Of the many improvements to be found in the in-development Linux 4.17 kernel -- nicely summarized in our Linux 4.17 feature overview -- one of the features I've been anxious the most to begin benchmarking has been the reported power management improvements. Here are my initial power/performance tests of Linux 4.17 that for some systems is seeing a measurable drop in power usage, even in some cases under load while without sacrificing the performance.
Linus Torvalds is expected by the end of the day to release Linux 4.17-rc1, thereby marking the end of the two-week merge window that saw a lot of changes and new features land for Linux 4.17. Here is our original feature overview of the changes to be found in this next major release of the Linux kernel, which should premiere as stable by the middle of June.
This past weekend I posted some Linux 3.17 to Linux 4.16 kernel benchmarks using two older Intel hardware platforms and one of the most frequent requests to come in following that article were some AMD benchmarks looking at the Intel Linux performance going a ways back. Here are some test results from Linux 4.4 to 4.16 using an AMD Ryzen Threadripper 1950X system.
Our latest benchmarking of the near-final Linux 4.16 kernel is checking on the performance of two Intel systems going back to the days of Linux 3.17, the oldest kernel that would successfully boot with the Ubuntu 18.04 user-space. Every major kernel release was tested as we see how the Linux kernel performance has evolved on these Haswell and Gulftown systems since October 2014.
Since last week's big release of LLVM 6.0 along with Clang 6.0, I have been carrying out some fresh compiler benchmarks of the previous Clang 5.0 to this new stable release that switches to C++14 by default, among many other changes to LLVM itself and this C/C++ compiler front-end.
A few days back I provided some fresh Linux 4.16 kernel benchmarks compared to recent stable kernel releases while also toggling the KPTI and Retpoline security features on Linux 4.16 Git for seeing the impact of the Spectre and Meltdown mitigation techniques on this latest kernel while using Intel Xeon hardware. For this latest round of tests is a similar comparison while using an AMD EPYC system.
Moving forward the Phoronix Test Suite will be offering Microsoft Windows support that's at near feature parity to the automated, reproducible, open-source benchmarking support we have offered the past decade for Linux as well as BSD, macOS, and Solaris platforms. This is brand new, rewritten Windows support with a focus on Windows 10 x64 and Windows Server 2016.
For those wondering how the Linux 4.16 kernel is performing with regards to the ongoing work around Spectre and Meltdown mitigation in the kernel, here are some fresh benchmark results.
Given the recent performance changes following the Spectre/Meltdown CPU vulnerability mitigation and having just wrapped up some fresh CPU bare metal benchmarks as part of that testing as well as the recent AMD Raven Ridge launch, I've carried out a fresh round this week of benchmarks on various Amazon EC2 on-demand instance types compared to a number of bare metal Intel and AMD processors in looking at how the compute performance compares.
Here are some preliminary benchmarks of the Linux 4.16 development kernel compared to Linux 4.15 stable on five different systems.
After the lengthy Linux 4.15 kernel cycle, the past two weeks have marked the Linux 4.16 merge window. Yet again it's been another heavy feature period for the kernel. There is still a lot of mitigation work going on for most CPU architectures surrounding Spectre and also Meltdown, the open-source graphics drivers have continued getting better, various CPU improvements are present, the VirtualBox Guest driver was mainlined, and dozens of other notable changes for Linux 4.16. Take a look.
A few days ago we found out that at the end of 2017 AMD quietly released their AOCC 1.1 C/C++ compiler. AOCC is AMD's compiler succeeding AMD Open64 that existed years ago as their optimized Fortran/C/C++ compiler for past CPU microarchitectures while the "AMD Optimizing C/C++ Compiler" is designed for current-generation Zen processors. Here are benchmarks of the new AMD AOCC 1.1 release compared to GCC 7, GCC 8, Clang 5.0, Clang 6.0, and Clang 7.0 SVN.
Here is a look at how the Linux kernel performance has evolved since the release of Linux 3.17 in October 2014. With all the major kernel releases over the past 3+ years, here is how the performance compares using two very different Intel Gulftown and Haswell systems.
Our latest in benchmarking the Linux 4.15 kernel is seeing how the performance has changed since Linux 4.0 and all subsequent releases on the same system. Here are those tests driven by curiosity, especially in light of the performance changes as a result of KPTI page table isolation and Retpoline additions.
For those wondering how the Intel (U)HD Graphics compare for games and other graphical benchmarks between desktop environments in 2018, here are some fresh benchmarks using GNOME Shell on X.Org/Wayland, KDE Plasma 5, Xfce, Unity 7, and LXDE.
We have looked several times already at the performance impact of Retpoline support in the Linux kernel, but what about building user-space packages with -mindirect-branch=thunk? Here is the performance cost to building some performance tests in user-space with -mindirect-branch=thunk and -mindirect-branch=thunk-inline.
