Early benchmark data shows Ubuntu 25.10 holding a measurable advantage over Microsoft’s Windows 11 25H2 in CPU‑heavy workloads on a high‑end AMD Ryzen 9 9950X testbed, with Phoronix’s first‑look geomean numbers pointing to roughly a ~15% edge for the Linux build in this specific profile. (phoronix.com)
Ubuntu 25.10 and Windows 11 25H2 are both due for broad release in the coming weeks, but they arrived in preview/daily snapshot form early enough for independent benchmarking. Phoronix ran a head‑to‑head comparison using clean installs and stock defaults to show where the operating systems land out of the box on identical hardware. That hardware—an AMD Ryzen 9 9950X (16 cores / 32 threads) with 32 GB DDR5‑6000, a 1 TB Crucial T705 PCIe 5.0 NVMe SSD, and an AMD Radeon RX 9070 GPU—was chosen to stress multi‑threaded CPU workloads typical of creators, render farms, and compile clusters. (phoronix.com)
Microsoft’s 25H2 update is delivered as an enablement package (eKB) that shares a servicing branch with 24H2; the package flips feature flags already shipped in monthly cumulative updates rather than replacing core OS binaries. That delivery model is explicit in Microsoft’s communications and explains why a large, system‑wide performance uplift from 25H2 alone was not expected. (blogs.windows.com, techcommunity.microsoft.com)
Practical implications for IT admins:
For creators, build‑farm operators, and CI managers, these numbers justify a concise pilot: clone representative jobs, reproduce the Phoronix‑style runs in your environment, and measure wall‑clock impact. For gamers and Windows‑dependent users, the headline is unlikely to change daily workflows—Windows retains strengths where it matters for those audiences.
Finally, treat this as the informed opening chapter of an ongoing story: drivers, firmware, compiler versions, and future microcode updates can—and often do—alter performance balances. The best engineering decision remains empirical measurement against the actual workloads you run, followed by automated, auditable benchmarking that you can reproduce across updates.
Source: extremetech.com Ubuntu 25.10 Outperforms Windows 11 25H2 in First CPU Benchmarks
Background
Ubuntu 25.10 and Windows 11 25H2 are both due for broad release in the coming weeks, but they arrived in preview/daily snapshot form early enough for independent benchmarking. Phoronix ran a head‑to‑head comparison using clean installs and stock defaults to show where the operating systems land out of the box on identical hardware. That hardware—an AMD Ryzen 9 9950X (16 cores / 32 threads) with 32 GB DDR5‑6000, a 1 TB Crucial T705 PCIe 5.0 NVMe SSD, and an AMD Radeon RX 9070 GPU—was chosen to stress multi‑threaded CPU workloads typical of creators, render farms, and compile clusters. (phoronix.com)Microsoft’s 25H2 update is delivered as an enablement package (eKB) that shares a servicing branch with 24H2; the package flips feature flags already shipped in monthly cumulative updates rather than replacing core OS binaries. That delivery model is explicit in Microsoft’s communications and explains why a large, system‑wide performance uplift from 25H2 alone was not expected. (blogs.windows.com, techcommunity.microsoft.com)
What the tests measured — methodology and scope
Phoronix focused deliberately on CPU‑bound, cross‑platform workloads to maximize sensitivity to kernel, scheduler, and toolchain differences rather than GPU bottlenecks. Key elements of the testing approach:- Clean OS installs for each distribution/version to avoid configuration carryover. (phoronix.com)
- Stock performance/power defaults on each OS—an out‑of‑the‑box comparison rather than a heavily tuned one. (phoronix.com)
- A mix of real‑world producer workloads and synthetics: Blender CPU renders, LuxCoreRender, Embree, Intel Open Image Denoise, OSPRay, IndigoBench, various encoders (Kvazaar/FFmpeg), ASTC texture compression, 7‑Zip, and others. The published suite included 41 cross‑platform tests in this run. (phoronix.com, tomshardware.com)
- Geometric means reported across the test set to summarize aggregate throughput while still showing per‑test variation. (phoronix.com)
Headline results — the numbers that matter
- On the chosen Ryzen 9 9950X testbed, Ubuntu 25.10’s geomean across the selected CPU workloads landed roughly ~15% faster than Windows 11 25H2 in Phoronix’s snapshot. (phoronix.com)
- Windows 11 25H2 showed no meaningful throughput gain over Windows 11 24H2 in this benchmark suite—the geomean difference between 24H2 and 25H2 was effectively zero within measurement noise. (phoronix.com, tomshardware.com)
- Individual tests varied: a small number favored Windows in specific workloads, but the majority of CPU‑heavy, multi‑threaded producer tasks favored Ubuntu in these runs. Out of the 41 cross‑platform tests, Windows scored only a handful of wins while Linux took most of the rest. (phoronix.com)
Why Ubuntu 25.10 often outperforms in this profile
The benchmark delta is not magic; it’s the result of several interacting factors where Linux has, in this snapshot, favorable defaults or newer components:Kernel and scheduler timeliness
