X3D, the 9070 XT and 5070 Ti are very competitive. The Radeon GPU is just 3% faster here. However, when using a much slower CPU, this changes dramatically, as now the 9070 XT is 17% faster when comparing the average frame rate and 21% faster for the 1% lows. Basically, the 9070 XT drops very little performance when paired with the 5600, while the 5070 Ti becomes a little over 20% slower due to its higher demand on the CPU. All of that said, Nvidia's CPU overhead issues are largely hidden at the more GPU-limited 1440p resolution.
Now, when switching over to the low preset, we see more of the same really. Using the 9800X3D, things look pretty normal. In fact, the 5070 Ti is up to 3% faster, seen when looking at the 1% lows. However, the tables very much turn when switching over to the slower Ryzen 5 processor. Now, the 9070 XT is 22% faster when comparing the average frame rate and 15% faster for the 1% lows. We're even seeing the effects of Nvidia's CPU overhead issue at 1440p. Testing with the 9800X3D, for example, revealed the 5070 Ti to be up to 7% faster. But when we switch over to the slower Ryzen 5 processor, it's now the Radeon GPU that's faster, up to 22% faster, in fact.
Now, Space Marine 2 is a very CPU-intensive title, so I'm not surprised to see Nvidia's CPU overhead rear its ugly head here. And as a result, we have a mixed bag of results, which look a bit wrong at first glance. For example, at 1080p, the 9070 XT is 4% faster than the 5070 Ti when paired with the 9800X3D, seen when we're looking at the 1% lows, and then 22% faster when using the slower Ryzen 5 processor. What seems odd at first, though, is the fact that at 1440p using the 9800X3D, the 5070 Ti ends up being 22% faster, again seen when looking at the 1% lows. Now, the reason we're seeing this is because the 1440p data is more GPU limited, especially in the case of the 9070 XT. That flips once again, though, when we test with the Ryzen 5 5600. Now the data is more CPU limited, and as a result, the 9070 XT ends up boosting 1% lows by 20%.
Retesting with the low preset reveals similar data to what we see when testing with Fortnite, that being very similar results at 1080p and 1440p as the data is heavily CPU limited. Not only that, but whereas the 9070 XT was just 5% faster when paired with the 9800X3D, that margin increased to around 15% when using the slower Ryzen 5 5600 processor. Not a massive margin, but it is another clear indication that GeForce GPUs require more of your CPU.
Next up, we have Assassin's Creed Black Flag Resynced, using the high preset with RT extended enabled. And at 1080p paired with the 9800X3D, the 9070 XT was 12% faster on average while delivering 19% higher 1% lows when compared to the 5070 Ti. However, dropping down to the Ryzen 5 5600 saw the 9070 XT's performance drop noticeably as this data is CPU limited, equalizing average frame rates at 94 FPS, while the 5070 Ti held virtually identical figures across both CPUs. Despite being CPU limited, we're not seeing Nvidia's overhead issue in this title. Then at 1440p, performance became heavily GPU-bound with the 9800X3D. The 9070 XT held a slight 4% advantage. But when we switch over to the 5600, the 5070 Ti edged ahead by 4% on average with slightly better 1% lows, again showing no signs of the GeForce CPU overhead issue holding back performance in this title.
Switching over to the ultra low preset, the 9070 XT maintained an 8% lead in average frame rate and a 14% uplift in 1% lows at 1080p when paired with the 9800X3D. Using the slower Ryzen 5600, the margins were somewhat compressed for the averages with the Radeon GPU just 4% faster, though it still produced 13% stronger 1% lows, which is evidence of a GeForce overhead issue. At 1440p, both GPUs performed similarly on average with the 9800X3D. The 9070 XT was within a couple of frames on average while delivering 11% better 1% lows. Then on the 5600, average frame rates were essentially identical at 121 and 122 FPS, but the 9070 XT still provided 12% stronger 1% lows.
Moving on to Battlefield 6. Tested using the badass preset at 1080p paired with the 9800X3D, the 5070 Ti held a slight 5% advantage over the 9070 XT, averaging 145 FPS. However, switching over to the slower Ryzen 5 5600 processor saw the 5070 Ti lose some ground, and this allowed the 9070 XT to nudge ahead by 3% for both the average frame rate and 1% lows. Still, overall performance was much the same between the two CPUs in this example. Then at 1440p, the results between both GPUs were again virtually identical when paired with the 9800X3D, with both cards delivering 110 to 111 FPS on average. Interestingly, dropping down to the Ryzen 5600 didn't shift things in AMD's favor here. Instead, the 5070 Ti retains its performance advantage, ending up 7% faster on average.
