NVIDIA Engineer Submitted Linux Scheduler Improvements
The patches optimize SMT performance for Vera CPU cores and aim for inclusion in the upcoming Linux v7.4 kernel.
Updated on Oct. 1, 2026 in Semiconductors
Live Poll
Do you believe software optimizations are key to getting the most out of your hardware?
NVIDIA engineer Andrea Righi has submitted the seventh revision of scheduler patches for Linux, designed to optimize performance on Vera CPU cores. The proposed updates are research-stage improvements currently pending integration into the mainline kernel.
Why it matters
This development aims to improve SMT throughput on hardware featuring Olympus cores, potentially increasing efficiency for scientific and high-performance computing workloads. The submission marks a refined approach to kernel-level scheduling for asymmetric hardware architectures.
The patch series consists of fewer than 100 lines of code and provides a 3 to 6 percent performance improvement in tested workloads. It introduces a new kernel parameter, sched_smt_asym_packing, to manage asymmetric packing within the SMT scheduling domain.
The players
NVIDIA
A designer of graphics processing units and high-performance CPUs including the Vera architecture with Olympus cores.
Andrea Righi
An NVIDIA engineer responsible for submitting the kernel patches for Vera CPU core optimization.
The details
The implementation optimizes Simultaneous Multithreading (SMT) performance by modifying how the kernel handles idle selection paths on Olympus cores. By honoring asymmetric SMT priority, the scheduler can better distribute tasks to the most efficient compute resources. The new parameter allows developers to force asymmetric packing, ensuring that threads are allocated to siblings that provide the highest throughput for the specific workload.
Timeline
August 2026: Author initiated work on the initial Linux scheduler improvements.
October 1, 2026: The seventh revision of the patches was submitted to the Linux kernel mailing list.
The Tech Race
This work follows the standard Linux kernel development process, where architectural optimizations are iteratively reviewed before mainline integration. The performance gains place it in the ongoing effort to refine thread scheduling for proprietary chip architectures within open-source operating systems.
These improvements will benefit users running high-performance computing tasks on Vera-based systems once the code is merged into the mainline kernel. The update requires no user-side installation beyond running a compatible future version of the Linux kernel.
The takeaway
The performance gains demonstrated in OpenBLAS and NVPL suggest a meaningful optimization for multi-threaded workloads. Observers should track the Linux kernel mailing list for the inclusion of these patches in the upcoming version 7.4 development cycle.
What happens next
The patches are currently under review with the intent to be integrated into the Linux v7.4 kernel release.
Further reading
For more on architectural optimizations, visit the Semiconductors section.
Source note: This article includes information reported by Phoronix.
Live Poll
Do you believe software optimizations are key to getting the most out of your hardware?
