You are not logged in.

#1 2011-11-15 17:22:43

Vamp898
Member
From: 東京
Registered: 2009-01-03
Posts: 891
Website

Do you have to leave Archlinux for the AMD bulldozer

Phoronix recently showed multiple Benchmarks that shows when you compile applications with -march=bdver1 a lot of applications gets more than double fast as with generic compilation and most applications gets a 25% boost.

With that boost you can even beat a Intel Core i7 990X. Without that boost you Bulldozer is slower than a i5 2500k at the most time.

So is there a easy way to get a system which is compiled with -march=bdver1

I mean we dont talk about 3% speed increase, we talk about 25-50% and if you pay 250 € for a processor we talk about 65-125 €. That is a lot of money which gets wasted because the OS is not optimized for the architecture.

Or do you think the difference from -march=generic to -march=bdver1 will minimize over time and so the generic compilation will get faster on the bulldozer?

i already tried to start a Archlinux optimized for K10 for my Phenom but the increase of speed with the optimisation wasn´t that much. On the Bulldozer it is a completely different thing. There is a _huge_ difference.

Last edited by Vamp898 (2011-11-15 18:02:33)

Offline

#2 2011-11-15 18:07:40

graysky
Wiki Maintainer
From: :wq
Registered: 2008-12-01
Posts: 10,595
Website

Re: Do you have to leave Archlinux for the AMD bulldozer

Can you post a link to the Phoronix article?  I have never seen a 25 % boost changing the march flag on any machine.


CPU-optimized Linux-ck packages @ Repo-ck  • AUR packagesZsh and other configs

Offline

#3 2011-11-15 18:20:18

Vamp898
Member
From: 東京
Registered: 2009-01-03
Posts: 891
Website

Re: Do you have to leave Archlinux for the AMD bulldozer

Phoronix always test some mysterious things, so it is hard to find a usefull benchmark. Often they mix a lot of compiler flags to check multiple things at a time. But i think on this image where they run POV-Ray there should be about 24% speed increase if i dont misread it

http://openbenchmarking.org/embed.php?i … fe0918&p=2

dont ask me why the used -Ofast instead of -O2 or -O3. I really dont know it.

But i have a Gentoo myself (beside a Archlinux) and as soon my bulldozer arrived i will do some real-world benchmarking with it and compare the results to Archlinux.

And here you have even about 40%

http://openbenchmarking.org/embed.php?i … 2c82f8&p=2

Last edited by Vamp898 (2011-11-15 18:23:43)

Offline

#4 2011-11-15 18:23:04

hadrons123
Member
From: chennai
Registered: 2011-10-07
Posts: 1,249

Re: Do you have to leave Archlinux for the AMD bulldozer


LENOVO Y 580 IVYBRIDGE 660M NVIDIA
Unix is user-friendly. It just isn't promiscuous about which users it's friendly with. - Steven King

Offline

#5 2011-11-15 18:24:52

Vamp898
Member
From: 東京
Registered: 2009-01-03
Posts: 891
Website

Re: Do you have to leave Archlinux for the AMD bulldozer

Yeah, thats the whole article. But there are not every Benchmarks inside of it. Epsecially C-Ray is missing and the LAME Benchmark. I dont understand why Phoronix makes it so complicated to get useful Benchmarks for a processor.

lol and on gcryped the even dont run the -O flag on the -march=bdver1 which makes it look like it is much slower than without -march. I just hope that my Bulldozer arrives soon that i can make the Benchmarks myself. "Its only done correct when you do it yourself"

Last edited by Vamp898 (2011-11-15 18:26:14)

Offline

#6 2011-11-15 18:34:26

mhertz
Member
From: Denmark
Registered: 2010-06-19
Posts: 681

Re: Do you have to leave Archlinux for the AMD bulldozer

I don't see why anyone would have to leave arch for such optimizations, as we allready have our nice makepkg abs/aur system available for e.g. things like that...

If I had that cpu arch. and the gains where considerable, then I would rebuild the apps that would benefit the most and then leave the rest as per default, where the difference would be neglible...

Offline

#7 2011-11-15 18:35:30

Vamp898
Member
From: 東京
Registered: 2009-01-03
Posts: 891
Website

Re: Do you have to leave Archlinux for the AMD bulldozer

I think i can expect some speed increase when compiling Qt/KDElibs and Plasma with those optimisations on my Desktop.

