A fresh round of benchmarks pitting Ubuntu 25.10 against Ubuntu 25.04 on AMD’s Zen 5 architecture reveals something that enterprise Linux administrators and workstation buyers should pay close attention to: the jump from Linux kernel 6.14 to the upcoming 6.15, combined with GCC 15 and Mesa 25.1, delivers measurable and sometimes dramatic performance gains across a wide range of workloads. The improvements aren’t uniform — some benchmarks barely budge, others leap by double-digit percentages — but the overall trajectory is clear. AMD’s newest silicon is still leaving performance on the table that newer software stacks are only now beginning to unlock.
Phoronix ran 250 benchmarks comparing Ubuntu 25.04 “Plucky Puffin” (shipping with Linux 6.14, GCC 14.2, and Mesa 25.0) against a daily development build of Ubuntu 25.10 “Questing Quokka” (running Linux 6.15, GCC 15, and Mesa 25.1) on an AMD Ryzen 9 9950X — the flagship 16-core, 32-thread Zen 5 desktop processor. The test platform used 32GB of DDR5-6000 memory and a Radeon RX 9070 XT GPU. Same hardware. Different software. The results tell a nuanced story.
Out of those 250 tests, Ubuntu 25.10 came out ahead 62% of the time. That’s not a blowout. But when it won, it often won big, while Ubuntu 25.04’s victories tended to be marginal. The geometric mean across all benchmarks showed a 3% overall advantage for the newer release — a number that understates the gains in specific workload categories.
Where the Gains Actually Land
Compiler performance showed some of the most consistent improvements. The LLVM compilation benchmark ran roughly 4% faster under Ubuntu 25.10. Timed Linux kernel compilations also improved. For shops running continuous integration pipelines on AMD hardware, these aren’t trivial numbers — they compound across hundreds or thousands of daily builds.
Cryptographic workloads saw even larger jumps. OpenSSL 3.5 performance on SHA-256 and SHA-512 operations improved substantially, with some sub-tests showing gains north of 10%. This matters for server workloads handling TLS termination, disk encryption, or blockchain-adjacent computation. The combination of kernel-level optimizations and compiler improvements appears to be giving the CPU’s instruction pipeline more room to breathe.
Scientific and HPC-style benchmarks were a mixed bag. Some OpenFOAM computational fluid dynamics tests improved. Others held steady. The NAS Parallel Benchmarks, a staple of supercomputing performance evaluation, showed modest gains in several kernels. Nothing earth-shattering, but consistently in the right direction.
Graphics told an interesting story too. Mesa 25.1 brought visible improvements in several Vulkan and OpenGL benchmarks, including notable gains in some gaming workloads on the Radeon RX 9070 XT. Titles tested through benchmarking suites like Geekbench Vulkan and various Phoronix Test Suite GPU tests showed the newer Mesa stack extracting more from RDNA 4 hardware. For creative professionals running GPU-accelerated rendering on Linux, this is directly relevant.
Database performance was another bright spot. PostgreSQL’s pgbench showed measurable throughput improvements under the 25.10 stack. Redis benchmarks moved in the right direction. These are workloads where kernel scheduler changes and memory management tweaks can have outsized effects, and Linux 6.15 appears to carry several such refinements.
Not everything improved. A handful of benchmarks regressed slightly — some audio encoding tasks, a few niche computational tests. The regressions were generally small, in the 1-2% range, and could reflect anything from measurement noise to intentional tradeoffs in the newer toolchain. None of the regressions looked alarming.
So what’s driving the gains? It’s not one thing. GCC 15 brings improved auto-vectorization and better optimization passes for Zen 5’s microarchitecture. Linux 6.15 carries scheduler refinements, memory management improvements, and updated AMD-specific driver code. Mesa 25.1 includes new compiler backends for RDNA 4. These layers interact. A faster compiler generates tighter code. A smarter scheduler places threads more effectively. A newer graphics driver translates API calls more efficiently. Stack them together and you get compounding returns.
This pattern — where AMD hardware takes a full kernel generation or two to reach its performance potential on Linux — has repeated with every recent Zen release. Zen 4 showed similar behavior. Early kernel support worked fine, but peak performance arrived months later as scheduler patches, power management fixes, and compiler tuning caught up. Zen 5 is following the same arc. Buyers who deployed Ryzen 9000 or EPYC Turin systems on day-one kernel support were functional but not optimal. The software is now catching up to the silicon.
For enterprise IT departments evaluating AMD workstations or servers, the implication is straightforward: the kernel version matters as much as the hardware SKU. Running a Zen 5 system on an older LTS kernel means leaving real performance on the floor. Ubuntu 24.04 LTS, for instance, ships with Linux 6.8 — two full versions behind what 25.10 will carry. The HWE (Hardware Enablement) kernel updates help, but they lag behind the mainline development branch by months.
And this isn’t just an Ubuntu story. Fedora, Arch, and openSUSE Tumbleweed users tracking upstream kernels will see similar gains. The improvements live in the mainline kernel, in GCC, and in Mesa — all upstream projects. Ubuntu’s release cadence simply provides a convenient snapshot for benchmarking.
The Phoronix results also carry implications for AMD’s competitive positioning against Intel. Intel’s Core Ultra processors have generally had tighter day-one Linux optimization, partly because Intel employs a large team of kernel developers who upstream patches before silicon ships. AMD has been closing this gap — the company’s Linux engineering investment has grown significantly — but the pattern of post-launch performance unlocks persists. Buyers making purchasing decisions based on launch-day Linux benchmarks may be undervaluing AMD hardware relative to what it delivers six to twelve months later.
Ubuntu 25.10 is scheduled for release in October 2025. The daily builds being tested now are unstable and not suitable for production. But the kernel, compiler, and Mesa versions they carry will be available to any distribution tracking upstream releases well before then. Fedora 43, expected in the same timeframe, will likely ship comparable versions. Rolling distributions already have access to most of these components.
The bottom line for anyone running AMD Zen 5 hardware on Linux: update aggressively. The hardware you bought is faster than the software you’re running on it — and the gap is closing with every kernel release. A 3% geometric mean improvement across 250 benchmarks, with double-digit gains in key workloads, isn’t a rounding error. It’s real money in compute-heavy environments. And it’s free — no new hardware required.
WebProNews is an iEntry Publication