Windows 10 vs Windows 11 2025 performance benchmarks shows holdouts they should upgrade

Microsoft is ending support for Windows 10 very soon, this October. Thus, users still on the decade-old OS may be wondering what kind of performance they can expect to gain, or lose, if they move from Windows 10 to Windows 11.

We last took a look at Windows 11 vs 10 performance towards the end of 2023 as we first compared an in-place upgrade, which was followed up by a clean installation test later on Steve's 14900K-RTX 4070 Ti PC. There, Windows 11 23H2 did not show a major difference or gap in performance vs Windows 10 22H2; yes there certainly were tests where we saw some substantial split between them, but for the most part we concluded that you could run either and be completely happy with your setup.

This time around, we have changed how things were tested. Last time, it was more of how the experience would be between the two OS when things were kept at default settings without any tweaking.

However, Steve and I later agreed that game testing in such default scenarios would represent GPU-bound situations much more often than not and hence our testing method was later updated to address this.

Thus, this time we have our graphics settings lowered and processor-intensive settings upped. If you want to check out how this impacts the CPU check out my dedicated article on the topic.

This methodology update only applies to games as synthetic benchmarks and in-built apps' benchmarks were already designed to test in such a way. We have divided our data into two parts:

• Gaming
• Productivity

We benchmarked in the following system and my colleague and friend Steven Parker contributed to the testing:

Test system

• Computer ATX Open Motherboard Rack
• AMD Ryzen 9 9950X3D
• Gigabyte X870E AORUS MASTER (BIOS N10)
• be quiet! Pure Power 12 M 850W
• Thermalright Phantom Spirit 120 CPU Air Cooler
• Thermal Grizzly Carbonaut (38x38x0,2mm)
• 2x 16GB G.SKILL Trident Z5 Neo RGB Series DDR5 6000MT/s CL28-36-36-96
• Samsung 990 Pro 1TB (PCIe Gen4)
• NVIDIA GeForce RTX 5090 Founders Edition
• Windows 11 24H2 Pro
• Windows 11 22H2 Pro

Gaming benchmarks

We start with UL Benchmark's 3DMark Fire Strike Physics test. This is a DirectX 11 benchmark and that means it does not utilize too many processor threads. Even then, at the default preset and 720p custom settings we see the CPU on Windows 10 being ahead of 11 by 4.3% and 6.2% respectively. The 720p test shows a wider split as it is even harder on the processor compared to the default setting test.

Windows 11 also picks up a win in the 720p graphics test, and this was surprising as it is counterintuitive given that the CPU test was won by Windows 10. It indicates that Windows 11 actually manages overheads better.

Up next, in Time Spy Extreme it was a landslide victory for Windows 10 as it was up to 29% better than Windows 11. This result was again very surprising for us considering that is Windows 11 that is supposed to be more optimized for the correct cross-core talk on dual-CCD Ryzen chips like the 9950X3D.

One thing to note is that the Time Spy Extreme (TSE) CPU benchmark is said to be "three times more demanding than the Time Spy CPU Test" and hence, consequently, it is even more demanding than the Fire Strike one above. This extra load on TSE definitely appears to expose Windows 11's worse capability as a gaming OS from the CPU side of things.

3DMark CPU Profile benchmark is next and like Time Spy Extreme, this is also a DirectX 12-based test and that measures the scaling behavior of processor cores and threads. It can be seen that CPU Profile does not show as big a difference as Time Spy Extreme did.

If we were to guess, it is likely because the TSE test presented a continuous heavy load, whereas the 3DMark CPU profile one starts and stops as it stresses the processor threads one by one via different test passes starting from one thread up to max threads. So perhaps Windows 11 was throttling faster when presented with a continuous load.

Next, we move to Steel Nomad and Steel Nomad Lite tests, which are more modern DirectX 12-based rasterization tests.

Once again, Windows 10 shows two clear wins on our custom 720p tests that were done to simulate CPU-heavy tasks as Steel Nomad (Lite) do not currently have a standalone CPU/physics benchmark. This indicates again that 11 is still not quite up there.

We wrap up our 3DMark data with custom 720p Speed Way where Windows 11 manages to eek out a small win over 10. Speed Way is a DirectX 12 Ultimate test that measures ray tracing (RT) performance and tracing all those rays of light is quite taxing on the processor, although perhaps not as much as it is on the graphics card.

