Linux vs. Windows results

Artem S. Tashkinov's Avatar

Artem S. Tashkinov

10 Jun, 2017 07:28 AM


I'm curious why Linux and Windows results differ by quite a large margin:

This is the same PC. Both tests have been run on a completely idle PC with next to zero background applications aside from system services. Windows performance profile was set to High Performance. I even changed Windows geekbench process priority to realtime - all to no avail.

Do you really use the same compiler for these two OS'es? If not, do you intend to? Otherwise your results between these two desktop OSes cannot be directly compared.


  1. 1 Posted by Anonymous on 23 Sep, 2017 12:49 AM

    Anonymous's Avatar

    I'm also interested.

  2. 2 Posted by Colin on 27 Sep, 2017 07:17 PM

    Colin's Avatar


    Thanks for letting us know about this, and I apologize for the delayed response. The deviation in scores for those two configurations are slightly larger than normal, but not atypical for the difference between two runs on a given machine. Some specific workload scores, mainly the HTML5 DOM score, show an unexpected discrepancy which could be the result of a hardware issue or an intermittent background process, though with two results it is difficult to draw firm conclusions about its cause.

    Linux and Windows do not use the same compiler. Some details on the compilers used for each platform are available via our page on the individual Geekbench 4 CPU workloads. Currently, there are no plans to change how Geekbench is compiled on a particular platform. Despite the different compilers, Geekbench should generate similar scores for the same hardware on average, though scores may fluctuate slightly.

    Primate Labs Inc.

  3. 3 Posted by mleise on 09 Apr, 2018 04:14 PM

    mleise's Avatar

    I am seeing the same effect for a 2013 notebook. HTML5 DOM Multi-Core looks like it is running single-core on Linux. Also the AES, LZMA, raytracing and the memory bandwith test is usually "won" by Windows, while the others are "won" by Linux.
    I believe most of it can be attributed to different compilers (at least in the single-core tests), while the HTML5 DOM discrepancy looks like a problem with threading support in the used library.

  4. 4 Posted by Colin on 10 Apr, 2018 02:48 PM

    Colin's Avatar

    Hi mleise,

    Thanks for letting us know about this. I'm not certain what would be causing your multi-core HTML5 DOM score on Linux to be similar to the single-core value. I've passed this information along to my team and will let you know as soon as I can if we can provide more information or an update to address this issue.

    Similarly, regarding the differences between your Windows and Linux runs of Geekbench, we're continuously investigating any potential issues which could cause scores to differ between operating systems, and I'll take a closer look based on your information.

    All the best,
    Primate Labs Inc.

  5. 5 Posted by b0b on 29 Apr, 2019 10:31 AM

    b0b's Avatar

    Look at this huge single-core difference for the Core i7-8850H between Linux (5817) and Windows (5233):

    Confirmed by pages and pages of Linux score before the first Windows score for the top chart list for this CPU:

    Can we conclude Geekbench is not really suitable to compare scores for the same machine running different OSes ?

  6. Support Staff 6 Posted by John on 30 Apr, 2019 03:50 AM

    John's Avatar

    From the results you posted it looks like the CPU is able to hit higher frequencies under Linux than Windows (4.3GHz vs 4.2GHz). If one OS enables the processor to run at higher frequencies (and for potentially longer periods of time) I would expect that to be reflected in benchmark results.

  7. 7 Posted by b0b on 01 May, 2019 01:38 PM

    b0b's Avatar

    Interesting. How do you know only 4.2 GHz was reached for the single core benchmark on Windows ? Have you looked to more detailed data of the benchmark report that is not displayed in the web UI ? Is there a way for users to check that in some way ?

  8. 8 Posted by Artem S. Tashki... on 15 Jun, 2019 08:54 PM

    Artem S. Tashkinov's Avatar

    The discrepancy between Linux and Windows in the most recent GeekBench v4.3.4 has increased even further.

    It's high time you guys compiled it using the same compiler.

