Microsoft this week lifted another upgrade block for Windows 11 24H2. Thus if you are looking to do an in-place upgrade from 23H2, you can check out our review to get an idea of what you could expect. We shared those results for Windows 11 24H2 performance last month.
While many tests showed margin of error differences, there were two interesting takeaways from that in terms of gaming. First the positive, there was a definite improvement in frame generation performance on Black Myth Wukong. We observed it on all three of our 24H2 runs and the gains were quite substantial too with over 17% improvement in averages and more than 15% in the lows.
On the flip side, single-threaded performance on 24H2 was, for some reason, lower and it was highlighted in more than one instance, be it a synthetic test or an actual game.
Besides gaming, application start-up performance also took a sizable hit on the newer feature update in PCMark 10.
With those in mind, we set out to see if doing a clean installation would reduce these differences, and also whether we would come across other new performance quirks.
Thus we proceeded to clean install Windows 11 November 2024 Patch Tuesday (KB5046617) on top of October 2024.
Just like last time, our test is not straight-up apples-to-apples comparisons. We are trying to replicate the usage and experience of a general user and as such, all settings would be kept at default. This is how we are evaluating the performance differentials of a clean install for Windows 11 24H2 vs an in-place upgrade. Hence, some of the settings like Core Isolation or Virtualization-based Security (VBS) have been kept enabled.
In case you are wondering about the hardware we tested this on, Steven and I worked on this together (remotely), on his test bench comprising Intel's Core i7-14700K CPU and AMD's 7800 XT GPU (provided by AMD for review). The latest GPU drivers were used at the time (Adrenalin version 24.10.1).
Again we start with synthetic benchmarks first.
In 3DMark's CPU/physics test, we see an immediate improvement with the clean installation. Our 14700K does better in both DirectX 11-based Fire Strike physics as well as DirectX 12-based Time Spy. This time, 24H2 draws nearly equal to 23H2 on Fire Strike Ultra and manages to beat it on Time Spy Extreme.
So an 8.2% gain in the more single-threaded Fire Strike Ultra and a 5.87% improvement in the more modern Time Spy Extreme.
The same trend continued on 3DMark CPU Profile which measures the scaling performance of a processor across threads on Time Spy. On our clean-installed 24H2 PC, we saw better scaling once the processor thread count exceeded eight. At up to 4 threads, the two systems were identical.
The biggest difference was seen at 16 threads where the clean installed 14700K was ~6.78% faster. With all threads maxed, the difference was significant too at ~6.6%.
Following that, we have our 3DMark GPU tests. The 7800 XT for some reason did worse on our clean-installed 24H2 system but it is something that may be considered within the margin of error (max. difference is ~2.23%). So we aren't going to sweat it out and are attributing it to run-to-run variance.
Hence, in the synthetic tests, the clean PC definitely fared significantly better. Neither 3DMark nor Windows received any specific optimizations in this period and thus the performance differential is likely to be down to the clean install.
With that, it's time to move on to some real-world games.
On Black Myth: Wukong, we again saw a huge uplift with frame generation (FG) enabled on 24H2. In fact, our results on the clean-installed system were identical to what we saw on the in-place upgrade.
Only the 95%ile score was better by 1 FPS in the case of FG off, the average was the same indicating that a clean PC or an upgraded PC hardly matters in this case.
We had another eventful result on Marvel's Guardians of the Galaxy as the 24H2 clean installed PC underperformed our 23H2 by a large degree.
Interestingly, Intel's APO (Application Optimization) actually helped this time as it smoothed out whatever issue it may have been that was bothering the 14700K as we once again saw performance parity between 24H2 and 23H2.
Shadow of the Tomb Raider's performance differences were within the margin of error mostly. Another noteworthy point here is that APO did not crash on 24H2 this time. We were able to evaluate its effect, and it most certainly helped on both 23H2 and 24H2.
The Callisto Protocol was a boring test run, there is not much going on here.
I was eager to see what would happen in Far Cry 6 since it presented an interesting case last time. Unfortunately, for some reason, our clean-installed system did not default to the same graphics settings that it had chosen during the in-place upgrade. Thus, we could not directly compare the results. Regardless, it was still interesting to note this difference between the two cases.
We noticed a similar thing in a couple of other games too, Assassins' Creed Odyssey and Final Fantasy XV. We are unsure why our 24H2 clean setup felt our RX 7800 XT was not capable of handling 1080p ultra settings such that it defaulted to a lower preset in these three titles.
Regardless, we could confirm that our 14700K was not bottlenecked even at these more CPU-demanding settings since we saw an increase in the FPS output in all the three games.
With the gaming portion wrapped up, we move to productivity. Cinebench 2024 shows the exact same performance on 24H2 (clean) and 23H2 in the CPU rendering test, and a 10-point increase over in-place upgraded 24H2.
The GPU test also shows an improvement on the clean installed PC but again nothing too exciting.
It was a bit disappointing to see that the clean installed 24H2 PC was slower than the in-place upgrade system on 7-Zip's compression test. While it was still slightly faster than 23H2, compared to the in-place upgrade, the clean 24H2 setup was ~3.5% slower.
PCMark 10 again revealed interesting results. If you recall, last time, on an in-place upgrade, we noted that the app startup performance was lower in the case of 24H2. Well, that trend continued and in fact, it got worse.
Windows 11 23H2 scored 24,096 in the application start metric. In contrast, 24H2 upgrade and 24H2 clean scored 21,376 and 20,813 respectively. Thus they were ~11.3% and ~13.625% slower respectively.
24H2 clean installed device was also much worse in photo editing, spreadsheet processing and writing tests. Overall, the 24H2 clean PC scored 9,999 points vs 10,734 points on 23H2 and 10,460 on 24H2 in-place upgrade.
Lastly, we have the memory allocation test on 24H2 vs 23H2. This shows how much system RAM each OS requests for caching page files, prefetch data, game/app data, among other things.
Speaking of RAM, we were using TeamGroup's 32 GB DDR5-7600 kit that we reviewed last year. Memory allocation is the amount of system RAM that an app requests and is not equal to the actual memory used up by that software. Regardless, it gives us an idea about the capacity of memory needed.
This is a major win for our clean installed 24H2 system. Not only does it do quite well against 23H2, it also massively reduces the amount of allocated RAM compared to what we had previously on our upgraded 24H2 PC. The improvement was seen in every single workload.
Perhaps there is a memory leak problem on an in-place upgrade that our clean-installed PC resolved. We can't say for sure, but our results consistently suggested this.
Conclusion
The idea of this clean installation review was to see if the performance nuances we saw on the in-place upgrade would be eliminated on a clean 24H2 PC and whether the latter would be the better way to upgrade from 23H2.
While the initial impressions were great as we saw significant improvements in the synthetic 3DMark benchmarks, they did not quite materialize in actual games. Neither did we see any notable improvements in non-gaming tasks. And in the case of PCMark, we saw a significant regression in performance.
Thus if you feel like trying a clean installation of Windows 11 24H2 in hopes of improving the performance, you will likely be disappointed; unless you are getting memory hogged like we experienced on our upgraded 24H2 setup. In that case, a move to clean installed 24H2 could help.
Our stance on upgrading from Windows 11 23H2 to 24H2 remains the same, it depends, but in most cases, you are not going to miss out on much if you stick to 23H2.
11 Comments - Add comment