Is Arch Linux Really Faster Than Ubuntu?

[Apopas]

Arch doesn't have many reasons to be faster than Ubuntu since both come with precompiled packages.
Gentoo on the ther hand can be faster with no much optimization and in many cases significantly faster.
Here are some very nice tests from Linux Magazine.

[nanonyme]

Doesn't everyone use -O2 everywhere anyway?

[BenderRodriguez]

I dissagree. Besides being the central place for all your interesting FLOSS news, Michael really listens to users and just wanted to benchmark Arch Vs. Ubuntu because there has been many situation where people said "Omg Arch is 100000 times faster than Ubuntu WTF!" in the forums.

Of course i understand completely what is this all about BUT making purely hardware tests (imagemagick, encoding) with such near the same packages (xorg, kernel almost the same (2.6.34 not much improved just a couple of fixes)) you can't expect to see responsiveness differencess between those three. I have used Arch in the past (now using Gentoo) i noticed that Arch is much more responsive (at least with the interface) but it doesn't automatically mean that it will compile/encode/decode faster as they are all mainly limited by the hardware/kernel. You can benchmark pure numbers crunching and you can prove that all in all they are all the same BUT you CAN'T prove with those which is snappier.

[BenderRodriguez]

Doesn't everyone use -O2 everywhere anyway?

It is not only the case of using -O2, Ubuntu building it's system MUST assume that there are some users not having some newer CPU capabilities like SSE to SSE4.2 and ONLY applications that make runtime cpu detection won't differ much then they must make a "generic" kernel (Arch must do the same too) for as many computers to run. Arch is more "optimized" for a desktop PC and locks itself out from some archaic PC's thus allowing for better optimizations towards them. What you see in those tests here at phoronix is at most cases kernel/hardware limitation (thus small differences). You will see difference in taska that require CPU bandwidth (Ubuntu got more as it is more generic than Arch which is more suited for a low latency desktop). Also take into account that i am also a greenhorn and it is best to check it for yourself

[nanonyme]

It is not only the case of using -O2, Ubuntu building it's system MUST assume that there are some users not having some newer CPU capabilities like SSE to SSE4.2 and ONLY applications that make runtime cpu detection won't differ much then they must make a "generic" kernel (Arch must do the same too) for as many computers to run.

Then again, the applications that actually benefit of the additional instructions (read: multimedia applications) tend to always do runtime CPU detection anyway.

[Ex-Cyber]

This was actually my point.
To everyone else: stop using funroll-loops, it sometimes makes code run slower. You can't find out which one it is for the particular program without profiling before compiling. Don't use it for whole system. kthxbye

It's not just that. The combination of -finline-functions and -funroll-loops with -Os is kind of pulling in two different directions (small code or not?), and I'm pretty sure that -Os already implies -foptimize-register-move and -fexpensive-optimizations. The whole thing just doesn't seem like it's pushing toward any coherent goal.

[b15hop]

and if you don't want to compile on the atom:
set up a chroot on a more powerfull box.
do the compiling there.
make the chroot a BINHOST.
set the binhost on the atom
pull all the precompiled packages from the binhost to the atom.
Install time: less then ubuntu.

Sometimes compiling isn't that useful because there is no gain out of it. If there is some gain then yes it's suited. I've compiled packages on different machines but again the average joe isn't going to know how to do this. Maybe portage could add this as a new feature. Ie, download source and the fast computer compiles it for all other machines.

[b15hop]

Arch doesn't have many reasons to be faster than Ubuntu since both come with precompiled packages.
Gentoo on the ther hand can be faster with no much optimization and in many cases significantly faster.
Here are some very nice tests from Linux Magazine.

Ah yes, but in my case gentoo wasn't any faster at all. In fact I think even after using gentoo for a whole year, I realised it was slower. It was all in my head that compiling everything == faster. Let someone else worry about compiling the system and optimising it. Then I can just use the OS for what I want.

back in the glorious k6-2 400 days compiling some libs (including glibc) was the difference between 'watchable video' and 'dia show'. So it was very much worth the time spent.

Yes I agree. Back in the day... I used gentoo for the very same reason, just so I could watch a movie on my PC with no stuttering. Compiling gave a huge benefit at the time. Why is it now though, that gentoo isn't showing any significant gains over Arch, let alone ubuntu.

Just a quick note (no edit).

The disadvantages of gentoo are:

- increased complexity compared to SuSE or Ubuntu
- compile time

Personally, I can live with these (I have a Quad-core phenom), but it's something everyone can decide on their own and I can accept that some people would prefer a different distro.

Personally, I'd rather take some compile time and complexity over pulling my hair out with Ubuntu after trying to customise it in a way that is not allowed (tm).

Technically SuSE and Ubuntu share the complexity, but it's all masked and there is no portage. x) I had the same problem when compiling stuff on Ubuntu. Just playing with it too much broke the system so I went back to Arch.

It's not just that. The combination of -finline-functions and -funroll-loops with -Os is kind of pulling in two different directions (small code or not?), and I'm pretty sure that -Os already implies -foptimize-register-move and -fexpensive-optimizations. The whole thing just doesn't seem like it's pushing toward any coherent goal.

Well I guess if someone was really desperate for speed increases. The small bit of code that does 80% of the work could be written in assembly. But then this will bring up the topic of assembly code vs modern compiler advancements. Since in many cases today, the compiler does a better job anyway.

[BenderRodriguez]

@b15shop

If you don't know that it is possible to let other machine compile code for another in Gentoo then you should ask yourself what have you learned for that whole year if i using gentoo second month already know that and many more. Too much optimization will actually hurt performance and don't tell me crap that i don't gain performance. I am using a netbook and needed source based distro to squeeze every bit out of it to actually normally work and i will tell you, Ubuntu is a cow, a big fat cow, even Ubuntu's Gnome which is supposedly memory usage friendlier was heavier than my KDE on Gentoo. And because all applications are exactly suited for my exact CPU IT IS running faster.

[godofgrunts]

Sorry guys, but no self respecting Arch user would allow EXT4 to be her filesystem of choice. Which why us Archers see faster performance in things like apache and such. reiserFS for the small files, XFS for the big stuff.