Ubuntu equally fast as Gentoo....WHAT???

[energyman76b]

[dE_logics]

So I was wondering you can 'flex' Gentoo to optimize for speed

[dE_logics]

I just installed Ubuntu 9.10, and these are the results -

Firefox consumes the same amount of memory -- ~47MB with Gentoo handbook opened.

Gnome system monitor consumes 8 mb memory on Ubuntu and 5.9 mb memory on Gentoo.

Smooth scroll in Firefox is horrible in Ubtunu...cannot be used where as in Gentoo I can easily use it...it's just a bit slow.

Best of all Gentoo uses 120 - 140 mb memory on startup and Ubuntu was using 304.

Now I'm using 2.6.30 on Gentoo and 2.6.31 on Ubtuntu.


I wont be comparing the startups now cause I've installed Ubuntu on a thumbdrive.

[i92guboj]

We must never forget the grandpa of all the ricing threads

http://forums.gentoo.org/viewtopic-t-309752-highlight-ricer.html

I've never said that Gentoo *can't* be made faster. But so can any other distro.

What I say is that if you are using Gentoo because you think that simply compiling your system yourself will give it some kind of magic understanding of the inner working of your hardware, then you are wrong. For that you will need a lot of fine tuning and experimenting, and not just set CFLAGS="-momg-t3h-fastest-of-all" and forget. The performance gain, if any at all, from compiling a whole system with general settings can be slightly worse or slightly better than the standard, either way. Most packages come with sane defaults *for that concrete package*.

I am sure that most people around already know this. I am just thinking aloud and trying to cool down the ricing spirit I've already detected in the thread

To get a real benefit from compiling, you would need to waste your entire life compiling and checking builds for every given package. You could as well just throw a thousand crazy CFLAGS in the middle instead of doing it the scientific way, and break some packages in the middle, cripple some others, and get a cool performance boost in one or two of them, it's what we call "ricing".

And, since you have to invest that much work, you could very well do it in any other distro. Installing a working toolchain and making the packages is gonna be the lesser of your problems, really.

On the contrary, things like USE flag provide a safe and direct benefit, they can potentially reduce the dependencies to a fraction, they can reduce the code footprint and they can do lot of funny things without having to worry whether your system is going to work right or accurately tomorrow due to some silly flag to get uber floating point maths speed or insane loop unrolling that will cripple your L1 cache, which is becoming a rare resource nowadays.
_________________
Gentoo Handbook

[i92guboj]

[energyman76b]

that is not unfair. That is a true advantage. Gentoo does not install tons of unneeded crap and starts all and everything. That is an important and cool feature.
_________________

[The Earth]

[dE_logics]

[dE_logics]

[ChojinDSL]

Didn't Phoronix or some other well known linux related site (or was it a gentoo user's blog?) recently compare benchmarks of different apps in gentoo and ubuntu? Too bad I can't remember which site that was anymore. But the results were interesting.

The guy compared ubuntu to Gentoo with -Os, gentoo -O2 and gentoo -O3.

In most of the applications, gentoo was faster. There were only one or two where ubuntu was faster. And it wasnt a small negligeble difference either. With some apps it was 15-30% increase in performance.

The tests also showed that -Os and -O3 where only faster in certain situations with certain apps. Overall -O2 performed the best in total.

Now, the numbers don't lie. But for day to day use, they're probably meaning less.
I mean, if you occasionally encode an mp3, then it's not gonna make much difference to you if the process takes 1minute or 1minute and 15 seconds.

On the other hand, if you do things such as video encoding or 3D Rendering (like me ) on a regular basis, then every little speed increase achieved purely through compiler optimisation IS gonna make a difference.
If a single frame takes 1minute 30seconds to render on ubuntu, but only 1minute on gentoo. That might not seem like much. But if you are rendering a animation with 7500 frames. That can be quite a difference.

That being said, I feel that the main advantage of gentoo is it's approach to package management and configuration. Compiler optimisations are just a nifty little perk.

But what I love about gentoo is that I'm not forced to use a certain framework or desktop, simply because the distro is tailored to it.

For example: I'm a kde guy. I don't feel comfortable with Gnome. I can use it, but I feel more at home with kde. Now your probably thinking, how is that problem with something like ubuntu?
Well from past experince, I find that ubuntu with gnome gets the most love from the ubuntu devs. All the other variants such as kubuntu or xubuntu seem like step children in comparison.

Yes, you can customise to your hearts content (its linux after all), but the point of something like *buntu, is that automagic out-of-the-box experience. If I need to "pop the hood" so to speak, I'd rather do it in an environment that was designed to do so from the ground up.

[energyman76b]

it was linux-mag:

http://www.linux-mag.com/id/7574
_________________

[dE_logics]

Today I did a bit of benchmarks of startup times, and yes, Gentoo was fast.

Gentoo is slow, it's not faster etc... is all a myth. I don't care what people say...point is I've measured and found it to be at least 30% faster (and at a max of 60%). Even time you benchmark Gentoo...it always comes ahead...it's not a matter of doubt that Gentoo is the fastest of the lot...CONSIDERABLY.

People saying Gentoo is slow and all...I just cant believe them. Either they have misconfiguration the system, have not optimized it by that much or are just biased...that's final with me.

