Windows 11 25H2 Preview: No CPU Gains; Linux Leads Creator Benchmarks

Microsoft’s preview of Windows 11 25H2 arrives under a familiar headline: no sweeping performance gains, and in independent creator‑workload benchmarks it still trails Linux by a noticeable margin.

Background​

Windows 11 25H2 is positioned as an incremental, enablement‑style update rather than a ground‑up overhaul. Under the hood it shares the same servicing branch as Windows 11 24H2; many features shipped in 24H2 were simply disabled and are being toggled on in 25H2. That means the public release is expected to behave like a compact enablement package for existing 24H2 machines rather than a dramatic performance reboot.
A fresh round of cross‑platform benchmarking led by a long‑standing hardware testing site compared Windows 11 25H2 (preview/release‑preview builds) against Windows 11 24H2 and Ubuntu 25.10 (plus Ubuntu 24.04.3 LTS as a baseline) on identical hardware. The headline result: across 41 CPU‑focused tests the Ubuntu instance delivered roughly a 15% geometric‑mean advantage over Windows 11 25H2, and 25H2 provided essentially no measurable performance uplift versus 24H2. Those numbers are representative of multi‑threaded creator workloads (rendering, encoding, compression, and other CPU‑bound tasks) rather than gaming or GPU‑bound tests.
This article summarizes those findings, verifies the technical specifics, probes benchmarking methodology, analyzes why Linux continues to lead in many creator workloads, and lays out practical implications — from the enthusiast desktop to enterprise deployments.

---

Overview of the benchmark results​

Test platform and what was measured​

• CPU: AMD Ryzen 9 9950X (16 cores / 32 threads) at stock frequencies.
• Memory: 2 × 16 GB DDR5‑6000.
• Storage: 1 TB Crucial T705 PCIe 5 NVMe.
• GPU: AMD Radeon RX 9070 (present but not the focus).
• Operating systems tested: Windows 11 25H2 (preview), Windows 11 24H2, Ubuntu 25.10 (daily/development snapshots), and Ubuntu 24.04.3 LTS.
• Tests: a suite of 41 CPU‑focused benchmarks including real‑world creator workloads such as LuxCoreRender, Embree, OSPRay, Intel Open Image Denoise, IndigoBench, and other rendering, compression and encoding tests.
• Procedure: clean installs for each OS, stock/default performance and power settings, and runs of binaries available for both Windows and Linux where possible — a methodology intended to reflect an out‑of‑the‑box user experience rather than a hand‑tuned maximum for any one platform.

Key numbers​

• Out of 41 benchmarks, Windows 11 captured just three first‑place finishes against Ubuntu.
• On the geometric mean of all tests, Ubuntu 25.10 was about ~15% faster than Windows 11 25H2 for this Ryzen 9 9950X testbed.
• Windows 11 25H2 produced no meaningful performance gain over Windows 11 24H2 across the measured suite — net average change effectively 0%.

These are not single‑test anomalies; they replicate a recurring pattern seen in similar cross‑platform comparisons performed on other AMD and Intel hardware where Linux variants often lead CPU‑bound workloads.

---

Why Linux continues to lead in many creator workloads​

Leaner userland and fewer legacy constraints​

Linux distributions — especially those optimized for performance or using modern compiler/toolchain defaults — typically ship with a lighter system userland compared to Windows. That means less background work, fewer legacy services, and smaller runtime overhead for the same raw CPU budget.

• Many Linux distros boot with fewer persistent system services by default than a typical Windows installation.
• Open‑source stacks (glibc, LLVM/Clang, GCC toolchains) can offer aggressive optimization tuned to target architectures; some distros also ship kernels configured for throughput rather than desktop responsiveness.
• Minimal background telemetry and fewer legacy‑compatibility layers reduce system noise for CPU‑bound workloads.

Kernel and scheduler differences​

The Linux kernel has seen sustained tuning in recent years for high‑thread‑count workloads and new CPU designs. Windows has made scheduler and power management improvements too, but differences remain:

• Linux’s kernel scheduler and its interaction with specific microarchitectures can benefit heavy parallel workloads.
• For newer AMD Zen designs, kernel development and compiler optimizations have sometimes led to better scaling across many cores.
• Conversely, there are documented cases where Linux initially lagged on hybrid CPU architectures (e.g., Alder Lake family) until scheduler improvements arrived; the performance story depends on the specific CPU generation and kernel version.

