Linux Patches Posted To Fix ~2x Performance Drop For CPU Workloads On NVIDIA Vera Rubin
([NVIDIA] 54 Minutes Ago
NVIDIA Vera Rubin SMT Issue)
- Reference: 0001623070
- News link: https://www.phoronix.com/news/NVIDIA-Vera-Rubin-SMT-Sched-Fix
- Source link:
An important set of Linux scheduler patches were posted for review on Thursday for improving the SMT-aware asymmetric CPU capacity handling. These patches to improve the Linux kernel scheduler around CPU Simultaneous Multi-Threading (SMT) is needed after NVIDIA engineers discovered up to a ~2x performance drop for CPU-intensive workloads on their upcoming Vera Rubin platform.
NVIDIA Linux engineer Andrea Righi has been working to improve the Linux kernel's asymmetric CPU capacity scheduling with better SMT awareness. The adapted behavior with the proposed patches is that when SMT is active, prefer fully-idle cores and avoiding treating partially-idle SMT siblings as full-capacity targets.
The key takeaway described by Righi in the mailing list patch series is:
"This patch set has been tested on the new NVIDIA Vera Rubin platform, where SMT is enabled and the firmware exposes small frequency variations (+/-~5%) as differences in CPU capacity, resulting in SD_ASYM_CPUCAPACITY being set.
Without these patches, performance can drop up to ~2x with CPU-intensive workloads, because the SD_ASYM_CPUCAPACITY idle selection policy does not account for busy SMT siblings.
Alternative approaches have been evaluated, such as equalizing CPU capacities, either by exposing uniform values via firmware (ACPI/CPPC) or normalizing them in the kernel by grouping CPUs within a small capacity window (+-5%), or enabling asympacking [3].
However, adding SMT awareness to SD_ASYM_CPUCAPACITY has shown better results so far. Improving this policy also seems worthwhile in general, as other platforms in the future may enable SMT with asymmetric CPU topologies."
Good they are getting this sorted out now before the NVIDIA Vera CPUs work their way out to data centers at scale.
[1]The patches for this SMT-aware asymmetric CPU capacity scheduler work is out for review on the LKML and will hopefully be ready for the mainline kernel in an upcoming merge window like Linux v7.1.
[1] https://lore.kernel.org/lkml/20260326151211.1862600-1-arighi@nvidia.com/
NVIDIA Linux engineer Andrea Righi has been working to improve the Linux kernel's asymmetric CPU capacity scheduling with better SMT awareness. The adapted behavior with the proposed patches is that when SMT is active, prefer fully-idle cores and avoiding treating partially-idle SMT siblings as full-capacity targets.
The key takeaway described by Righi in the mailing list patch series is:
"This patch set has been tested on the new NVIDIA Vera Rubin platform, where SMT is enabled and the firmware exposes small frequency variations (+/-~5%) as differences in CPU capacity, resulting in SD_ASYM_CPUCAPACITY being set.
Without these patches, performance can drop up to ~2x with CPU-intensive workloads, because the SD_ASYM_CPUCAPACITY idle selection policy does not account for busy SMT siblings.
Alternative approaches have been evaluated, such as equalizing CPU capacities, either by exposing uniform values via firmware (ACPI/CPPC) or normalizing them in the kernel by grouping CPUs within a small capacity window (+-5%), or enabling asympacking [3].
However, adding SMT awareness to SD_ASYM_CPUCAPACITY has shown better results so far. Improving this policy also seems worthwhile in general, as other platforms in the future may enable SMT with asymmetric CPU topologies."
Good they are getting this sorted out now before the NVIDIA Vera CPUs work their way out to data centers at scale.
[1]The patches for this SMT-aware asymmetric CPU capacity scheduler work is out for review on the LKML and will hopefully be ready for the mainline kernel in an upcoming merge window like Linux v7.1.
[1] https://lore.kernel.org/lkml/20260326151211.1862600-1-arighi@nvidia.com/