Announcement

Collapse
No announcement yet.

Windows 10 vs. Ubuntu 18.04 LTS Performance On AMD Ryzen 9 3900X

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • ari_linux
    replied
    Originally posted by stormcrow View Post

    18.04 used 4.15.x, but the current actual state of 18.04.2 uses 4.19.x with patches. Ubuntu devs update to more recent kernel long term release versions as they deem necessary.
    Current HWE kernel version is 4.18.
    https://packages.ubuntu.com/source/b...linux-meta-hwe

    Leave a comment:


  • ari_linux
    replied
    Originally posted by rbmorse View Post

    Ubuntu has patched systemd in the 19.04 repository so the Dingo can boot on Ryzen3000/x570 although the installer .iso hasn't been updated. Now that 18.10 is dead, you can change software sources to notify for all upgrades, apt update, then run:

    Code:
    sudo do-release-upgrade
    to upgrade an existing 18.04 to 19.04.

    Probably not useful for testing/comparing purposes as it won't be a "clean" install, but for those of us with Ryzen 3000 series CPUs and a need to run a 5.x series kernel, this will work.
    You just need to install hwe-edge kernel packages.
    https://packages.ubuntu.com/source/b...-meta-hwe-edge

    Leave a comment:


  • Dr. Righteous
    replied
    Originally posted by stormcrow View Post

    18.04 used 4.15.x, but the current actual state of 18.04.2 uses 4.19.x with patches. Ubuntu devs update to more recent kernel long term release versions as they deem necessary.
    Very true; because kernel updates just applied without checking can break software and drivers.
    But, in the case of benchmarking; there is a lot of development and optimizations that happened since v4.19

    Leave a comment:


  • gregzeng
    replied
    Summarising these benchtests, Linux betters Windows 10, by about 8%. This is on the racetrack, if we use benchmarks in isolation, one at a time.
    Real life DESKTOP operating systems are very different. Simultaneous operations are needed. These have different priorities at different times. My quietly running Windows-10 has "only" 2346 threads, running with "only" 208 processes. Often my notebook is running various backup utilities, defragmenting programs, indexing my main operating drives, providing real-time process monitors, etc. At these times it runs about 2514 threads in 216 processes.
    The real time process monitor mainly used are the Windows "gadget" widgets. "Top Process Monitor version 9.4" shows the "running summary. The third party application "Process Lasso" logs the real time performances, and gives priority to my main applications.
    The main advantage to Windows are the many applications & utilities it has, compared to Linux.
    Given the more capable computer hardware we have now, Windows 10 can easily overcome this 8% "handicap". My 2013 model notebook computer still performs ok.
    More details on hardware & operating system:
    Dell XPS-15 L521x notebook computer. i7, Nvidia, 16 GB DDR3, 4TB HDD, TB SSD. Windows-10 Home, Single Language. Build 18932.rs_prelease.190628-1650. Insider Preview version, fast-release cycle.
    Last edited by gregzeng; 07-28-2019, 03:48 AM.

    Leave a comment:


  • F.Ultra
    replied
    Originally posted by tildearrow View Post

    Windows context switch faster than Linux one. This is not good considering Linux usually handles lots of tasks with ease.

    (Or maybe Microsoft C faster than GNU C?)
    Hard to say, at least some years ago the context switch overhead in Linux was way way less than that of Windows but one really shitty thing with all the new Meltdown and Spectre vulnerabilities is that all the fixes increases this overhead quite a bit.

    What also have to be considered here is what the resolution and accuracy of the timers used to measure these benchmarks are. Benchmarking code changes to my own code shows highly irregular results even from CLOCK_MONOTONIC_RAW so I would not be surprised if some of the differences measured between systems here is actually within the error bar of the timer itself.

    Leave a comment:


  • Tshawkins
    replied
    Originally posted by bug77 View Post
    If you used Windows 10 1903, why didn't you pit it against Ubuntu 19.04?
    (I'll see myself out)
    Its actualy a good question, does the kernel in 18.04 have all the spectre patchs in it, 1903 certainly does, plus some other branch prediction vulnerability fixes. The spectre et al fixes are known to slow a kernel down by up to 15% worse case. You need to ensure that you are using a current kernel in head to head testing. I woukd suggest using fedora 30, asmit always has a bang up to date kernel, rarely more than a week old,

    Leave a comment:


  • user556
    replied
    The cooling is a big factor in how fast a Ryzen part runs. Put a bigger cooler on the CPU and it'll run faster in the multi-core tests in particular. The Prism might be good for a stock cooler but it still nerfs any CPU that it's supplied with.

    Leave a comment:


  • HenryM
    replied
    Originally posted by TomasC View Post
    Was the test done with the same CPU, or two different CPUs (but still the same model)? The Ryzen 3xxx CPUs seem to have a large individual variance.
    I think most variance comes from the particular ram-motherboard combo. ram makes a big difference with Ryzen, and some apparently doesn't play nice with AMD.

    Leave a comment:


  • andrei_me
    replied
    Originally posted by schmidtbag View Post
    Also, I'm a big fan of the 2nd to last comparison graph, though as a suggestion:
    I'd like to see it sorted based on the percentage
    Cool idea, it could be sorted by the highest advantage for the left axis going to the highest advantage for the right axis.
    So Windows ordered desc, Linux ordered as

    E.g: starts showing FFmpeg, Selenium ARES 6, Selenium WebXPRT, which Windows wins, and then go to Linux starting with Novabench RAM, Cpuminer, ending with PHPBench and IndigoBench

    Michael

    Leave a comment:


  • smitty3268
    replied
    Originally posted by edwaleni View Post

    Let's see what happens with Windows 1903 on Threadripper, specifically the 2990WX.

    It will be interesting to see if MSFT has fixed the scheduler and their NUMA support to something more current.

    I read that MSFT had been working on new NUMA support for some multi-package Intel CPU's coming out.

    Being Microsoft, they probably decided to do all the NUMA changes at once.
    The NUMA bug affecting Threadripper is still present in 1903.

    Leave a comment:

Working...
X