New performance data from Phoronix reveals that Ubuntu 25.10 daily builds outpace Microsoft’s Windows 11 25H2 by roughly 15% in CPU-intensive workloads on AMD’s flagship Ryzen 9 9950X processor. The independent testing, conducted with clean OS installations and stock settings, highlights a measurable throughput advantage for the Linux distribution in multi-threaded rendering, encoding, and compression tasks—while Windows 11 25H2 itself delivers no meaningful performance gain over the existing 24H2 release.

The Test Setup: A High-End Zen 5 Workload Cruncher

Phoronix built its comparison around a system tailored to stress CPU parallelism: an AMD Ryzen 9 9950X with 16 cores and 32 threads, 32 GB of DDR5-6000 RAM, a 1 TB Crucial T705 PCIe 5.0 NVMe SSD, and an AMD Radeon RX 9070 GPU. Both operating systems were installed from scratch with default settings, mirroring the out-of-box experience for most users and enterprises. Ubuntu 25.10 was based on pre-release daily snapshots, while Windows 11 25H2 was applied as an enablement package (eKB) over a 24H2 foundation—Microsoft’s planned delivery mechanism for this update.

The test suite deliberately excluded GPU-bound gaming to focus on long-running, multi-threaded CPU tasks where kernel scheduler efficiency, compiler optimizations, and background service overhead can swing results significantly. A total of 41 cross-platform benchmarks were run, spanning Blender CPU rendering, LuxCoreRender, Embree, Intel Open Image Denoise, OSPRay, IndigoBench, Kvazaar/FFmpeg video encoding, ASTC texture compression, 7-Zip compression, and more.

Headline Numbers: A 15% Geomean Gap

Phoronix’s geomean across all 41 tests landed Ubuntu 25.10 approximately 15% faster than Windows 11 25H2 on this specific hardware. Individual tests varied, but the aggregate trend favored Linux across the majority of multi-threaded workloads. Windows managed only a handful of wins, primarily in single-threaded synthetic tests and certain encoder configurations where Windows-specific runtime optimizations gave it an edge.

Crucially, the comparison between Windows 11 24H2 and 25H2 showed no performance delta—the geomean difference was effectively zero, within measurement noise. This aligns with Microsoft’s documentation that 25H2 is an enablement package flipping feature flags already present in monthly cumulative updates, not a kernel or scheduler overhaul.

Why Ubuntu 25.10 Pulls Ahead

Several technical factors combine to give Ubuntu 25.10 its lead in these benchmarks:

Newer Kernel and Scheduler Enhancements

Ubuntu 25.10 ships with development versions of the Linux 6.16/6.17 kernel series, which include refinements for Zen 5 processors—better thread placement, frequency scaling, and cache affinity. For workloads that hammer all 32 threads, these scheduler tweaks keep more cores fed with work and reduce contention.

Modern Toolchain and Compiler Optimizations

The daily builds use GCC 15 (or near-final snapshots), which generates more efficient vectorization and instruction scheduling for AMD’s latest microarchitecture. When benchmark binaries are compiled natively with these up-to-date toolchains, they can extract higher instructions-per-clock than older compilers typically available on Windows for cross-platform open-source projects.

Leaner Default Services and Telemetry

Windows 11’s default configuration includes a variety of background services, telemetry collection, and virtualization-based security features that can consume CPU cycles and affect scheduling. While these can be tuned or disabled, Phoronix intentionally used stock settings to represent the typical user experience. Ubuntu’s default install has a far smaller footprint of always-on processes, reducing noise during sustained number-crunching.

Where Windows Still Holds Ground

The results are not a clean sweep for Linux. Windows retains important advantages:

  • Single-threaded responsiveness: Several latency-sensitive and single-threaded tests showed parity or a slight Windows lead, reflecting its mature per-thread scheduling for interactive tasks.
  • Proprietary software and drivers: Applications optimized specifically for Windows—including many creative suites and vendor-tuned binaries—can still outperform Linux counterparts because of platform-specific optimizations and better driver integration.
  • Gaming dominance: This CPU-focused phase did not test gaming, where Windows’ GPU driver stack and per-title optimizations remain the gold standard for most gamers.

