Dev

Linux 7.3 Improves Strix Halo NUMA Performance

Ubuntu 26.10 comparison of Linux 7.2 vs 7.3 shows improved NUMA on Strix Halo with minor graphics and latency gains.

7 min read Reviewed & edited by the SINGULISM Editorial Team

Linux 7.3 Improves Strix Halo NUMA Performance
Photo by Olivier Collet on Unsplash

In the development version of Ubuntu 26.10, the shipped kernel has switched from Linux 7.2 to Linux 7.3. Following the switch, a performance comparison was conducted in reporting by Phoronix’s Michael Larabel. The target was a Framework Desktop equipped with the AMD Ryzen AI Max+ 395. Using the same machine, the differences between Linux 7.2 and Linux 7.3 were tested.

Last week’s testing covered the combination of Ubuntu 26.10 and Linux 7.2. Improvements over the combination of Ubuntu 26.04 LTS and Linux 7.0 had been confirmed. This time, a direct comparison of 7.2 and 7.3 was performed to follow the subsequent kernel transition. As the October final release is expected to ship 7.3, the test can be positioned as close to real-world use.

The graphics benchmarks were largely the same with Linux 7.3 over Linux 7.2 for the Radeon 8060S Graphics

The above is the original article’s summary of graphics performance. While there was no major change overall, small improvements were recorded under some conditions. No regressions from the development kernel update were observed. This suggests maturity was already high at Linux 7.2.

Ubuntu 26.10 Kernel Transition and Test Conditions

Canonical updated the default kernel during development of Ubuntu 26.10. Shortly after daily builds used Linux 7.2, it moved to Linux 7.3. This comparison is a follow-up test prompted by that transition. The user space and test configuration were kept identical.

The test machine is a Framework Desktop equipped with the Ryzen AI Max+ 395. This chip, known as Strix Halo, integrates CPU and integrated graphics in a single design. It features a unified architecture that shares memory. It is attracting attention for portable high-performance computing and inference uses.

Testing included dozens of real-world workloads and kernel micro-benchmarks. Graphics, network latency, NUMA, and responsiveness were covered. Real-world workloads were run under conditions close to everyday processing. The micro-benchmarks were intended to isolate behavioral differences inside the kernel.

In Linux desktop environments, the impact of kernel updates is drawing interest. Linux Mint Unveils Xepub EPUB Reader and Clockenstein Calendar also reported updates to the desktop environment. While the kernel and desktop environment are separate layers, the user experience is determined by their combination. This testing is significant in isolating changes at the kernel layer.

Changes in Radeon 8060S Graphics Performance

Results for the integrated Radeon 8060S graphics were largely flat. Updating to Linux 7.3 produced neither major gains nor losses. This shows that the mature driver stack is already stable. For existing users, the absence of regressions is what matters.

As an exception, a small improvement appeared in the side-view action game DDNet. The difference emerged under lightweight graphics loads. It may reflect fine-tuning of drivers or synchronization handling. It can be seen as progress in optimization for specific conditions.

No significant differences were confirmed in other graphics tests. Under heavy 3D rendering loads, differences between kernels were buried. Graphics performance depends more on the maturity of user-space drivers and Mesa than on the kernel. This can be interpreted as optimization already being advanced at 7.2.

Despite being integrated, Strix Halo is close in scale to discrete graphics. As a result, changes in kernel memory management or power control could have an effect. This time, that effect was limited. The update can be evaluated as maintaining stability.

Major NUMA Improvement and Lower Latency

The clearest change was in NUMA testing with Stress-NG. Measurements improved substantially on Linux 7.3. Handling of the non-uniform memory access architecture appears to have become more efficient. It is a favorable result for Strix Halo with its large integrated unified memory.

This trend is also consistent with server-side measurements. Phoronix reported gains from kernel improvements in separate testing targeting EPYC. The report states that 2026 kernel improvements delivered about an 8% improvement in geometric mean. Common improvements can be seen as spreading from server to client.

