Introduction
According to the Phoronix Test Suite website, PTS
is the most comprehensive testing and benchmarking platform available that provides an extensible framework for which new tests can be easily added. The software is designed to effectively carry out both qualitative and quantitative benchmarks in a clean, reproducible, and easy-to-use manner. The Phoronix Test Suite can be used for simply comparing your computer's performance with your friends and colleagues or can be used within your organization for internal quality assurance purposes, hardware validation, and continuous integration / performance management.
Among the strengths of Phoronix Test Suite are its ease of use, the large selection of tests available, the ability to automatically process results and, present them graphically. The software also integrates the OpenBenchmarking.org infrastructure, where users can review information on available tests and test suites, upload results to share these results with other users, and access previously uploaded results and the results of other users for comparison of results.
To benchmark the CachyOS kernels, Phoronix Test Suite was installed and configured on CachyOS 260308 as described in Benchmarking with Phoronix Test Suite: Installation and Configuration, which although not relevant for this benchmark, allows for sharing PTS resources among distributions installed on a multi-boot system.
The tests were performed on an installation of CachyOS 260308 with updates current as of May 16 2026. The hardware was a Lenove Legion 5i Pro (16ITH6) with an Intel Core-i7 11800H CPU, an integrated Intel UHD-630 integrated graphics card, a discrete Nvidia GeForce GTX-3050 Mobile graphics card, 500 GB SK Hynix PCIe NVMe SSD, 1 TB Samsung Evo 980 PRO PCIe NVMe SSD, and 24 GB of RAM. The details of the hardware, according to Phoronix are shown in the following image, althoug the Phoronix hardware detection does not recognize a second 8GB Samsung RAM module.
The Details of the System Configuration
The benchmark tests were collected in a custom PTS test suite, the creation of which was described in Benchmarking with Phoronix Test Suite: Installation and Configuration. The selected tests were the nine tests (the number of tests according to the Phoronixarticle's results summary, not as identified by the Phoronix output in the following listing) that produced results with the largest variation in performance of the kernels in a Phornix benchmark compariosn of the Liquorix kernel against the upstream kernel. Using the same tests as those used by the Phornix site to compare the performance of different kernels is reasonable as the goal of the current test is to compare the performance of the various kernels offered by CachyOS. The tests in the current test was limited to a subset of those used by Phoronix in its comparison of the Liquorix kernel in order to reduce the test time. Readers with ample time and are interested in perfroming a comparison of the kernels can use the article referenced above as a guide for installing Phoronix Test Suite and creating a custom test suite that includes all of the tests in the Phoronix article.
The determination of the best performing CachyOS for overall system responsiveness was based on the tests that were filtered by the Phoronix Test Suite Result Viewer to not include outliers or those that that were "noisy" -- a term used by the Phoronix Test Suite Result Viewer.
Custom Test Suite
The custom test suite consisted of the individual tests listed in the output of phoronix-test-suite info custom-responsiveness, as shown in the following listing. The details of each of these tests can be viewed at OpenBenchmarking.org/tests.
░▒▓ │ ~/.config/nvchad
▓▒░─────────────────────────────────────────────────────────────────────────────────░▒▓ at 09:38:54 PM ▓▒░─╮
❯ sudo phoronix-test-suite info custom-responsiveness ─╯
[sudo] password for brook:
Phoronix Test Suite v10.8.6
custom-responsiveness
Suite Description: Kernel benchmark comparison based on the ten individual tests with the largest differences
between tested kernels in https://www.phoronix.com/news/Linux-6.12-Liquorix-Performanc
Run Identifier: local/custom-responsiveness
Suite Version: 1.0.0
Maintainer: ORDINATECHNIC
Status:
Suite Type: System
Unique Tests: 3
Contained Tests:
Stress-NG Test: Semaphores
Stress-NG Test: Socket Activity
Stress-NG Test: Context Switching
srsRAN Project Test: PUSCH Processor Benchmark, Throughput Total
srsRAN Project Test: PUSCH Processor Benchmark, Throughput Thread
Hackbench Count: 16 - Type: Process
Hackbench Count: 16 - Type: Thread
Hackbench Count: 32 - Type: Process
Hackbench Count: 32 - Type: Thread
16 Tests / 3 Unique Tests
░▒▓ │ ~/.config/nvchad ▓▒░─────────────────────────────────────────────────────────────────────░▒▓ took 5s │
at 09:40:55 PM ▓▒░─╮
❯ ─╯
Tested Kernels
Tested Kernel Properties
| Kernel |
LTO |
PGO |
CPU Process Scheduler |
| cachyos |
X |
X |
EEVDF |
| cachyos-bmq |
|
|
BMQ |
| cachyos-bmq-lto |
X |
|
BMQ |
| cachyos-bore |
|
|
BORE |
| cachyos-bore-lto |
X |
|
BORE |
Test Execution
The test runs were executed after booting into CachyOS and selecting the kernel to be executed from the GRUB menu. No other applications were running. The WiFi interface was connected and the same default services were enabled for each run. In each case the, the test suite was executed using the Phoronix Test Suite command
sudo phoronix-test-suite benchmark custom-responsiveness
where
custom-responsiveness is the custom test suite created as described in
Benchmarking with Phoronix Test Suite: Installation and Configuration.
