macOS 10.13 High Sierra vs. Ubuntu Linux Performance

Apple this week released macOS 10.13 "High Sierra" as the latest version of its operating system. Of course, curiosity got the best of me so here are benchmarks of macOS 10.12.6, macOS 10.13, Ubuntu 16.04 LTS, and Ubuntu 17.10 from a MacBook Air to see how the performance compares.

Arguably the most interesting change with macOS High Sierra is under-the-hood and that's the automatic migration from an HFS+ file-system to using Apple's new APFS file-system. After decades of use, HFS+ is finally being replaced by this new file-system that brings support for 64-bit inodes, exclusively optimized for flash storage, built-in encryption, supports snapshotting, better data integrity, etc. I already posted some benchmarks showing APFS generally having much better performance than HFS+ while it wil lbe interesting to see how it compares to EXT4 on Ubuntu.

MacOS High Sierra also brings Metal 2 as the new version of the Apple-exclusive graphics API, though that isn't being tested for this article today. As part of their graphics update also comes support for VR devices, external GPUs, and an updated Quartz compositor. OpenGL hasn't received much love in 10.13 and Apple still is not supporting the Vulkan graphics API. Other macOS 10.13 changes include HEVC hardware extensions, kernel security improvements, Siri voice improvements, and other refinements to its various bundled applications.

macOS 10.13 High Sierra was compared to macOS 10.12.6 before loading up the Linux distributions. The macOS builds were making use of their native Xcode compilation stack powered by LLVM/Clang.

Ubuntu 16.04.3 LTS was up next for testing with its Linux 4.10 kernel, GCC 5.4 compiler, Mesa 17.0.7, etc.

For a look ahead at where things are going with Ubuntu 17.10 and then Ubuntu 18.04 LTS, a daily snapshot of 17.10 Artful Aardvark from this week was also benchmarked. Ubuntu 17.10 is shipping with the Linux 4.13 kernel, GCC 7.2.0 code compiler, EXT4 file-system, Mesa 17.2.1, and has transitioned to its GNOME 3.26 desktop environment atop Wayland. A second run of Ubuntu 17.10 was also done when using LLVM Clang 5.0 as the C/C++ code compiler rather than GCC to help isolated differences in performance due to each vendor's default choice as the system code compiler.

The same Apple MacBook Air was used throughout all of the testing and was Haswell era with a Core i5 4250U, 4GB DDR3 memory, 120GB Apple SSD, and HD 5000 series graphics. I'm working on getting access to some other MacBook hardware that's more modern for some Metal vs. OpenGL/Vulkan tests and other comparisons. If I get access to that, there will be follow-up High Sierra tests in the near future.

All of these macOS High Sierra benchmarks and Linux tests were done in a fully-automated and reproducible manner using the open-source Phoronix Test Suite benchmarking software.