Reductions were also confirmed in network latency measurements. The difference was small but reproduced across multiple runs. It may reflect adjustments to kernel networking processing and timer mechanisms. For latency-sensitive uses, it is a change that cannot be ignored.

Partial improvements also appeared in other micro-benchmarks. They do not directly raise overall system throughput. Still, more efficient latency and interrupt handling is meaningful as an accumulation. For a kernel under development, this is steady progress.

Limited Differences in Real-World Workloads

Across dozens of real-world workloads, overall differences were small. No clear differences emerged in document processing, encoding, compression, or scripting-language processing. Perceived differences in daily use are likely to be minimal. This is also the flip side of the high polish of Linux 7.2.

The previous round of testing had confirmed improvements over Ubuntu 26.04 LTS. The change from Linux 7.0 to 7.2 was relatively large. By contrast, the change from 7.2 to 7.3 remained fine-tuning. It is a natural result as part of a gradual maturation process.

The Framework Desktop is a machine with mature Linux support. User-space driver and power-management issues have already been resolved. As a result, the effects of kernel updates are hard to see in measurements. Unsupported machines could show different results.

Beyond performance, Linux 7.3 is highly significant. It includes new features, hardware support, and security hardening. For hardware outside the scope of this testing, the benefits may be larger. Flat performance does not mean the update has little value.

Linux 7.3 Feature Enhancements and Outlook

Linux 7.3 incorporates many changes as a development kernel. Expanded hardware support and internal restructuring have progressed. For new chips like Strix Halo, fuller support is important. It includes stability improvements not reflected in measurements.

Ubuntu 26.10 has frozen the kernel ahead of its October release. Remaining bug fixes will proceed on the 7.3 base. This testing reflects the state just before release. The same trend is expected to hold in the final release.

Development toward Linux 7.4 will now accelerate. Faster read-ahead processing, among other improvements, has already been reported. Its effects on Strix Halo will also need fresh verification. The combination of kernel evolution and integrated chips bears watching.

Editorial Opinion

Looking at short-term impact. The Ubuntu 26.10 release is approaching in October. With the kernel frozen at 7.3, validation by hardware makers and distributors is seen as entering its final stage. Users of Strix Halo systems are assessed as gaining NUMA and latency improvements without regressions. In three to six months, Strix Halo is likely to solidify as an option for new desktop Linux deployments.

Taking a long-term view. NUMA and memory-management improvements are assessed as showing spillover from server to client. The unified large-memory design pairs well with inference and creative workloads. In one to three years, kernel efficiency gains are seen as supporting wider adoption of integrated chips. If power and responsiveness improvements accumulate, they could change the foundation of portable high-performance computing.

Note questions from the editors. This testing is based on a single machine and limited workloads. It remains unclear whether the lack of difference in real-world workloads reflects kernel maturity or bias in test selection. The mechanism behind gains under lightweight conditions like DDNet also remains unverified. Are hardware support maturity and coverage of test conditions now the question?

References

Frequently Asked Questions

How much did Strix Halo performance improve with Linux 7.3?
Overall, improvements remained modest. Radeon 8060S graphics were largely flat, with a small gain in DDNet. NUMA measurements with Stress-NG improved substantially, and network latency fell slightly. No clear differences were confirmed across dozens of real-world workloads.
Which kernel will Ubuntu 26.10 use?
During development it switched from Linux 7.2 to Linux 7.3. The October final release is expected to ship Linux 7.3. The previous Ubuntu 26.04 LTS uses Linux 7.0, so 26.10 is expected to benefit from the generational kernel difference. This comparison reflects the state just before release.
Should users of Strix Halo systems update?
For performance alone, no dramatic change is expected. However, no regressions were confirmed, and NUMA and latency improvements were gained. Linux 7.3 also includes hardware support and feature enhancements. Moving to the Ubuntu 26.10 final release is meaningful from a stability standpoint.
Source: Phoronix

Comments

← Back to Home