Toolchain and ABI effects​

Workloads that heavily rely on compiler optimization (compilers, renderers) will show sensitivity to the toolchain used to build the binaries:

• Linux builds often use GCC or LLVM/Clang with specific flags tuned for the platform, which can influence single‑thread and vectorized performance.
• Windows builds may use MSVC or different runtime libraries, which alters hot paths and can widen or narrow the gap depending on the benchmark.

Filesystem, I/O and runtime libraries​

For I/O‑sensitive tasks, the underlying filesystem and I/O stack matter. Linux offers a number of high‑throughput filesystem choices and kernel improvements (e.g., EXT4, XFS, newer IO schedulers) that can advantage certain mixed CPU/I/O workloads.

---

What to watch in methodology — why the headline number isn’t the whole story​

Benchmarking is valuable, but results must be read with nuance. The tests described here were designed to reflect default user experience rather than a maximal tuned environment; that matters.

• Clean installs and stock settings are appropriate for “what users see right after installing,” but they don’t reflect the performance possible after careful tuning (power plans, affinity, background services, kernel cmdline, specific driver flags).
• Cross‑platform parity is difficult. Even when choosing binaries available for both Windows and Linux, differences in compiler, runtime, and dynamic libraries create unavoidable variance.
• The Linux systems in these tests were development snapshots (Ubuntu 25.10 daily builds) — that produces a preview of the distro’s final kernels and libraries but can vary slightly versus the eventual release. Likewise, Windows 25H2 testing used preview/release‑preview builds, not the final retail release.
• Thermal and power configuration: stock BIOS/UEFI defaults differ machine to machine. Ensuring identical microcode, firmware, BIOS settings, and cooling behavior matters for reproducibility.
• GPU drivers and GPU usage: these CPU‑focused benchmarks intentionally deprioritized GPU workloads. Gaming and graphics tests often favor Windows due to mature DirectX drivers and vendor optimizations; these results do not overturn the conventional wisdom that Windows remains dominant for gaming.

Because the test suite emphasized CPU rendering and encoding, the results reflect areas where Linux historically excels. Different workloads (interactive desktop, gaming, certain Windows‑exclusive professional apps) can flip the balance.

---

Strengths and limitations of Windows 11 25H2 as revealed by the testing​

Strengths​

• Stability and feature parity: 25H2’s approach of enabling features already present in 24H2 minimizes risk and simplifies rollout for enterprise and consumer users who prefer incremental upgrades.
• Incremental security and compatibility updates: by shipping as an enablement package, 25H2 provides a small, low‑risk update path.
• Driver and ecosystem compatibility: Windows still benefits from the broadest ISV and driver ecosystem, especially for professional Windows‑only applications and GPU‑heavy games.

Limitations (from the benchmarks)​

• No generational performance gains for CPU‑bound creator workloads: across the measured suite, 25H2 does not close the gap with Ubuntu 25.10; it’s essentially the same in raw CPU throughput as 24H2.
• Out‑of‑the‑box performance differences persist: for users who prioritize raw creator throughput without tuning, Linux distributions maintain an advantage.
• Windows feature bloat and background services: although Microsoft has made progress improving responsiveness, the additional services and telemetry surface area still create overhead compared to many Linux setups.

---

Practical implications for different user groups​

Creators and content producers​

If your primary need is raw multi‑threaded CPU throughput — rendering, large image or video encoding pipelines, heavy compression jobs — and you’re comfortable with Linux tooling and workflow, the benchmark evidence suggests Linux will likely provide the best out‑of‑the‑box performance for those tasks.

• For studio pipelines where reproducibility and maximum render throughput are required, Linux (or a performance‑oriented distro) is worth considering.
• If you require Windows‑only software (certain Adobe suite features, specialized capture/codec tools), Windows remains functionally necessary. In those cases, consider tuning Windows: power plan adjustments, disabling unnecessary background services, and ensuring drivers are current.

Gamers​

These tests focused on CPU creator workloads, not gaming. For gaming and GPU‑heavy tasks, Windows retains advantages due to DirectX, driver maturity, vendor optimizations, and broader developer attention. The Windows ecosystem still provides the highest compatibility for the latest triple‑A titles and GPU feature sets.

Developers and sysadmins​

Developers who need reproducible build systems or containerized work might prefer Linux for faster compile and build times in many cases. However, enterprise sysadmins must weigh:

• Compatibility with corporate Windows tooling and management.
• The potential to adopt mixed environments: Windows desktops for interactive applications, Linux servers for rendering/backends.

