Early Linux 7.3 Kernel Benchmarks Show Some Performance Improvements, Some Regressions
([Software] 2 Hours Ago
5 Comments)
- Reference: 0001653923
- News link: https://www.phoronix.com/review/linux-73-early-benchmarks
- Source link:
Being halfway through the [1]Linux 7.3 kernel merge window and prior to wrapping up and sending back the HP Z4 G6i workstation powered by the [2]Xeon 678X Granite Rapids WS processor, I decided to run some benchmarks to see how Linux 7.3 Git is performing relative to the prior stable kernel series. Here are the first benchmarks of this early state of Linux 7.3 with some performance improvements as well as some regressions.
[3]
From the HP Z4 G6i workstation I carried out benchmarks of Linux 7.0, 7.1, 7.2, and 7.3 Git kernels with the Ubuntu 26.10 daily software stack otherwise being the same. This workstation is powered by the 48-core / 96-thread Intel Xeon 678X GNR-WS processor, 4 x 32GB DDR5-6400 memory, SK Hynix 2TB NVMe SSD, and the NVIDIA RTX PRO 6000 Blackwell graphics card.
[4]
For the purposes of the upstream kernel comparison, the Nouveau driver was in use for the RTX PRO 6000 Blackwell. That included Nouveau working with Zink/NVK on Linux 7.3 just as it was in prior releases... That is important to note with some of the data in this article.
The same standard kernel configuration was used across all tested kernel versions with just assuming the Kconfig defaults on each succeeding kernel release tested. All kernels were built from source using the same compiler. While normally the Ubuntu Mainline Kernel PPA is preferred for easy reproducibility and transparency, with the Ubuntu Kernel PPA builds still failing and being less than reliable in recent months, the local kernel builds were used for this comparison.
Dozens of different workloads were tested across the Linux 7.0 / 7.1 / 7.2 / 7.3 Git (as of 23 August) kernels with no other hardware or software changes between runs. Thanks to HP for having provided the Z4 G6i workstation for some interesting Linux benchmarking over the past month on Phoronix.
[1] https://www.phoronix.com/search/Linux+7.3
[2] https://www.phoronix.com/search/Xeon+678X
[3] https://www.phoronix.com/image-viewer.php?id=linux-73-early-benchmarks&image=linux_73_hp_1_lrg
[4] https://www.phoronix.com/image-viewer.php?id=linux-73-early-benchmarks&image=linux_73_hp_2_lrg
[3]
From the HP Z4 G6i workstation I carried out benchmarks of Linux 7.0, 7.1, 7.2, and 7.3 Git kernels with the Ubuntu 26.10 daily software stack otherwise being the same. This workstation is powered by the 48-core / 96-thread Intel Xeon 678X GNR-WS processor, 4 x 32GB DDR5-6400 memory, SK Hynix 2TB NVMe SSD, and the NVIDIA RTX PRO 6000 Blackwell graphics card.
[4]
For the purposes of the upstream kernel comparison, the Nouveau driver was in use for the RTX PRO 6000 Blackwell. That included Nouveau working with Zink/NVK on Linux 7.3 just as it was in prior releases... That is important to note with some of the data in this article.
The same standard kernel configuration was used across all tested kernel versions with just assuming the Kconfig defaults on each succeeding kernel release tested. All kernels were built from source using the same compiler. While normally the Ubuntu Mainline Kernel PPA is preferred for easy reproducibility and transparency, with the Ubuntu Kernel PPA builds still failing and being less than reliable in recent months, the local kernel builds were used for this comparison.
Dozens of different workloads were tested across the Linux 7.0 / 7.1 / 7.2 / 7.3 Git (as of 23 August) kernels with no other hardware or software changes between runs. Thanks to HP for having provided the Z4 G6i workstation for some interesting Linux benchmarking over the past month on Phoronix.
[1] https://www.phoronix.com/search/Linux+7.3
[2] https://www.phoronix.com/search/Xeon+678X
[3] https://www.phoronix.com/image-viewer.php?id=linux-73-early-benchmarks&image=linux_73_hp_1_lrg
[4] https://www.phoronix.com/image-viewer.php?id=linux-73-early-benchmarks&image=linux_73_hp_2_lrg