AMD Ryzen 5 4500U Performance On Windows 10 vs. Six Linux Distributions

[CochainComplex]

I'm pretty sure the geometric mean is calculated wrong. The values should be normalized, so around 1, not around 10+.

micheal I think something in the way your software generates the geometric means is wrong. The values should be normalised before being multiplied. See this example.

You might be right I once mentioned inconsistences here the underlying Data looks like this
Geometric mean in the bar plot 363.01 vs. 368.48 compared to in the rawdata 363.010 vs 357.149

[King Mucus]

Well, Clear Linux had the most 1st place finishes and performed significantly better than the competition on CPU intensive tasks. It performed poorly on games, which are normally more GPU than CPU intensive. As such, I don't think it would be reasonable to state that Intel is making an effort to artificially hamper AMD processors performance on their distro.

Other than just setting mtune and march to a specific Intel architecture, it would require meddling on the compilers themselves. The latter would be pointless. If they made the former, it proves that new AMD processors are largely compatible with Intel processors instructions, I think.

Alas, if Michael actually decides to begin benchmarking Gentoo, it would be nice if he activated the ~amd64 mask for the kernel itself and every package installed (or even globally). That would contribute even more to the overall 1337ness and awesomeness, at least in theory. It would also be much appreciated.

[edwaleni]

Is Clear Linux falling behind because it is running on an AMD or are this the first effects of - we will focus on server and cloud not desktop ?

I wondered the same and suspect its probably a little of both.

I already posted with the Intel Clear Team on how their builds can't enable certain features of the Ryzen Mobile line.

With all the reorgs going on at Intel and especially in the open source area, I doubt they will put any energy into AMD support, especially mobile.

Unless someone is in the mood to compile Clear Linux and regression test it on a modern AMD mobile product, I wouldn't place too much more expectations on it.

As for Clear on AMD Mobile:

Will it boot? Yes
Will it run? Yes
Will it perform advanced features? Increasingly No.

[AndyChow]

geometric mean says something different. It can not be always the last otherwise it would end up last in the geometric mean.

I'm pretty sure the geometric mean is calculated wrong. The values should be normalized, so around 1, not around 10+.

micheal I think something in the way your software generates the geometric means is wrong. The values should be normalised before being multiplied. See this example.

[vladimir86]

Yeah, I am also missing some Gentoo here: In the end of the day I am super curious to know how a binary based distro would compare with a port based one!

[Vistaus]

Very insightful. Loved those benchmarks, Michael. Congratulations on your great job!

They really show that Intel is not prioritizing gaming performance at all on their distro and that Firefox is underoptimized on Linux under its default settings.

Now if you could get someone to optimize Gentoo for you, with custom kernel, -O3 and the gentoolto overlay, that would be an order of magnitude of awesomeness. I know it might be a lot of work, but can envision willing volunteers to do it for you on Freenode's #gentoo channel or in the Gentoo forums.

I would become a Premium subscriber if you managed to get a überriced, super optimized Gentoo setup on your benchmarks from now on, even if the results end up disappointing the fanboy who is writing this message to you.

-O3? If you're going for the most optimized build possible, you might as well use -Ofast.

[Veto]

It seems like there still are a couple of places where the open source AMD GPU driver is significantly behind the closed Windows driver (GiMark and TessMark). Hopefully someone (@Marek?? ) will find the reason and make the Linux driver rock even harder

[skeevy420]

simply, they were not using the same kernel..

it's not a fair test since in ubuntu he puts the latest stable kernel and in manjaro he puts older 5.6 kernel on such a modern laptop. Fedora has the manpower needed to backport things to the older kernel series so..
The answer is simply, if manjaro was in them latest stable kernel I think manjaro wouldn't be in the last position. I think like ubuntu, in majaro being so easy to change the kernel, a more fair test would have been doing it with 5.7 kernel in almost every distro..

It makes me wonder if they're using 5.6 by default since ZFS doesn't officially support Linux 5.7 just yet.

Even though Manjaro has a 5.7 available, Michael benchmarking what gets installed OOTB whenever he does the benchmarks makes it a perfectly fair test.

My nitpick is that I'd rather Michael test the GNOME or Plasma version of Manjaro since every other distribution tested was on GNOME or Plasma...or better yet, use the GNOME versions of Manjaro and OpenSUSE so we can maybe eliminate desktop environment factors. XFCE could very well be what is causing Manjaro to have lower numbers on some tests. First page is OpenGL tests and Manjaro is easily the worst across the board while OpenSUSE is right there with Ubuntu so we can deduct that both GNOME and Plasma are performing well in regards to the graphics tests. With no other XFCE distribution tested it's hard to draw an accurate conclusion, but since Ubuntu and SUSE had big first place finish pieces of the pie and the graphics tests are where Manjaro had its ass handed to it (it was pretty competitive elsewhere), the Desktop Environment appears to be the cause of their "stellar" performance.

[wizard69]

Fedora does seem to have really bad Java support. I run F32 but then again have zero interest in Java so that isn't a big bother. What I find harder to describe or quantify, is the feel that running Windows 10 leaves you with. Honestly it sucks! That is based on running it at work. there really isn't a test for that.

In any event I'm waiting on Micheal to run some of that coming Apple Silicon hardware. It will be interesting if they can deliver on all of the hype around ARM of late. Some of that hype comes from me wanting a laptop that actually does run all day on a battery, maybe even all weekend.

[amdtesterman]

why is manjaro always is the last ?

the kernel.. the version of kernel was the EOL 5.6 while in ubuntu he has updated to 5.7 and in opensuse tumblewee is 5.7 so.. except fedora that has 5.6 but much manpower to backport things.. Michael sometimes do tests very unfair..