AMD Ryzen 7 7840U Windows 11 vs. Linux CPU Performance

[smitty3268]

When you do benchmarking you want your system behavior to be predictable and tests to be repeatable.

If you are trying to compare and evaluate hardware, then sure. But Phoronix generally only does Windows tests to compare it to Linux, not keeping it static to compare to other runs of Windows using different hardware.

And if you are comparing it to other OS's, then I think it's fair to test what end users will actually see - which means less predictability on Windows. That's fair since he uses defaults on linux too, and gets all sorts of complaints constantly that he's not optimizing linux properly the way he needs to in order to be "fair". No, testing the defaults is the only fair thing to do.

Besides, generally speaking Phoronix's automated benchmarks are pretty garbage, and Michael doesn't dig into results that make no sense to figure out what's going on, be that an error in the testing or something more interesting being exposed. If you want to complain about how Michael tests, that's by far the more important thing to fix first. Optimizing OS defaults is way down the list of improvements.

[andre30correia]

As i have said in the past, I care more about stability than raw performance and this foes for most things, from computers to cars to power tools.

I used to do consulting for video work and one of my favorite tests for testing hardware and software was this long, complex, 4k encode, with lots of filters, both GPU and CPU powered and a couple of audio tracks; this project was done using Shotcut. This specific encode would take over 9 hours if done with no GPU acceleration on a Ryzen 1600, with GPU acceleration of some of the filters, it would drop to about 5.5 hours. The performance different between Windows 10 and Ubuntu was negligible.

What mattered was stability, with projects of this type, I found both Windows and Ubuntu would sometimes either lock up, Shotcut would crash outright or the system would reboot. I looked at all the usual suspects, ram, power supply, cooling, etc, but the thing was it was never consistent, sometimes the projects would finish fine and other times i would lose hours worth of work.

I finally bought a cheap 4790 based Xeon and a cheap Gigabyte board, with the same ram, power supply and drives, the system was about 20% slower in CPU only tasks but is was rock solid stable. Swapping out the GTX960 for a cheap Fermi based Quadro and I could finish jobs like this one after the other, processing video for days completely reliably.

Fast forward to today, and as I mentioned in March i lost my job. As a way of keeping myself busy and improving my skill set, I started a couple of podcasts, one politically themed and one sports themed and I have been recording and editing audio with Audacity. I have also put up a few website related to these podcasts that feature JS powered AI, this is meant more as a learning exercise for myself and to show employers some of the skills i bring to the table.

If you want to get the real measure of a distro, you need to use it for real work, stuff that takes hours, taxes the system and is critical, stuff were you can't afford to lose data.

I have found that most of the distros available are not reliable, for one reason or another. some of the issues that i have encountered are:

1) Audacity failing to launch if installed via package manager even though the appimage works fine.

2) Audacity stops working after a while, with no ability to save your work for extended periods of time, then if you wait long enough it comes back to life and you can save your recording. This one is very nerve-racking

3) Pluma just stops working. I use Pluma for HTML, CSS, JS and Python, and sometimes it will just stop accepting inputs, such as the return key or backspace and it will stop allowing you to save your work.

4) I use a laptop that has a 17" 1600x900 monitor and I hook it up to a 50" 1080P TV that i use as a monitor. Windows 10 handles this setup properly and without a hiccup, regardless of whether I mirror the displays, or extend them and set one or the other as the primary and it also handles it properly when i shut down the TV to pull the HDMI out and switch to using just the laptop monitor.

I have tried Kubuntu, Ubuntu Cinnamon, Ubuntu, PoP OS, Mint Mate, Mint Proper, Mint Debian, Fedora Workstation, Fedora Mate, Fedora KDE, Fedora Cinnamon, PC Linux OS, MX Linux (both editions), Manjaro (all of them), every variant of Gecko Linux (which I normally like), every variant of Spiral Linux, if you name a distro, I tried it, none of them can properly handle this monitor setup.

The only one that handles it properly is the latest edition of Ubuntu Mate, this wasn't the case before, it used to have problems as well, but this latest editions based on Lunar Lobster is the only one that gives me the Windows 10 experience of dependability.