However, switching over to the low preset reveals a massive change in how both configurations behaved. When using the 9800X3D at 1080p, the 5070 Ti comfortably outperformed the 9070 XT, delivering a 20% higher average frame rate and 19% greater 1% lows. But when paired with the slower Ryzen 5 5600, Nvidia's driver overhead took a heavy toll, slashing the 5070 Ti's average frame rate by over 50% down to 124 FPS. This allowed the 9070 XT to completely reverse the standings, coming in 22% faster on average and 19% faster for the 1% lows. And we saw a near-identical pattern play out at 1440p. With the 9800X3D, the 5070 Ti maintained a solid 15% lead over the Radeon GPU on average. However, as soon as the slower 5600 was introduced, the 5070 Ti suffered another massive CPU bottleneck, and this allowed the 9070 XT to pull ahead by 33% on average while boosting 1% lows by 26%.
Next up is Horizon Zero Dawn Remastered, tested with the very high preset. And again, we'll start at 1080p with the 9800X3D. And here the 9070 XT was 7% faster on average when compared to the 5070 Ti, while also delivering 10% stronger 1% lows. Swapping over to the Ryzen 5 5600 caused both configurations to take a notable performance hit, but the Radeon GPU extended its lead ever so slightly, ending up 11% faster for both the average frame rate and 1% lows. Moving up to 1440p with the 9800X3D saw the 9070 XT outperform the 5070 Ti by 11% again for both the average frame rate and 1% lows. But when we tested with the slower Ryzen 5 5600, that margin remained virtually identical, with the Radeon GPU once again finishing 10 to 12% ahead across the board.
Switching over to the very low quality preset at 1080p paired with the 9800X3D, the 9070 XT held a modest 4% advantage for the average frame rate, though its 1% lows were 9% higher. Then when dropping down to the Ryzen 5600, the margin stayed fairly consistent with the 9070 XT delivering 7% higher average frame rates and 4% better 1% lows, showing only minor hints of CPU overhead impacting the GeForce GPU. Here at 1440p, the 9070 XT pulled ahead more noticeably with the 9800X3D, beating the 5070 Ti by 11% on average with 13% stronger 1% lows. Then switching to the Ryzen 5 5600, it narrowed the gap slightly, leaving the Radeon GPU 7% ahead for the average frame rate and 1% lows.
Next up, we have Spider-Man 2, tested with the very high preset and ultimate ray tracing enabled. Starting at 1080p with the 9800X3D, the heavy ray tracing workload clearly favors the GeForce GPU, and this saw the 5070 Ti outpace the 9070 XT by 11% on average while delivering 23% higher 1% lows. However, when dropping down to the Ryzen 5 5600, performance became heavily CPU bottlenecked for both configurations, completely wiping out Nvidia's advantage, as both cards tied at around 60 to 61 FPS on average while matching 1% lows at 50 FPS. But more importantly, what we're looking at here is a typical CPU bottleneck where performance ends up neutralized at similar frame rates, meaning Nvidia's CPU overhead doesn't appear to be an issue here.
At 1440p using the 9800X3D, the 5070 Ti was once again faster, leading by 8% on average and 15% for the 1% lows. But when paired with the slower Ryzen 5 5600, CPU limitations flattened the field once more, resulting in an exact tie across both GPUs at 57 FPS on average and 47 FPS for the 1% lows. Switching over to the very low quality preset at 1080p using the 9800X3D, both GPUs were virtually tied, each delivering right around 250 FPS on average with essentially identical 1% lows. When switching to the slower Ryzen 5 5600, the 9070 XT managed to pull ahead by 8% on average and produced slightly better 1% lows, showing some signs of the 5070 Ti giving up performance to a CPU overhead issue.
At 1440p with the 9800X3D, performance was once again dead even, with both GPUs delivering 205 to 208 FPS on average and near-identical 1% lows. Dropping down to the Ryzen 5 5600 saw performance become CPU limited to identical levels at 1080p, allowing the 9070 XT to maintain a small 3% edge on average.
Moving on to Cities Skylines 2, we'll start with the high preset data at 1080p using the 9800X3D. The 9070 XT was 6% faster than the 5070 Ti, though slightly slower when looking at the 1% lows, but with just 1 FPS in it, let's just call it identical. So, not much in it with the 9800X3D. And when we switch to the 5600, we again end up with identical performance. So, despite being heavily CPU limited, Nvidia's CPU overhead issue, well, it's not an issue in this game. We also see the same thing at 1440p here. The 9070 XT and 5070 Ti are evenly matched, and regardless of whether you use the 9800X3D or the old 5600, the results are heavily GPU limited to just over 50 FPS.
Now, this data is very unexpected. It appears as though the 9070 XT is quite a bit faster than the 5070 Ti when using the very low preset, but not because of a CPU issue. Although the Radeon GPU was up to 40% faster when paired with the 9800X3D, we actually saw identical results when using the slower Ryzen 5 5600. Here the data is heavily CPU limited, and Nvidia's CPU overhead issue isn't a problem again for this title.