The following set of images show some of the output when executing the command.
The Output Produced by PTSDuring Test Execution
Results
The benchmark results were surprising in that the default kernel performed better than the kernels with alternative CPU process schedulers. This was true even in the case of the alternative scheduler kernels that are -- like the default kernel -- optimized with LTO. The following set of images from Phoronix Test Suite Results Viewer summarizes the test results.
Results Overview
Results Overview
All five kernels tested had similar results for most tests except the Context Switching component of Stress-NG. In that test, cachyos, cachyos-bore, and cachyos-bore-lto, clearly performed better. cachyos-bmq and cachyos-bmq-lto performed better than the others in the Semaphores component of Stress-NG.
-
Image 1
-
Result overview with noisy and incomplete test data filtered out. Performance of all kernels was similar except for the Stress-NG:Socket Activity test and the Stress-NG:Context Switching test. The cachyos, cachyos-bore-lto, and cachyos-bore kernels performed significantly better than the other two kernels in the Stress-NG: Context Switching test and to a lesser extent better than the other kernels in the Stress-NG: Socket Activity test. These kernels also performed better, but only slightly, than the other two in the Hackbench: 32-Process test.
-
Image 2
-
Presents a summary of the fliteed results. It provides the average of all runs for each test for each tested kerenl.
-
Image 3
-
Corresponds to Image 1 but with unfiltered results.
-
Image 4
-
Corresponds to Imge 2 but with unfiltered results.
Overall Winner
The tests determined that the kernel with the best performance was the default CachyOS kernel cachyos, as illustrated in the image below, which shows the geometric mean of all tests for each kernel. This presentation of the results also shows that the two variants using the BORE process scheduler performed almost as well as the cachyos kernel. These two kernels had the second and third largest geometric mean values, which were very close to the value of the best performing kernel. The two kernels that use the BMQ kernels, however, performed significantly worse than any of the other kernels.
The Geometric Mean of all Results
The cachyps kernel finished first in more tests than any of the other kernels, followed by cachyos-bore-lto. In tests where noisy and incomplete measurements were discarded, it had the best performance in 60% of the tests, followed by cachyos-bore-lto which came in first in 40% of the tests. In tests where none of the results were discarded, both cachyos and cachyos-bore-lto came in first in 37% of the tests, but in this case cachyos-bmq also came in first in 25% of the tests. This is shown in the following set of images.
The Default CachyOS Kernel, cachyos, Finished First In More Tests
cachyos finished first in the most tests. The first image in this set is
based on tests where incomplete and noisy data was removed before producing the graph, and the second image
is based on all tests.
Not only did the cachyos and cachyos-bore-lto, come in first, in all of the tests (filtered results), they did not come in last in any of the tests. The following set of images show that in the filtered results (first image of the set), cachyos-bmq notably had the worst performance in 80% of the tests, with cachyos-bmq-lto having the worst performance in the remaining 20% of the tests. When test results are unfiltered cachyos-bmq also had the worst performance in most of the tests at 62% (second image in the set). The notable feature of this reduction of the test results is that all of the kernels performed the worst in at least one of the tests, except the cachyoskernel.
The Default CachyOS Kernel, cachyos, Finished First In More Tests
cachyos finished first in the most tests. The first image in this set is
based on tests where incomplete and noisy data was removed before producing the graph, and the second image
is based on all tests.
Individual Test Results
The average of measurements of each individual test for each kernel is presented in the following set of images. It should be noted that the noisy and incomplete test data mentioned above that PTS`''` offered to filter out when preparing graphical presentation of test results was limited to the Stress-NG:Socket Activity test. The results of this test are shown in Slide 6 and Slide 7, where the former is of filtered measurements and the latter is of unfiltered measurements.
Individual Test Results
Conclusion
These tests, as noted above were those used by Phoronix to make comparisons of kernel performance. However, in this evaluation only a subset of those used by Phoronix were utlized, although the ones selected here were the ones that produced the largest variance in measurements between the kernels in the Phoronix test.
It would seem that, based on these tests, the default CachyOS kernel, cachyos is the best performing kernel of the five tested in this article. It
-
had the best performance in the most tests
-
did not have the worst performance in any of the tests
-
had the highest geometric mean of test results
It would also seem that, based on these tests that the cachyos-bmq and cachyos-bmq-lto kernels are the worst performing kernels, having
-
the worst performanc in more tests than any of the other kernels
-
the smallest geomtric mean values of any of the kernels
It is probably safe to conclude that the cachyos kernel's LTO and PGO optimizations make a difference in its performance. It may also be safe to conclude that the Linux kernel's default process scheduler for normal user space porcesses, EEVDF performs better than the BMQ, and to a lesser extent, the BORE process schedulers.