Ubuntu 25.10 daily builds are shipping newer upstream Linux kernels (6.16/6.17 series in development snapshots) and policy defaults that can better exploit high core‑count Zen‑5 behavior for sustained throughput. Those kernel and scheduler improvements can affect thread placement, frequency scaling, and cache affinity—all material for long, parallel jobs. (phoronix.com)Toolchain and compiler versions
Newer compilers and libraries (for example, GCC 15 series in the 25.10 toolchain) produce more optimal vectorization and code generation for modern instruction sets. When cross‑platform workloads are built natively with those up‑to‑date toolchains on Linux, they can extract additional throughput versus older toolchains or Windows toolchains used for the comparative runs. (phoronix.com)Leaner default services and telemetry
Windows ships with a set of background services, telemetry, and virtualization‑based security features enabled by default that can consume CPU and scheduling attention in the broad out‑of‑the‑box case. Phoronix intentionally used stock settings to reflect what a typical user would see; in that configuration, Linux’s leaner service footprint can be a practical advantage for raw throughput. That’s not a claim that Windows cannot be tuned to match—it can—but Phoronix’s approach was to measure default behavior. (phoronix.com)Where Windows still wins
The results are nuanced: Windows retained advantages in certain classes of tests and remains the practical choice for many users.- Single‑threaded and latency‑sensitive tasks frequently favored Windows or tied, reflecting longstanding differences in per‑thread responsiveness. (phoronix.com)
- Windows‑only applications and vendor‑tuned binaries still benefit from platform‑specific optimizations and proprietary drivers. In real‑world deployments where those applications are required, Windows remains the necessary platform. (phoronix.com)
- The testing did not examine gaming in this phase—GPU driver stacks and per‑title optimizations continue to make Windows the dominant gaming platform for most titles and driver scenarios.
Enterprise and management context: why 25H2 didn’t move the needle
Microsoft is shipping 25H2 as an enablement package on top of the shared servicing branch used by 24H2. An eKB toggles features delivered earlier in monthly LCUs from disabled to enabled; it does not perform a monolithic binary rebase of the OS. That engineering choice prioritizes fast, low‑disruption deployments and compatibility over sweeping kernel rewrites that could yield systemic throughput changes. (blogs.windows.com, techcommunity.microsoft.com)Practical implications for IT admins:
- Fast rollout and low downtime for managed fleets (a single reboot for fully patched systems). (techcommunity.microsoft.com)
- Targeted removals and manageability features (PowerShell 2.0 removal, WMIC deprecation, and options for removing select preinstalled Store apps on Enterprise/Edu devices) matter operationally—but they do not equate to a kernel‑level performance transformation. (blogs.windows.com, learn.microsoft.com)
Deep dive: workload‑by‑workload behavior
Multi‑threaded rendering and producers (Blender, LuxCore, Embree)
These workloads scale across many cores and thread groups. In this profile, Ubuntu 25.10’s newer kernel, scheduler heuristics, and toolchain gave it consistent leads—often several single‑digit percent advantages, and in some cases up to the reported ~10–15% range. For creators, a 5–15% rendering speedup on long renders multiplies strongly across production pipelines. (phoronix.com)Video encoding and compression
Multi‑file batch encodes and parallel compression tests also tended to favor Ubuntu in this snapshot. Differences here often trace back to thread scheduling, syscall overhead, and how frequency governors respond under prolonged load. That said, Windows sometimes led on specific encoder builds where Windows runtimes or libraries were tuned differently. Cross‑platform identical binaries reduce this variance. (phoronix.com)Single‑threaded synthetics and small IO jobs
Windows retained parity or modest leads in several single‑threaded synthetic tests and small‑IO scenarios where latency and per‑thread responsiveness trump raw parallel throughput. These workloads represent everyday interactivity and legacy software behavior where Windows remains consistently competitive. (phoronix.com)Reproducibility, caveats, and the “first‑look” warning
Phoronix and other outlets stress that these runs are snapshots, not canonized final verdicts. Key caveats:- Drivers and firmware matter: BIOS/UEFI microcode, chipset drivers, NVMe driver behavior, and GPU drivers can move numbers materially; a BIOS or driver update can flip results on short notice. (phoronix.com, tomshardware.com)
- Builds and compiler choices matter: Running identical binaries cross‑platform reduces one large source of variance. Native builds using newer compilers on Linux advantage Ubuntu; recompiling with comparable modern toolchains on Windows (where possible) narrows that gap. (phoronix.com)
- Default vs tuned systems: Phoronix measured stock defaults to represent the typical experience. Aggressive tuning on either platform (disabling specific services, adjusting governors or power plans, or applying vendor‑recommended driver settings) changes outcomes and may favor either OS depending on the effort put in. (phoronix.com)
- Reproduce the exact testbed (same CPU stepping, BIOS, microcode).