So, having now gone over all of that data, how serious is the Nvidia CPU overhead issue for GeForce 50 series GPUs? Overall, I'd say not very. While still an issue, it is far from a dealbreaker or something that GeForce owners need to be particularly concerned with. As I found with my original investigation into this many years ago now, the overhead problem is really only a concern for those of you using older CPUs who are upgrading to a new GPU while hoping to get a bit more out of your existing platform. Even then, though, in most examples, the GeForce GPU will work just fine, as we saw with the Ryzen 5600. For example, we didn't see the CPU overhead issue rear its ugly head in Cities Skylines 2, Spider-Man 2, and Assassin's Creed Black Flag Resynced. Moreover, there was only a very small performance loss in Pragmata, Horizon Zero Dawn Remastered, and 007 First Light.
We saw mixed results in Space Marine 2, but generally slower CPUs will see better performance when paired with the Radeon GPU, especially when looking at the more demanding sections of the game, like the Horde Battle that we test, for example. And we did see at 1440p using the ultra settings that the 5070 Ti was 22% faster than the 9070 XT when comparing 1% lows with the 9800X3D. But switching over to the Ryzen 5600 flipped those results, and now the Radeon GPU was 20% faster, going from just 59 FPS to a far more impressive 71 FPS.
The overhead issue was also spotted in Fortnite using the performance mode, though both configurations did push well over 300 FPS, so it's less of an issue here unless you have a very, very slow CPU. Also, if you are a competitive Fortnite player, you'd likely see a bigger difference for stacked end games in tournaments, but I can't really test that. Nvidia's overhead problem could also be observed in Marvel Rivals, even with the ultra quality settings, where the 9070 XT went from improving 1% lows over the 5070 Ti by just 1% with the 9800X3D to an incredible 21% when paired with the slower Ryzen 5 processor.
But the most eye-opening results were probably those of Battlefield 6. The badass preset did show a switcheroo for the 1080p data, but it wasn't that dramatic, not when compared to what we saw with the low quality preset. Here we saw the 5070 Ti lead the 9070 XT by a massive 20% margin at 1080p when using the 9800X3D, only to have that margin completely reversed with the 5600. As we found when using the slower CPU, the 9070 XT was 22% faster, and interestingly, that margin actually increased to 33% at 1440p.
As a side note, the overall performance margins based on the 11 games used here for testing roughly lines up with our most recent 52-game average, where the 5070 Ti was just 1.5% faster at 1440p. This data shows no performance difference between the two, at least when paired with the 9800X3D. The margin did favor the Radeon GPU when using the 5600, though overall just a 4% advantage. The low quality settings did extend that margin to 9%, up from 4% with the 9800X3D.
This is why pairing a GeForce GPU with an older CPU could be a poor choice, depending on the games you play and how you want to play them. If the goal of your next GPU upgrade is to render, say, 150 FPS or more in Battlefield 6, you'll likely end up sacrificing performance with a GeForce GPU. And as we just saw, it could be as much as 20 to 30%, simply because the GeForce GPU is maxing out your CPU sooner than a Radeon GPU would.
And this will really only start to become an issue for high-end PC builds once AMD has a truly high-end Radeon GPU, so something for this generation that would rival an RTX 5090. Though the 5090 hasn't been a gaming GPU for some time now. But the point is, at present, that is the kind of GPU performance level you would need to be in trouble with current-generation flagship CPUs.
So in short, six years ago now, I found the GeForce 30 series had a serious CPU overhead issue, and the problem has persisted to this very day with the GeForce 40 and 50 series. Back in 2021, I was testing with much slower processors in the Ryzen 5 1600X and 2600X. But as predicted, games have continued to become more and more CPU demanding, to the point where we can easily demonstrate this issue using a Zen 3 processor. But even so, as I've said, this really is only a problem for those of you using older processors like Zen 3 and older. But if you have a more modern processor, the overhead issue is unlikely to be a problem for you, and therefore not a dealbreaker for buying a GeForce GPU.
Anyway, we have all the updated information, so you can go ahead and make informed buying decisions. It's not that I suspect many of you would be buying a GeForce GPU now anyway, because pricing for anything with more than 8 GB of VRAM is, well, it's disgusting. So that's that. I guess that's, we'll end the video there. Not much more to say really. If you enjoyed it, give it a like. Subscribe for more content. We also have the join button or Patreon if you want to support the channel more directly and get more involved, if you would like to, by joining the Discord server and chatting with us there and maybe giving video suggestions and whatnot. We also have the Q&A series where you can ask a question on Discord and get priority, behind-the-scenes content, and I don't know if there's other stuff, I forget now. There's things. Check it out if you're interested. If not, that's perfectly fine. And I would like to thank you for watching this video. I'm your host Steve. See you again next time.