Ubuntu 23.04 Laptop Performance Mixed Against Ubuntu 22.10

[phoronix]

Phoronix: Ubuntu 23.04 Laptop Performance Mixed Against Ubuntu 22.10

Today marks the release of Ubuntu 23.04 "Lunar Lobster" and I've already been trying it out on a number of test systems. Up today are some initial Ubuntu 23.04 vs. 22.10 laptop benchmarks. If you were hoping though for this release to improve performance, unfortunately that doesn't appear to be the case with overall across a range of workloads Ubuntu 23.04 is similar to -- or in some areas trailing -- Ubuntu 22.10 on both Intel and AMD hardware.

https://www.phoronix.com/review/ubuntu-2304-laptops

[Hans Bull]

That's some pretty serious regressions.

[avis]

It's depressing to see such major regressions as if no one is actually performing (regression) testing but then it's not new. People can hate Microsoft all they want but I don't remember them having such performance related SNAFUs ever except for Vista which had so much new stuff, Microsoft hadn't had the time to optimize everything. And then they had normal actual releases back then, not rolling Windows releases with beta testing as they do now.

[stormcrow]

It's depressing to see such major regressions as if no one is actually performing (regression) testing but then it's not new. People can hate Microsoft all they want but I don't remember them having such performance related SNAFUs ever except for Vista which had so much new stuff, Microsoft hadn't had the time to optimize everything. And then they had normal actual releases back then, not rolling Windows releases with beta testing as they do now.

Mm. Happening all the time with Windows. The most recent one is a (not completely fixed) performance regression copying large files over SMB. There's another one that can sometimes occur in copying large files via file explorer on local storage. That doesn't include all the times they've broken necessary functionality in monthly updates or functionality regressions in Windows 11 compared against 10. Microsoft's software problems are a cautionary tale that automated testing can't catch everything. The only way to do that is to actively use the product day in and day out. It's slow. It's not cool. It doesn't make for catchy headlines unless something breaks spectacularly. But, it's the only way to catch some kinds of bugs. This is why most companies prefer either a managed service provider who does that kind of testing for them, or having the tech team gate in updates only after limited deployment testing. Much as benchmarking can be done wrong and it is certainly gamed by mfgs for marketing purposes; it can identify performance related regressions in a semi-automated way when properly evaluated. Benchmarks for (and part of) marketing should be ignored. Benchmarks for spotting performance problems is good. The later is generally what Michael does and is a net benefit to the FOSS community.

[dreamcat4]

Phoronix: Ubuntu 23.04 Laptop Performance Mixed Against Ubuntu 22.10
were hoping though for this release to improve performance, unfortunately that doesn't appear to be the case...
https://www.phoronix.com/review/ubuntu-2304-laptops

hi michael,
any chance you can install the kde plasma desktop and re-test on the current setup? that would be nice alternative to see against the gnome / mutter....

other than that i am out of ideas. since it could be other things. for example mesa perhaps. having said that, to try a different kernel out then it's actually dead easy to add 3rd party ppas to install both xanmod and liquorix. as alternative kernel builds. that also (like kde) does not interfere with the other stock ubuntu packages. for an easy testing situation. can just reboot back into which ever one(s). to rerun the same tests

[edit] but the reason i bring it up is the other thing can do with those 3rd party xanmod and liquorix kernel would be to run same exact 6.2 kernel back on the older 22.10 release. thus eliminating 6.2 kernel as general reason for this. depending on whatever the outcome of testing the older ubuntu release too. if that makes sense

[piotrj3]

Mm. Happening all the time with Windows. The most recent one is a (not completely fixed) performance regression copying large files over SMB. There's another one that can sometimes occur in copying large files via file explorer on local storage. That doesn't include all the times they've broken necessary functionality in monthly updates or functionality regressions in Windows 11 compared against 10. Microsoft's software problems are a cautionary tale that automated testing can't catch everything. The only way to do that is to actively use the product day in and day out. It's slow. It's not cool. It doesn't make for catchy headlines unless something breaks spectacularly. But, it's the only way to catch some kinds of bugs. This is why most companies prefer either a managed service provider who does that kind of testing for them, or having the tech team gate in updates only after limited deployment testing. Much as benchmarking can be done wrong and it is certainly gamed by mfgs for marketing purposes; it can identify performance related regressions in a semi-automated way when properly evaluated. Benchmarks for (and part of) marketing should be ignored. Benchmarks for spotting performance problems is good. The later is generally what Michael does and is a net benefit to the FOSS community.

That is like singular performance regression over certain task. People meanwhile benchmarked entire windows 11 vs 10 vs 8 vs 7, and for example found in typical benchmarks like Michael provides windows 11 is on average faster then 10. And most of time Windows performance is very consistent.

Here you have 50% regression on geometric mean on popular configuration and noticable performance regression in graphical work on all configurations. That is nothing Windows ever had.

[NeoMorpheus]

For shit and giggles, I would love to see these tests done on SteamOS/HoloISO and compare them.

[hax0r]

Is there a OpenBenchmarking Result ID for the tests that were run in this review? I'm trying to find it on openbenchmarking.org, I would like benchmark Ubuntu 22.10 vs 23.04 on my older 5GHz i7-4970K with mitigations=off and cpupower set to 'performance' and GTX 980 on nouveua and see if I get any regressions with Ubuntu 23.04.

[oiaohm]

Here you have 50% regression on geometric mean on popular configuration and noticable performance regression in graphical work on all configurations. That is nothing Windows ever had.

True and false. Microsoft has never successfully had a perfect 50% regression.

https://stackoverflow.com/questions/45024029/windows-10-poor-performance-compared-to-windows-7-page-fault-handling-is-not-sc

Microsoft has successfully with different Windows updates windows 10 and Vista did 90% performance regressions. Software development be it open source or closed source it possible to goof up badly.

And most of time Windows performance is very consistent.

Most reviewer of Windows don't have time to benchmark every single Windows update to catch the goofs. Yes also you find reviewers not using registered version of windows or version of windows with updates turned off so no windows updates.

Windows 11 has been for regressions surprisingly light. Windows 7 was also light in performance regressions. This also happens with Linux distributions.

[ll1025]

True and false. Microsoft has never successfully had a perfect 50% regression.

https://stackoverflow.com/questions/45024029/windows-10-poor-performance-compared-to-windows-7-page-fault-handling-is-not-sc

Microsoft has successfully with different Windows updates windows 10 and Vista did 90% performance regressions. Software development be it open source or closed source it possible to goof up badly.


Most reviewer of Windows don't have time to benchmark every single Windows update to catch the goofs. Yes also you find reviewers not using registered version of windows or version of windows with updates turned off so no windows updates.

Windows 11 has been for regressions surprisingly light. Windows 7 was also light in performance regressions. This also happens with Linux distributions.

Vista was bad but "90% regression" isn't really accurate.

It's reputation came from a few factors

• Marketing laptops with way too little RAM as "vista-ready", leading to serious swapping
• An SMB / file-copy bug that led to very slow speeds (due to pre-calculating transfer size, among other things)
• A very heavy-handed UAC implementation that people were not ready for

Most of the performance issues were just RAM exhaustion on underspecced machines.