A new Phoronix Premium member was hypothesizing in the forums whether Ubuntu's low-latency kernel would be performing better in the wake of the Kernel Page Table Isolation (KPTI) support in the kernel for fending off the Meltdown CPU vulnerability. With always aiming to deliver on test requests by premium members, I ran some benchmarks of the Ubuntu low-latency vs. generic kernels and I also tossed in the Liquorix kernel for benchmarking too.
At the beginning of January I posted some early LLVM Clang 6.0 benchmarks on AMD EPYC while in this article is comparing the tentative Clang 6.0 performance to that of the in-development GCC 8.0. Both compilers are now into their feature freeze and this testing looked at the performance of generated binaries both for generic x86_64 as well as being tuned for AMD's Zen "znver1" microarchitecture.
Over the past week and a half of running many benchmarks looking at the performance impact of the Linux KPTI and Retpoline patches for Spectre and Meltdown mitigation, one of the most common test requests is some thorough benchmarks on older systems. Why that's important is with older (pre-Westmere) CPUs there isn't PCID (Process Context Identifier) support that's used by KPTI, which helps offset some of the performance loss. So for some test results to share today are two old ThinkPads from the Clarksfield and Penryn days compared to a newer Broadwell ThinkPad in looking at the performance difference.
While the Kernel Page Table Isolation (KPTI) patches were quick to land in the mainline Linux kernel for addressing the Meltdown CPU vulnerability, the "Retpoline" patches are still being worked on as the leading approach on the Linux side for dealing with the Spectre CPU vulnerability. The Retpoline patches are said to have little impact on performance, but here are our benchmarks of these kernel patches for seeing how they affect a variety of AMD and Intel systems.
Yet another one of the avenues we have been exploring with our Linux Page Table Isolation (KPTI) testing has been looking at any impact of this security feature in the wake of the Meltdown vulnerability when testing with an older Linux Long Term Support (LTS) release. In particular, when using a kernel prior to the PCID (Process Context Identifier) support in the Linux kernel that is used to lessen the impact of KPTI.
Continuing on with our Linux Kernel Page Table Isolation (KPTI) performance testing are some benchmark results when running tests within a virtual machine on Xeon class hardware.
Over the past day you've likely heard lots of hysteria about a yet-to-be-fully-disclosed vulnerability that appears to affect at least several generations of Intel CPUs and affects not only Linux but also Windows and macOS. The Intel CPU issue comes down to leaking information about the kernel memory to user-space, but the full scope isn't public yet until the bug's embargo, but it's expected to be a doozy in the data center / cloud deployments. Due to the amount of interest in this issue, here are benchmarks of a patched kernel showing the performance impact of the page table isolation patches.
With LLVM 6.0 being branched this week and that marking the end of feature development on this next compiler update before its stable debut in February, here are some benchmarks of the very latest LLVM Clang 6.0 compiler on AMD's EPYC 7601 32-core / 64-thread processor as we see how well the AMD Zen "znver1" tuning is working out.
I've been working on some AMD EPYC virtualization tests on and off the past few weeks. For your viewing before ending out the year are some initial VirtualBox vs. Linux KVM benchmarks for seeing how the guest VM performance compares.
For those wondering how the LLVM Clang vs. GCC C/C++ compiler performance is comparing as we end out 2017, here are some recent benchmarks using the latest Clang 6.0 SVN and GCC 8.0.0 compilers in a range of benchmarks.
This week marked the release of the long-awaited POCL 1.0 release candidate. For the uninformed POCL, or the Portable Computing Language, is a portable implementation of OpenCL 1.2~2.0 that can run on CPUs with its LLVM code generation and has also seen back-ends for its OpenCL implementation atop AMD HSA and even NVIDIA CUDA. I've been trying out POCL 1.0-RC1 on various Intel and AMD CPUs.
The merge window is effectively over for Linux 4.15 with it being the 14th day of the process, although 4.15-rc1 might not end up coming out today due to Linus Torvalds' traveling around the US Thanksgiving holiday. But with Torvalds tending to not approve major last minute additions to new kernels, we don't anticipate any last minute surprises and therefore here is our feature overview of the changes and new features of Linux 4.15. This is arguably the most exciting and feature-packed kernel update ever.
Given the release of a new Blender "Barbershop" benchmark file, I decided to test this new scene plus the other benchmark files with Blender 2.79 on a variety of Intel/AMD CPUs for some fresh results of how various newer CPUs compare for this open-source 3D modeling software.
Our latest Linux file-system benchmarking is looking at the performance of the mainline Btrfs, EXT4, F2FS, and XFS file-systems on the Linux 4.14 kernel compared to 4.13 and 4.12.
592 software articles published on Phoronix.