  9. 9 Posted by Artem S. Tashki... on 24 Jun, 2019 08:53 PM

    Artem S. Tashkinov's Avatar
  10. 10 Posted by joe9000 on 02 Jul, 2019 07:34 AM

    joe9000's Avatar

    John is likely correct. The dynamic power management throttling cores is a big unknown.

    Throttling is the default mode of operation these days. Particularly true for stock coolers and laptops.

    Without the clock rates of each core being recorded by each test as it runs, the amount of throttling can only be guessed at.

  11. 11 Posted by joe9000 on 02 Jul, 2019 06:31 PM

    joe9000's Avatar

    I should have also have added that the 13% can be attributed to the higher percentages of desktops vs laptops that Linux installs will have. Desktops will have better cooling so won't throttle as strongly as the laptops do.

  12. 12 Posted by joe9000 on 02 Jul, 2019 06:46 PM

    joe9000's Avatar

    Historically, the CPU clockrate was a known fixed value when running the benchmark. It has been a while since that was true, benchmarking tools just haven't kept up.

    Basically, benchmarking done without recording the associated clockrates lose much value.

  13. Support Staff 13 Posted by John on 03 Jul, 2019 07:23 PM

    John's Avatar

    For the Ryzen 3600 results, this is another situation where the clock speed is higher under Linux than Windows. For the Linux result the system is running at ~4.4GHz, while for the Windows result the system is running at ~4.2GHz.

    Switching to a different compiler won't solve this problem.

  14. 14 Posted by joe9000 on 03 Jul, 2019 09:01 PM

    joe9000's Avatar

    As you've indicated with the "~", those clock rates are your best guess at explaining the difference. If the actual dynamic clock rates had been recorded then that component of the score wouldn't be guesswork any longer.

  15. Support Staff 15 Posted by John on 04 Jul, 2019 04:48 AM

    John's Avatar


    Geekbench 4 records the maximum clockrate the system can achieve before running the workloads. You can see this data by appending .gb4 to the end of any Geekbench 4 result:

    The ~ was meant to indicate the approximate maximums -- it's much easier to understand ~4.2GHz than 4149MHz.

  16. 16 Posted by joe9000 on 04 Jul, 2019 05:17 AM

    joe9000's Avatar

    Cool! Didn't know about that. The Processor Frequency Mean could be what I'm looking for. Just need it added alongside each score.

  17. 17 Posted by Marco.Leise on 04 Jul, 2019 05:18 AM

    Marco.Leise's Avatar


  18. 18 Posted by joe9000 on 04 Jul, 2019 05:36 AM

    joe9000's Avatar

    There is something missing there though. I can't see a distinction between single-core workloads and multi-core workloads. Clearly, on some systems at least, the clock rate will come out quite different between the two.

    This would need resolved to produce score-frequency pairs for both single-core and multi-core.

  19. 19 Posted by Vitali on 22 Jun, 2020 10:49 PM

    Vitali's Avatar

    There is a very large difference between Windows and Linux multi-core scores on AMD 3990X CPU. In my case it's 26052 on Windows and 36217 on Linux. This is 28% difference. Is it due to the fact that Windows can't properly utilize more than 64 threads?

  20. 20 Posted by Artem S. Tashki... on 26 Jun, 2020 09:17 AM

    Artem S. Tashkinov's Avatar


    That might be the case and also it might be due to the power plan you're using in Windows.

    Also check AnandTech Threadripper 3990x review (can't give you the link because it's been rejected).

  21. 21 Posted by Pete on 03 Aug, 2020 01:00 AM

    Pete's Avatar


    Windows 10 can use way more than 64 logical processors, but the application need to be compiled with that in mind, and it needs to understand processor group. If Geekbench doesn't do that, you won't get anywhere near a valid result.

Reply to this discussion

Internal reply

Formatting help / Preview (switch to plain text) No formatting (switch to Markdown)

Attaching KB article:


Attached Files

You can attach files up to 10MB

If you don't have an account yet, we need to confirm you're human and not a machine trying to post spam.

Keyboard shortcuts


? Show this help
ESC Blurs the current field

Comment Form

r Focus the comment reply box
^ + ↩ Submit the comment

You can use Command ⌘ instead of Control ^ on Mac