[shickapooka800]

[szczerb]

[dE_logics]

Mod Edit: Fixed some broken tags, changed [code] tags to [quote] tags because the latter supports font-size changes, and then changed font size to try to reduce the extremely large vertical space of this post.

To read the text in the quote blocks, please highlight it and copy-paste it into your favorite editor for viewing there.

In the future when you need to post large quantities of output, please use one of the free text-pasting services that are available, such as http://pastebin.com/. Then in your post, simply link to the url the service provides you with.

(And before I go, a comment regarding the content of this post: there's a lot of warnings, error messages, and timings of sub-processes here. It is very difficult to follow what's going on and what's being measured. Generally, when something is this long, it is a good idea to filter the output and post only the relevant information. For example, the total time start to finish and perhaps one or two intermediate times if those are important. As it stands, it takes a lot of work to read and understand. Most people won't bother they'll skip the thread instead of trying to figure out what it is saying and what advice to offer.)

-- Akkara

[cokehabit]

[shickapooka800]

omg use pastebin
_________________
The Dose Makes the Poison

[cokehabit]

[szczerb]

And what exactly does that show? You fired up a bunch of GUI applications and timed them. With no comment or anything, so I suppose that you manually clicked the same stuff each time - which proves nothing. You just wasted a lot of your time and a bit of ours (and got reported for abusing the scrolling part of my touchpad).

[shickapooka800]

i'd like to refer the spammer to the concept of dynamically loaded, shared libraries and their ... dynamic ... effect on program execution times. if you want to even try and use 'time' in this manner, you'd need to start from a cold boot, with exactly the same software running on each trial (library versions, kernel version, etc etc). this is not even close to a fair comparison.

there you have it. I have been trolled.
_________________
The Dose Makes the Poison

[predatorfreak]

Christ I don't think I've seen nutty ricery since I made kernel patchsets and playfool was still maintaining no-sources... I feel old. Anyway on topic.

Right, first let's clear up some serious misconceptions with CFLAGS. The major performance differences that you will get are usually in applications with very little hand-written assembly code, which do very very math-heavy tasks. In general, with no assembly involved, good test cases for compiler performance include: xvid, x264, mplayer/ffmpeg, blender, pov-ray and applications like that.

Note, most (especially xvid/x264/mplayer/ffmpeg) will see significantly higher gains by enabling their hand-optimized assembly than you will ever see with any combination of CFLAGS, period. In general, from testing encoding a 1000-frame DVD test clip from straight YUV4MPEG, I got these results with xvid with no asm (Yes I am too lazy to do x264):

-march=x86-64 -mtune=generic -O2 -pipe (single-pass, single-thread, bitrate 900): Average 18FPS
-march=x86-64 -mtune=generic -O2 -pipe (single-pass, 8-threads, bitrate 900): Average 41FPS
-march=x86-64 -mtune=generic -O3 -pipe (single-pass, single-thread, bitrate 900): Average 20FPS
-march=x86-64 -mtune=generic -O3 -pipe (single-pass, 8-threads, bitrate 900): Average 44FPS
-march=native -msse4 -O2 -pipe (single-pass, single-thread, bitrate 900): Average 18FPS
-march=native -msse4 -O2 -pipe (single-pass, 8-threads, bitrate 900): Average 42FPS
-march=native -msse4 -O3 -pipe (single-pass, single-thread, bitrate 900): Average 20FPS
-march=native -msse4 -O3 -pipe (single-pass, 8-threads, bitrate 900): Average 45FPS

Tests done with GCC 4.4.2 on a Core i7 @3.0GHz (Turbo-enabled, so it hits about 3.3GHz single-threaded, 3.2GHz multithreaded in load), 9GB of RAM (I'm nutty and decided that 9GB triple-channel is cool), USE flag pic to disable all assembly optimizations in XVID. So, let's sum this up: A math intensive media-encoding application (best-case test scenario for compiler performance) gains 2FPS from -O3 single-threaded and 3FPS from multi-threaded. BTW, optimizing explicitly for my CPU did about jack crap for my average FPS, with only the multithreaded -O2 build showing a noticeable difference.

To contrast, -march=native -msse4 -O2 -pipe, same options as above, with the pic USE flag disabled (asm enabled), single-threaded is 72FPS, multi-threaded is 134FPS. Seriously, the compiler does significantly worse than hand optimization where it counts.

All that said, the reasons I use Gentoo (atm, I'm a flip-flopper) are:
1. USE flags
2. LDFLAGS (I notice a decent difference with everything using GNU Hash and -O1, with regard to load-times)
3. x264/mplayer/ffmpeg aren't ancient/crippled to death/etc, and live ebuilds exist for ffmpeg/mplayer.

(BTW encode your stuff with x264, it averages 130FPS here on DVD content with standard options, about 80FPS with higher quality settings, and it produces significantly better results than XVID)
_________________
System: predatorbox
Distro: Arch Linux x86_64
Current projects: blackhole, convmedia and anything else I cook up.

[cokehabit]

[dE_logics]

[predatorfreak]

[speeddemon]

Wow.
I opened this thread fully expecting a joke, and what do you know. I haven't seen one of these threads in years. As much as I want to comment, I'll leave this one to the ricers.
_________________
Cats are deadly animals. If you stick your nose up their crotch and snort their piss, THEY CAN KILL YOU!!!