The Fastest Linux Distribution For Ryzen: A 10-Way Linux OS Comparison On Ryzen 7 & Threadripper

[sa666666]

Once again proving that anybody who actually likes AMD should be overjoyed any and every time Intel makes yet another contribution to the open source ecosystem since if you removed Intel's contributions from Linux and only relied on what AMD did, you literally couldn't boot a RyZen machine to a command line much less have the performance optimizations that Intel has provided here.

Yep, just like the "improvement" that Intel contributed to the kernel that slowed down AMD CPUs for no reason (until an AMD dev spotted it and fixed it). Yes, AMD users should definitely be thankful for that.

[JeansenVaars]

Was trying to catch a conclusion out of Clear Linux here but I didn't find anything. Winners can be losers in all. There isn't a second position overall best here.. opensuse and ubuntu are good averages, antergos and solus are best in some scenarios and worse in others.. thoughts?

[slacka]

Mostly compiler settings optimized for recent CPUs.

No, they do more then just change compiler flags. Here are some of their patches to improve performance:

https://github.com/clearlinux-pkgs/linux

Yep, that's why I personally would like to see 32-bit distro's disappear. Then we can assume that the lowest common denominator isn't a CPU from 20 years ago, and actually make use of compiler flags for CPUs made in this decade.

WAT? You are very confused. 32-bit distro builds do not affect the performance of 64 bit distro builds. 64-bit distro's are only held back by the fact that they sometimes limit themselves by targeting CPUs of the first gen x86-64 processors. Eliminated their 32-bit companions wouldn't magically make any of these distros any faster.

If x86 magically disappeared tomorrow, no distro would get any faster, but you would end up with buggier software because building software on diverse platforms helps catch bugs. It also makes software more portable, allowing projects like Raspbian and RaspBSD to exist with a minimum amount of work. You'd also give up the ability to spin up low memory VMs as x86-64 requires about 30% more RAM than identical x86 programs. That said, I'd love to see x32 ABI make a comeback, if x86 ever starts to fade.

[duby229]

Hey, I was just thinking, Michael, On the bar graphs, how hard would it be to add a little popup with that bars name when the mouse is hovered over it? It would help make deciphering graphs with many bars a bit easier.

[Michael]

Hey, I was just thinking, Michael, On the bar graphs, how hard would it be to add a little popup with that bars name when the mouse is hovered over it? It would help make deciphering graphs with many bars a bit easier.

Technically there is tooltip support there, but due to changes in modern browsers, doesn't generally work these days :/ Particularly now with Chrome's per-site isolation feature, our SVG graphs need to be served within an <img> tag otherwise they will not load at all... Compared to other sites just relying upon JavaScript+SVG graphs these days, these graphs are dynamically served from OpenBenchmarking.org, there is JPEG/PNG fallback support (not so important for browsers these days, but still need PNG graph support for dealing with PDF result file generation, etc). Anyhow, due to that cross-domain SVG serving compilation with per-site isolation, until I find a better workaround than <img> that breaks some interactivity and also not wanting to load the graphs with a bunch of JavaScript anyhow.

In Firefox, the SVG graphs still should be served through an object tag or so, but the browser no longer seems to show the xtooltip data.

[duby229]

It would be really interesting to see Gentoo in this race, since that should be one of the distributions trying to battle with Clear Linux.

Well, as a Gentoo user I can say this, it's really up to the user how much optimization he's willing to do. Gentoo provides the tools, but it's up to the user to set it up right. Sabayon or Calculate might make decent premade environments though.

[Michael]

slacka sometimes the forum software acts up until I manually clear the queue.

[eva2000]

Possibly, but I have a feeling Clear linux uses both lto and gpo, the later is only available for a few packages in Gentoo and the former is a bit of a pain to get working across your whole system

Not just LTO and PGO but Intel also use Automatic Feedback Directed Optimizations AutoFDO

https://github.com/google/autofdo

https://gcc.gnu.org/wiki/AutoFDO

What is AutoFDO

AutoFDO uses sampling based profile to drive feedback directed optimizations.

AutoFDO uses perf to collect sample profiles. A standalone tool is used to convert the perf.data file into gcov format. GCC reads in the gcov file and interprets the profile into a set of hashmaps. A standalone pass is added to use the processed profile data to annotate the basic block counts and estimate branch probabilities.

[Zan Lynx]

FDO / PGO is great stuff. I use GCC's -fprofile-generate on some code libraries I build at work and it's a 10 - 30% performance improvement. It's a time consuming build that requires a full test-set run and then a full rebuild.

That's why AutoFDO is really great. You can just run apps normally while collecting profile data and there's no need for automated test sets and slow compiler options (running code built with -fprofile-generate is slow).

[dungeon]

Ok seriously what are they putting in Clear Linux... crack? It's kind of silly that an Intel-based distro is so clearly ahead of the competition. Even on AMD hardware, when it was developed specifically for Intel.

So what gives? I know you did a post recently Michael about trying to supercharge Ubuntu w/ Clear Linux tweaks, but to no avail really, so what's the secret sauce here? Do we know?

Of course we know, these are about 3 main things (i think it is in this order, but you can change order if you like ):

1. Ignorance
2, Out of tree patches
3. Compiler optimizations with recent CPUs in mind