Enterprise and business IT​

25H2 being an enablement package makes it easier for enterprises to adopt without significant retraining or revalidation cycles. Performance parity with 24H2 means minimal disruption — but for teams evaluating platform migrations for throughput reasons, Linux servers or dedicated render farms remain the more efficient choice for CPU-bound batch workloads.

---

How Microsoft is responding (and what to expect next)​

Microsoft recognizes performance criticism around Windows 11 in some scenarios. Recent efforts include:

• A new Feedback Hub logging capability in Insider builds to automatically gather performance traces when users report sluggishness — intended to accelerate root‑cause analysis.
• Stricter driver certification measures, such as requiring static analysis during driver submission, aimed at improving driver quality before deployment.

These actions target stability and responsiveness rather than raw throughput improvements. Given 25H2’s enablement nature, any substantive architecture‑level performance gains are more likely to appear in future, larger releases.

---

Risks, caveats and unverifiable claims​

• Benchmark variability: microcode updates, BIOS versions, and drivers can materially influence outcomes. Results reported here reflect a specific hardware/firmware/software snapshot.
• Preview OS builds and daily distro snapshots: because Windows 25H2 testing used preview builds and Ubuntu used development snapshots, final release numbers could shift slightly; the broad pattern is unlikely to reverse, but marginal differences may change.
• Generalization risk: these tests focused on CPU‑bound creator tasks. Extrapolating these findings to all workloads (interactive office tasks, legacy enterprise applications, or all gaming) would be incorrect.
• Vendor or configuration‑specific optimizations: vendor‑supplied drivers or tuned distros (Clear Linux, for example) can produce materially different numbers than the Ubuntu baseline and may widen the Linux advantage in similar tests.

Any claim about “this makes one OS always better than another” should be treated cautiously; real choices depend on workload, ecosystem constraints, tooling, and whether the user will tune systems further.

---

Actionable recommendations​

• If you’re a content creator focused on multi‑threaded rendering or encoding and can operate in a Linux environment:
• Evaluate Ubuntu 25.10 or a performance‑oriented distro for production pipelines.
• Run representative tests using your actual workloads before deciding to migrate.
• If you must remain on Windows for application compatibility:
• Keep drivers and firmware updated and consider a targeted performance audit (power plan, background processes).
• For heavy batch jobs, consider dedicated Linux build/render servers while retaining Windows workstations for interactive tasks.
• For gamers:
• Stick with Windows for the broadest compatibility and best GPU performance for modern titles.
• For IT managers:
• Treat 25H2 as a low‑risk enablement update; test internal applications but expect minimal performance regression or gain relative to 24H2.
• If throughput is a strategic requirement, design hybrid environments that leverage Linux where appropriate.

---

The larger takeaways for the Windows ecosystem​

• Windows 11 25H2’s incremental nature reflects Microsoft’s current update cadence: refine, enable, and stabilize rather than rearchitect at every cadence. That’s sensible for enterprise stability but means fewer headline performance wins for enthusiast users.
• Linux’s consistent advantage in CPU‑bound creator workloads remains a combination of kernel evolution, lighter system footprint, and toolchain effects.
• The competitive landscape is nuanced: Windows still dominates gaming and many professional app ecosystems, while Linux remains compelling for batch throughput and developer workflows.
• Expect Microsoft to continue iterative improvements (feedback‑driven fixes, driver certification updates). Major shifts in raw CPU workload performance are more likely to stem from future, larger architectural changes than from 25H2.

---

Conclusion​

The early 25H2 preview benchmarks reaffirm a recurring narrative: an incremental Windows update does not automatically close performance gaps where Linux has an edge. For creators running CPU‑heavy production tasks, Linux distributions continue to deliver better out‑of‑the‑box throughput in multi‑threaded workloads. For most users, Windows 11 25H2 will feel familiar — updated and slightly refined but not faster in a way that changes the performance calculus many professionals and enthusiasts care about.
These findings are a call to choose the right tool for the job. Where software compatibility or gaming matters most, Windows remains the pragmatic choice. Where raw throughput and scale of CPU‑bound jobs are priorities and the software stack permits it, Linux still pulls ahead. The wise path for many organizations is hybrid: retain Windows for interactive work and compatibility, offload heavy batch workloads to tuned Linux environments — and continue to test with your actual workloads before making platform decisions.

---

Source: HotHardware Microsoft's Upcoming Windows 11 25H2 Battles Linux In Benchmark Showdown