Which brings me to the point, it doesn't matter if Linux is a bit faster in this benchmark or that one. What matter is can i rely on it, does it behave as expected, can i count on it to finish a big project? If the answer is No, then it does me no good.

Maybe your laptop is not compatible with Linux, i have no such problems with mine

[Sonadow]

It seems you forget about other tests where Linux eats windows for breakfast. Now ask yourself who is brainwashed fanboy? Furthermore, as a fanboys you ignored the fact windows isn't patched for the latest vulnerability, so it cheats in performance. It also depends on test choices, but such things are obvious for most people..

Windows has a driver for hardware X used by > 60% of the world's PC users.
Linux has no driver and no alternative for hardware X.


Windows has a software package for Enterprise Software Y used by easily 70% of businesses.
Linux has no such application and no compatible alternative for Enterprise Software Y


Windows eats Linux for breakfast. Period.

[xAlt7x]

I guess those performance regressions could be easily mitigated with power-profiles-daemon (or TLP v1.6 and newer)
With power-profiles-daemon on laptop I can automate profile switching using udev rule

if [[ $(powerprofilesctl | grep balanced) ]] ; then printf 'SUBSYSTEM=="power_supply",ENV{POWER_SUPPLY_ONLINE}=="1",RUN+="/bin/sh -c '\''if [ $(powerprofilesctl | grep performance) ] ; then powerprofilesctl set performance ; else powerprofilesctl set balanced; fi'\''"\nSUBSYSTEM=="power_supply",ENV{POWER_SUPPLY_ONLINE}=="0",RUN+="/bin/sh -c '\''if [ $(powerprofilesctl get) != power-saver ] ; then powerprofilesctl set balanced ; fi'\''"\n' | sudo tee /etc/udev/rules.d/power-profiles-daemon--laptop.rules ; fi

and test it without reboot

sudo udevadm control --reload-rules  && sudo udevadm trigger --subsystem-match=power_supply

P.S. Downside of this approach is that the fan will activate more often.

[Sonadow]

As i have said in the past, I care more about stability than raw performance and this foes for most things, from computers to cars to power tools.

I used to do consulting for video work and one of my favorite tests for testing hardware and software was this long, complex, 4k encode, with lots of filters, both GPU and CPU powered and a couple of audio tracks; this project was done using Shotcut. This specific encode would take over 9 hours if done with no GPU acceleration on a Ryzen 1600, with GPU acceleration of some of the filters, it would drop to about 5.5 hours. The performance different between Windows 10 and Ubuntu was negligible.

What mattered was stability, with projects of this type, I found both Windows and Ubuntu would sometimes either lock up, Shotcut would crash outright or the system would reboot. I looked at all the usual suspects, ram, power supply, cooling, etc, but the thing was it was never consistent, sometimes the projects would finish fine and other times i would lose hours worth of work.

I finally bought a cheap 4790 based Xeon and a cheap Gigabyte board, with the same ram, power supply and drives, the system was about 20% slower in CPU only tasks but is was rock solid stable. Swapping out the GTX960 for a cheap Fermi based Quadro and I could finish jobs like this one after the other, processing video for days completely reliably.

Fast forward to today, and as I mentioned in March i lost my job. As a way of keeping myself busy and improving my skill set, I started a couple of podcasts, one politically themed and one sports themed and I have been recording and editing audio with Audacity. I have also put up a few website related to these podcasts that feature JS powered AI, this is meant more as a learning exercise for myself and to show employers some of the skills i bring to the table.

If you want to get the real measure of a distro, you need to use it for real work, stuff that takes hours, taxes the system and is critical, stuff were you can't afford to lose data.

I have found that most of the distros available are not reliable, for one reason or another. some of the issues that i have encountered are:

1) Audacity failing to launch if installed via package manager even though the appimage works fine.

2) Audacity stops working after a while, with no ability to save your work for extended periods of time, then if you wait long enough it comes back to life and you can save your recording. This one is very nerve-racking

3) Pluma just stops working. I use Pluma for HTML, CSS, JS and Python, and sometimes it will just stop accepting inputs, such as the return key or backspace and it will stop allowing you to save your work.