- Use identical binary builds where possible (containerized or statically built cross‑platform tools).
- Capture OpenBenchmarking or equivalent logs to make runs auditable.
- Re‑run after driver and firmware updates, and track variance across runs.
What this means for different user groups
Content creators and render farms
If your workload is dominated by long, CPU‑heavy renders or batch encodes, the Phoronix snapshot suggests Ubuntu 25.10 is worth piloting—even a modest percentage gain compounds across hours of render time and multiple render nodes. Test your own scenes and encoder presets, but the potential payoff is real for throughput‑sensitive operations. (phoronix.com)Developers and CI pipelines
Build farms and compile clusters that can run on Linux may benefit from updated kernels and compilers that often appear earlier in Linux distributions. For critical pipelines, measure end‑to‑end wall time with your actual workloads before committing to platform changes. (phoronix.com)Gamers and Windows‑dependent users
For gaming, Windows remains the practical platform for the vast majority of titles and driver optimizations; Phoronix’s CPU‑focused snapshot does not move that needle. If your workflow relies on Windows‑only applications (creative suites, vendor‑locked tools), those software dependencies outweigh raw OS throughput for most users. (phoronix.com)Enterprises and IT operations
25H2’s enablement package model offers operational advantages—fast installs, fewer reboots, and targeted removals—but it’s not a performance re‑engineering of the OS. If you are planning broad updates for performance reasons, the enablement package should be treated as a low‑disruption update; expect substantial performance work to come from drivers, firmware, or separate targeted updates. (blogs.windows.com, techcommunity.microsoft.com)Practical recommendations (a short checklist)
- For creators and builders: run an A/B test of a representative job suite on Ubuntu 25.10 (or updated LTS kernels) vs your Windows 24H2/25H2 image. Use identical binaries or containers when possible. (phoronix.com)
- For IT admins: treat 25H2 as an operational enablement package—plan pilot upgrades and validate removal of legacy components (PowerShell 2.0, WMIC) in your estate. (blogs.windows.com, learn.microsoft.com)
- For mixed shops: consider hybrid designs—Linux worker nodes for rendering/CI, Windows workstations for specialized proprietary apps, and shared artifact storage to avoid cross‑platform build drift. (phoronix.com)
- Always capture logs and version information (kernel, BIOS, microcode, driver versions) when benchmarking; small differences here explain most variance.
Strengths and risks — critical analysis
Strengths of the findings
- The test suite is intentionally sensitive to OS scheduling and compiler/toolchain differences; for the targeted audience (creators, HPC, CI), this is the right lens. (phoronix.com)
- Phoronix’s methodology (clean installs, stock defaults, OpenBenchmarking traces) is transparent and reproducible, enabling further community verification. (phoronix.com)
- The 25H2 enablement model is well documented by Microsoft; the absence of a large kernel‑level change is consistent with what Microsoft publicly described. That alignment between disclosure and observed results strengthens the credibility of the measurements. (blogs.windows.com, techcommunity.microsoft.com)
Risks and limitations
- Benchmarks are hardware‑ and workload‑dependent. The Ryzen 9 9950X is a high‑core‑count Zen‑5 chip where scheduler and frequency scaling matter; different CPUs (Intel, lower core counts, or mobile SKUs) may exhibit different relative behavior. (phoronix.com)
- Driver/firmware updates can change results quickly. Relying on a single snapshot for purchasing or architecture decisions is risky; continuous measurement in your environment is recommended. (phoronix.com)
- Default vs tuned configurations: the out‑of‑the‑box advantage for Ubuntu does not mean Windows cannot match or beat Linux after tuning. Enterprises often tune Windows images; Phoronix intentionally did not apply that level of tuning to reflect typical defaults. (phoronix.com)
Final analysis and conclusion
Phoronix’s first benchmarks present a clear, reproducible signal: on the tested Ryzen 9 9950X workstation and using an array of CPU‑heavy producer workloads, Ubuntu 25.10 daily snapshots outperformed Windows 11 25H2 by an appreciable margin in aggregate throughput—roughly ~15% geomean in these runs—while Windows 11 25H2 itself showed no meaningful throughput improvement over 24H2 due to Microsoft’s enablement‑package delivery model. (phoronix.com, tomshardware.com)For creators, build‑farm operators, and CI managers, these numbers justify a concise pilot: clone representative jobs, reproduce the Phoronix‑style runs in your environment, and measure wall‑clock impact. For gamers and Windows‑dependent users, the headline is unlikely to change daily workflows—Windows retains strengths where it matters for those audiences.
Finally, treat this as the informed opening chapter of an ongoing story: drivers, firmware, compiler versions, and future microcode updates can—and often do—alter performance balances. The best engineering decision remains empirical measurement against the actual workloads you run, followed by automated, auditable benchmarking that you can reproduce across updates.
Source: extremetech.com Ubuntu 25.10 Outperforms Windows 11 25H2 in First CPU Benchmarks