Why Windows 11 25H2 Is Not a Performance Update

Microsoft has been transparent that 25H2 will be delivered as an enablement package, sharing the same core servicing branch as 24H2. When a fully patched 24H2 system applies the eKB, a single reboot flips feature flags to enable new surface elements like updated UI touches and the removal of legacy components (PowerShell 2.0, WMIC) on certain editions. No major kernel binaries are replaced, so no significant throughput uplift should be expected—a point Phoronix’s data confirms.

For IT administrators, the enablement model means fast, low-risk deployments with minimal downtime. But it also means organizations looking for raw performance gains from Windows updates should focus on driver, firmware, and future feature updates rather than expecting 25H2 alone to deliver a speed boost.

Workload-by-Workload Deep Dive

Multi-Threaded Rendering and Production

Blender, LuxCoreRender, Embree, and OSPRay all scale across many cores. Here, Ubuntu’s lead was often 5–15%—a gain that compounds significantly in render farms where hours-long jobs run repeatedly. Creators who rely on open-source renderers may find switching to Linux for rendering tasks worth piloting.

Video Encoding and Compression

Batch encoding with Kvazaar (HEVC) and FFmpeg, plus heavy 7-Zip compression, leaned toward Ubuntu in the geomean. Differences trace back to thread scheduling overhead, syscall paths, and how the CPU frequency governor reacts under sustained load. Windows did win sporadic encoder tests, likely due to library-level tuning on that platform.

Single-Threaded and Small I/O Tasks

Windows often matched or edged ahead in single-core synthetics and small-file operations. These workloads emphasize per-thread latency and interactivity, an area where Windows’ scheduler traditionally excels.

Reproducibility, Caveats, and the “First-Look” Label

Phoronix and independent analysts caution that these are early snapshot results, not a universal verdict. Several variables can shift outcomes:

  • Firmware and microcode: BIOS updates, AMD chipset drivers, and AGESA versions can alter Zen 5 behavior significantly.
  • Compiler and build parity: Using identical binaries compiled with the same toolchain (e.g., via containers) would narrow the gap; here, native builds on Linux benefited from newer compilers.
  • Tuned configurations: Enterprises often strip down Windows images or tune power plans; such optimizations could close the gap, but Phoronix measured defaults to reflect the typical user scenario.

For those replicating the tests, Phoronix has published OpenBenchmarking logs enabling audit trails. The advice is clear: test your own exact workloads, capture BIOS/driver versions, and re-measure after updates before making platform decisions.

What This Means for Different Audiences

Content Creators and Render Farms: A 15% render time reduction can translate into significant cost savings. Pilot Ubuntu 25.10 on a render node with your own scenes and presets—even half that improvement compounds over projects.

Developers and CI/CD Pipelines: Build farms running Linux may benefit from the latest kernels and compilers sooner than Windows environments. Measure end-to-end build times with your actual toolchains before migrating.

Gamers and Windows-Only Application Users: This data set does not address gaming, and Windows remains the platform of choice for most titles and proprietary professional software. The raw CPU throughput edge matters less if your workflow is anchored to Windows-specific tools.

Enterprise IT Operations: 25H2’s enablement package simplifies rollout and minimizes reboots, which is operationally valuable. But it is not a performance re-engineering; expect real performance improvements to come from driver and firmware updates, not from the eKB itself.

A Reproducible Signal, Not a Final Answer

Phoronix’s benchmarks provide a clear, reproducible signal on one configuration: a high-core-count Ryzen 9 9950X workstation running CPU-bound open-source workloads. Under those conditions, Ubuntu 25.10 delivers a significant throughput advantage over Windows 11 25H2, while Windows’ own update shows no delta from 24H2. The reasons are traceable to kernel freshness, compiler versions, and default service overhead—not to a fundamental inability of Windows to match Linux when tuned.

As both operating systems approach general availability, these results should inform testing strategies for performance-conscious users. The broader takeaway is not that one OS is universally faster, but that the toolchain and scheduler stack matter enormously for modern high-core-count processors—and that Linux distributions often get first access to those innovations. For Windows enthusiasts, the data underscores that 25H2 is an operational rather than a performance milestone, keeping the door open for future kernel-oriented updates to close the gap.