4) I use a laptop that has a 17" 1600x900 monitor and I hook it up to a 50" 1080P TV that i use as a monitor. Windows 10 handles this setup properly and without a hiccup, regardless of whether I mirror the displays, or extend them and set one or the other as the primary and it also handles it properly when i shut down the TV to pull the HDMI out and switch to using just the laptop monitor.

I have tried Kubuntu, Ubuntu Cinnamon, Ubuntu, PoP OS, Mint Mate, Mint Proper, Mint Debian, Fedora Workstation, Fedora Mate, Fedora KDE, Fedora Cinnamon, PC Linux OS, MX Linux (both editions), Manjaro (all of them), every variant of Gecko Linux (which I normally like), every variant of Spiral Linux, if you name a distro, I tried it, none of them can properly handle this monitor setup.

The only one that handles it properly is the latest edition of Ubuntu Mate, this wasn't the case before, it used to have problems as well, but this latest editions based on Lunar Lobster is the only one that gives me the Windows 10 experience of dependability.

Which brings me to the point, it doesn't matter if Linux is a bit faster in this benchmark or that one. What matter is can i rely on it, does it behave as expected, can i count on it to finish a big project? If the answer is No, then it does me no good.

Well said!

There is always a random chance a segfault would suddenly happen on any program or application running on Linux, or the filesystem suddenly gets corrupted. The filesystem thing I have experienced with my own systems at least four times. On a Mandriva installation back in 2009-ish, on a Mageia installation sometime around 2012, and two more times on a Debian 10 and a Debian 11 installation. Completely and utterly random.

Debian 10 and Debian 11 would also have rare instances where Plasma would just randomly segfault and kill the entire graphical desktop, dumping me back into GDM or a TTY if the desktop was started from a TTY. I never, ever had these sudden graphical desktop terminations happen on a Windows system. Not even during the days of Windows 3.1, Windows 95 and Windows 98 where Windows was merely a GUI for MS-DOS.

[xAlt7x]

@sophisticles , @ Sonadow
Сan we please stay on topic and not provoke flame-wars? This article specifically compares Windows and Linux performance on the same hardware.
Arguments like "Windows might lose some benchmarks but at least it doesn't suck like Linux for X and Y use-cases" seem fanboy-ish to me.

[intelfx]

@sophisticles , @ Sonadow
Сan we please stay on topic and not provoke flame-wars? This article specifically compares Windows and Linux performance on the same hardware.
Arguments like "Windows might lose some benchmarks but at least it doesn't suck like Linux for X and Y use-cases" seem fanboy-ish to me.

You'd have better time preaching to the wall than trying to educate the two biggest MS/Apple fanboys on this forum about the fallacies of being a fanboy.

[avis]

birdie you have fundamentally misunderstood the reason for W1zzard's Windows tweaks. Techpowerup is benchmarking hardware, not software. Therefore, the Windows install is neutered to be as static as possible so that all of the difference between results can be attributed to the choice of graphics card. But in this article, the software is the varied parameter.

Fundamentally your comment cannot be further from the truth.

When you test software you want your hardware to be 100% idle or otherwise your results are not worth anything.

If Windows performs inconsistently in its default configuration, tough luck for Windows.

This inconsistency takes roughly 2 minutes to be eliminated but people here seem to be totally fine with getting whatever results. Again, Linux beats Windows, hope you're happy.

[xAlt7x]

When you test software you want your hardware to be 100% idle or otherwise your results are not worth anything.

By that logic we should tweak Linux as well (turning off services like Gnome Tracker, Baloo). And although this sounds good in theory, but in practice it will to lead to inconsistent results when someone tries to repeat Michael's benchmarks.

[Mario Junior]

Michael has never mentioned whether he prepares Windows for benchmarking properly.

And that's a lot of work that needs to be done otherwise you could be getting quite low results.

https://www.techpowerup.com/forums/t...chmark.287480/

Much BS and schizo shi* for ZERO benefit.