But i use kde-meta and so i would have to re-build about 300 packages...

Offline

#8 2011-11-15 23:25:45

eldragon
Member
From: Buenos Aires
Registered: 2008-11-18
Posts: 1,029

Re: Do you have to leave Archlinux for the AMD bulldozer

phoronix is usually sensasionalist crap. if they got these percentages...it must have been in some crazy situation which you are most likely never to hit in real life.

if you want to make sure. use abs. test the kernel / xorg and some heavy ass app... but i doubt you will feel a difference.

Last edited by eldragon (2011-11-15 23:26:10)

Offline

#9 2011-11-16 04:45:41

bwat47
Member
Registered: 2009-10-07
Posts: 638

Re: Do you have to leave Archlinux for the AMD bulldozer

if you read the comments in the forums on that article, as usual phoronix has misleading benchmarks.

Offline

#10 2011-11-17 09:43:50

Vamp898
Member
From: 東京
Registered: 2009-01-03
Posts: 891
Website

Re: Do you have to leave Archlinux for the AMD bulldozer

Recompiling the Kernel doesnt help because the Kernel always ignore whatever you set in CFLAGS for stability/security reasons. If you have an AMD, the best optimisation you can get is -march=k8 if you set -march=bdver1 the kernel will just ignore it.

Offline

#11 2011-11-17 22:24:51

spctrl
Member
Registered: 2010-06-20
Posts: 32

Re: Do you have to leave Archlinux for the AMD bulldozer

Phoronix.. never believe anything that comes from there. Sponsored by AMD/ATI so it's always going to be biased, or even fabricated. Had to edit this post a couple of times in order to keep foul language out. Yes, I really hate phoronix.

Offline

#12 2011-11-18 17:25:42

Leonid.I
Member
From: Aethyr
Registered: 2009-03-22
Posts: 999

Re: Do you have to leave Archlinux for the AMD bulldozer

Vamp898 wrote:

Recompiling the Kernel doesnt help because the Kernel always ignore whatever you set in CFLAGS for stability/security reasons. If you have an AMD, the best optimisation you can get is -march=k8 if you set -march=bdver1 the kernel will just ignore it.

Optimizing kernel is irrelevant for overall performance because the code is not being executed in real time. A much better idea is to identify bottleneck apps and optimize them. For example, look at how the HPC software is compiled: the basic distro is -O2 only, for stability. But the codes which have to perform are usually very agressively optimized.

Having said that, what about -march=native?

@spctrl:
I agree with you that all CPU benchmarks are pure crap (phoronix or any other source). Simply because performance is not all about CPU, but also motherboard/ram/bus/etc... I even saw extremetech.com review of bulldozer saying that it's a failure because "windows 7 can't effectively distribute load accross cores" (or something in this spirit, total nonsense). But given the overall low activity of AMD when it comes to advertisements (especially compared to intel), them sponsoring such benchmarks is highly unlikely.

Last edited by Leonid.I (2011-11-18 17:32:55)


Arch Linux is more than just GNU/Linux -- it's an adventure
pkill -9 systemd

Offline

#13 2011-11-18 22:51:36

Grinch
Member
Registered: 2010-11-07
Posts: 265

Re: Do you have to leave Archlinux for the AMD bulldozer

Leonid.I wrote:

Optimizing kernel is irrelevant for overall performance because the code is not being executed in real time.

Yes the low latency of a kernel means that there's not alot of 'raw' performance to be had by more aggressive optimization levels due to:

1. Most (I think) of the performance critical parts of the kernel is already manually optimized using compiler extensions which more directly define the code to be generated and doesn't really benefit from higher optimization levels.

2. Unlike cpu intensive apps the kernel is designed to be very low latency and as such tight loop optimizations which benefit cpu intensive apps won't benefit the kernel the same way. That said, the already low kernel latency can be lowered further by harder optimization, but you can forget anything remotely like 25%.

Leonid.I wrote:

A much better idea is to identify bottleneck apps and optimize them.

Exactly, if you are using compressors, encoders, emulators, rendering, or other computationally intensive programs on a regular basis then there can certainly be worthwhile to compile them with higher optimization levels than the defaults (often -O2) and also natively for your hardware (-march=native). If you REALLY want the best possible optimization you should probably use Profile Guided Optimization aswell which typically results in 5-20% improvement on cpu intensive stuff. Be aware though that alot of programs in this category uses hand-optimized assembly which a compiler can't optimize and as such won't benefit from a recompile with higher optimization (although you can often disable assembly code and see if the compiler can generate faster code for your platform).

