Windows 11 Performance Comparison: 23H2 vs 24H2 vs 25H2 — What’s Really Improved?
Table of Contents
Microsoft is preparing Windows 11 26H2 as its next annual feature update, and it is already available for testing through the Windows Insider Program. For every new OS release, the critical question for the EUC community remains the same: what is the performance impact for virtual desktop scenarios? Rolling a new release out to production without understanding its effect on resource consumption, user density, and application responsiveness is a gamble that could negatively affect both the user experience and the scalability of your environment.
This research takes an early look at the Windows 11 26H2 preview and compares it directly against the current 25H2 release, to judge ahead of general availability what the overall difference between the two versions is.
What is new in Windows 11 26H2?
Windows 11 26H2 is the second major annual update in Microsoft’s standard Windows 11 update cadence. Because 26H2 shares the same servicing branch as 24H2 and 25H2, the update is delivered as a small enablement package (eKB) that activates features already present but dormant on the device. This means devices running 24H2 or 25H2 can move to 26H2 with a single restart, while the build number changes to 26300. For image management and VDI lifecycle stability, this is a welcome approach, as it avoids a full OS replacement.
At the time of writing, 26H2 is available to Windows Insiders in the Experimental channel. Highlights of the preview cycle so far include the Low Latency Profile, which boosts CPU frequency during UI interactions to improve responsiveness, along with File Explorer refinements and various reliability fixes. More information can be found on the Windows IT Pro Blog.
It is important to emphasize that 26H2 is still a preview release. The results in this research reflect a point-in-time snapshot of the following builds:
- Windows 11 25H2 (OS Build 26200.8655)
- Windows 11 26H2 preview (OS Build 26300.8772)
As Microsoft continues to iterate on the preview, the final general availability release may behave differently.
Setup and scenarios
To test the hypothesis, Login Enterprise is used to generate a consistent user load, running the standard Knowledge Worker 2022 workload. This workload simulates a typical office user working with applications such as Microsoft Office, Edge, and Adobe Reader, making it representative of the average knowledge worker profile.
The following two scenarios are part of this research:
- Windows 11 25H2
- Windows 11 26H2 preview
Both scenarios run as a single virtual machine on a KVM hypervisor with the guest drivers and guest agent installed. Each virtual machine is configured with 8 vCPUs, 16 GB of memory, and a 64 GB disk hosted on NVMe storage.
To ensure consistent and reliable results, Microsoft Defender is completely disabled so it does not influence the benchmark results. Please note that this is not considered a best practice in a production environment, but it is a common approach in benchmarking scenarios where the goal is to isolate and measure the impact of a specific component or configuration change. Beyond that, no additional services are running, and the only other change to the images is the installation of PowerToys, with Awake applied to keep the machine always on and awake during the test runs.
Apart from the Windows version, all hardware, software, benchmark settings, and virtual machine configurations remain identical throughout the research.
During each benchmark run, telemetry is collected from the virtual machine, including CPU utilization, memory usage, disk reads and writes per second, disk queue length, and the number of running processes.
Hypothesis and results
As 26H2 is delivered as an enablement package on the same servicing branch as 25H2, no major architectural or kernel-level changes are expected. Based on previous GO-EUC research comparing 23H2, 24H2, and 25H2, where the differences between releases were marginal, a similar outcome could reasonably be expected here: a comparable overall footprint, with perhaps a slight increase in resource usage due to newly enabled features.
Let’s start with the CPU utilization from the virtual machine perspective.
This is where the results immediately deviate from the hypothesis. The average CPU utilization on the 26H2 preview is approximately 7.6%, compared to 18.2% on 25H2, a reduction of more than half while running the exact same workload. Both versions show similar peaks during the more demanding phases of the workload, which is expected as the workload activity itself has not changed, but the baseline CPU activity between application actions is consistently and significantly lower on 26H2.
A difference of this magnitude is not typical for an enablement package release, so the obvious question is: where does this reduction come from? The number of running processes provides a strong clue.
The 26H2 preview runs an average of 151 processes during the test, compared to 234 on 25H2. That is roughly 80 fewer processes running on an identically configured image with the same workload. This largely explains the lower CPU baseline: fewer background processes means fewer background CPU cycles. It is worth noting that preview builds do not always ship with the full set of shipped features and services enabled, so part of this reduction may not survive into the general availability release. This is something to re-validate once 26H2 reaches Release Preview or general availability.
The reduced process count is also clearly reflected in the memory footprint.
The average memory usage drops from 32.8% on 25H2 to 26.5% on the 26H2 preview. To put this into perspective, a 6.3% difference on a 16 GB virtual machine equates to approximately 1 GB of memory. In a multi-session or high-density VDI environment, a 1 GB per-VM reduction would be a meaningful improvement for consolidation ratios, which makes this one of the more interesting numbers to watch as 26H2 moves towards general availability.
The next charts zoom in on the storage perspective.
Disk reads are nearly identical between both versions, averaging around 7 reads per second, which is expected since read activity is primarily driven by the workload, and the workload remains consistent across both tests. Disk writes show a slight increase on the 26H2 preview, averaging 64 writes per second compared to 56 on 25H2, with a marginally higher peak. This could be related to additional logging or telemetry activity in the preview build, but the difference is small enough that it is unlikely to be noticeable on modern NVMe-backed storage.
The current disk queue length confirms this: with an average of 0.078 on both versions, the storage subsystem is never under pressure during the test, and there is effectively no difference between the two releases from a storage latency perspective.
Conclusion
Based on this early comparison, the Windows 11 26H2 preview shows a surprisingly lighter resource footprint than 25H2 when running an identical knowledge worker workload. The average CPU utilization was reduced by more than half, memory usage dropped by approximately 1 GB on a 16 GB virtual machine, and disk behavior remained effectively unchanged. The most likely driver behind these improvements is the significantly lower number of running processes, roughly 80 fewer on average, observed on the 26H2 preview build.
For EUC architects and VDI administrators, these results are promising: if these numbers hold through to general availability, 26H2 could translate directly into higher user density and improved consolidation ratios without any configuration changes, simply by adopting the new release.
However, a healthy dose of caution is required. This research compares a shipping release against a preview build, and preview builds do not always represent the final product. Services and features that are dormant or absent in the Experimental channel today may well return at general availability, and with them, part of the resource footprint. The significant reduction in running processes is the clearest indicator that the preview build is not yet feature-complete, rather than proof of a dramatic optimization effort.
The key takeaway is that the 26H2 preview currently runs noticeably leaner than 25H2, but these results should be treated as an early indication rather than a final verdict. This research will be worth repeating once 26H2 reaches Release Preview or general availability, to validate whether the reduced footprint survives the journey to the final release.
Photo by Jason Dent on Unsplash
Question to our readers: If Windows 11 26H2 does ship with a materially lower CPU and memory footprint, would that change your upgrade timeline, or would you wait for the first cumulative updates as usual?