Up next we have some real games. The custom graphics settings we chose were designed to maximize the CPU load and minimize the same for the GPU. (For reference, RT = Ray Tracing, PT = Path Tracing, FG = Frame Generation, RR = Ray Reconstruction, VRS = variable rate shading)

We hit it off with 2016's Assassins' Creed Odyssey - based on AnvilNext 2.0 - where Windows 11 secures an absolutely massive win. If we were to guess we would say it was Ryzen's X3D cross CCD cores not playing well with the title on Windows 10, otherwise we really do not see a reason for such a humongous split. In the average framerate metric, Windows 11 was 55% ahead.

Shadow of the Tomb Raider from 2018 is up next. Based on the Foundation Engine, the Ryzen 9950X3D sees almost no difference in output between the two OS despite being quite CPU heavy.

Cyberpunk 2077 from 2020 running on the RED engine is next. At our custom settings, we saw that the Ryzen 9 9950X3D was a very substantial 19% better on the Windows 11 side in average framerates; in the minimums the gap was slightly lower at 17%.

We have a couple of games from 2021, starting with Hitman 3, which is based on the Glacier 2 engine. Windows 10 gets a small win here as it was nearly 9% better. This is unlikely to be noticeable but still it counts given that it is statistically significant.

Far Cry 6 is the other 2021 title. Running on the Dunia engine, it presents another very interesting case of a huge performance split between the two OS. Far Cry games are known to be single-thread heavy. While the 9950X3D was 22% better at average FPS on Windows 11 vs 10, the gap increased to 26% for the minimums.

F1 24, based on EGO 4.0 engine, was again a small but significant victory for Windows 11 as it was close to 14% better. Like Hitman 3, this is not something users will notice without a frame-rate counter atop, but it's still statistically significant.

Monster Hunter Wilds is the latest game in our test suite and it runs on RE Engine, and again we come across yet another baffling result. So far we have seen Windows 11 either registering massive wins or smaller victories. However, this time things have completely flipped as we noted 33% better performance on Windows 10 vs 11 in the title.

Productivity

With our gaming tests done, we move on to productivity.

Starting with Cinebench 2024, Windows 10 gets an 8.5% lead over Windows 11 here. Since this is a rendering test and utilizes all available threads, the dual CCD architecture of the 9950X3D was not of much concern here.

UL Procyon's Office suite productivity test provided a bit of mixed bag of results. While Windows 11 was the winner on Outlook and PowerPoint, Windows 10 won in Excel, and it was a tie in Word.

The results are certainly interesting. It appears that Windows 10 excels in number crunching scenarios while Windows 11 is better at presentation and personal information management-based tasks. Perhaps underlying graphics processing and networking upgrades may be helping? We can not say for sure as that is out of our scope.

AI is one area where Windows 11, in particular, version 24H2, seems to have made the biggest progress. We observed a similar thing in case of graphics performance as well when we compared Windows 1124H2 with 23H2 (both on in-place upgrade and clean install).

We saw the biggest divide in the case of the Phi 3.5 model with a 46% lead in favor of Windows 11, and this was followed closely with a 43% gap in the case of Llama 3.1.

Finally, we benchmarked Computer Vision also in the Procyon suite. This tests inferencing and here, the scores were within a percentage of each other, so very much unlike text generation.

Conclusion

We mentioned up top that we updated the way we tested as we thought our previous methodology, although realistic, did not quite capture the full picture and the extent to which moving between OS could impact the processor's performance.

However, I will be honest, going in, I did not expect to see such massive differences between them. Barring a few exceptions, we saw substantial splits almost everywhere and this was especially pronounced in certain games (engines). And productivity workloads also saw major differences too with Windows 11 24H2 shining bright in the case of AI text generation.

So there is little to no subtlety this time with both OS swinging hard and sharing clear victories, with Windows 11 being ahead more often. This also means that there is less reason to hold on to Windows 10, at least from the performance point of view; hence, I feel it will be a pretty good idea to upgrade especially with support for Windows 10 also ending.

While Windows 10 did enjoy some big wins, over time these victories will likely fade out as Windows 11-specific optimizations start taking over entirely.

As an Amazon Associate we earn from qualifying purchases.

Source

Hope you enjoyed this news post.

Posted Monday 11 August 2025 at 6:33 pm AEST (my time).

News posts... 2023: 5,800+ | 2024: 5,700+ | 2025 (till end of July): 3,458

RIP Matrix | Farewell my friend