As for Phoronix's 'benchmarks', take them with a LARGE grain of salt. The tests have been shown lacking again and again and most of the time it seems the site-owner who is doing them doesn't have a clue of what he is doing. This isn't even limited to compiler benchmarks, he's been benchmarking 3d game performance on systems with 3d compositing on against systems without 3d compositing and presented them as fair comparisons. Overall the tests are of extremely poor quality, not surprising then that he seems to go for quantity.

edit: Oh and I'd like to hear of the results you get when recompiling for the bulldozer architecture as I'm thinking of going for that next time I upgrade my machines (won't be for another 1-1 1/2 year though so no hurry wink )

Last edited by Grinch (2011-11-18 22:56:14)

Offline

#14 2011-11-19 16:58:19

Mr.Elendig
#archlinux@freenode channel op
From: The intertubes
Registered: 2004-11-07
Posts: 4,092

Re: Do you have to leave Archlinux for the AMD bulldozer

And for the original question: No, you don't have to leave arch. I doubt there is a single binary distro that builds their packages with that cflag, and if you go gentoo or similar, then you could just as well use abs to rebuild. It wouldn't really be any significant slower to do it using abs than using a source based distro.


Evil #archlinux@libera.chat channel op and general support dude.
. files on github, Screenshots, Random pics and the rest

Offline

#15 2011-11-20 19:10:39

Vamp898
Member
From: 東京
Registered: 2009-01-03
Posts: 891
Website

Re: Do you have to leave Archlinux for the AMD bulldozer

With the difference that on Gentoo i can have Debug Symbols (not in the binaries with the splitdebug feature) and also i have the USE-Flags so i can much more customize/optimize the system than ABS would allow (on Gentoo setting one USE-Flag which affects all packages would mean editing 300 PKGBUILDs on Archlinux)

So seems if you want to optimize on an architecture that there is not much way around Gentoo

btw. i bought myself an Intel Core i7 2600 just to benchmark it (340 € with Mainboard!) and the result was not suprising


http://img526.imageshack.us/img526/5034/bulldozer.png

i just send back the i7 tomorrow and keep the bulldozer.

Benchmarks are done using Hardinfo with -O3 and -march=native (hardinfo incl. all benchmarks inside it) (re-compiled for every processor)

Just one more thing

Simply because performance is not all about CPU

Nobody ever said that. But CPU Performance is all about CPU and a CPU Benchmark which shows the CPU Performance (correctly) is right in what it does. Showing CPU Performance.

btw. some interesting things about that Benchmarks

Intels HT makes there Processors really faster (without HT there results would be much worse, about 30% less but i dont wanted to create a separate benchmark in case im not showing the performance difference from no HT to HT)
The Intel Core i7 2600 had problems to keep with the AMD Phenom II X6 1100T, on all Windows Benchmarks the AMD Phenom II X6 1100T loose with a huge difference to the i7 2600.

Even when the i7 2600 is faster on Linux than on Windows, the AMD Phenom II X6 1100T wins anyway (what the hell does the windows benchmarks do?)

and as you can see, especially on the CryptoHash Benchmark and Blowfish Benchmark the Bulldozer wins at full line.

Very mysterious, the Windows Benchmark suites also run CryptoHash and Blowfish and there the Bulldozer looses at the full line.

The most difference i can see from the hardinfo benchmarks to the windows benchmarks is that hardinfo is opensource and you can read every line.

That is _really_ mysterious and let me think that there is something wrong with the ClosedSource Benchmarks. Especially that every ClosedSource Benchmark Suites on Windows, even with the same thing benchmarked, have different results.

Offline

#16 2011-11-20 19:54:50

Pierre
Developer
From: Bonn
Registered: 2004-07-05
Posts: 1,964
Website

Re: Do you have to leave Archlinux for the AMD bulldozer

I am not sure if only running hardinfo is the right tool here. It runs way to short to measure anything meaningful. Most tests are run in a fraction of a second. It's quite possible that for a relative long time the cpu was on its lowest power saving state during the test. These tests also don't challenge recent cpus at all. I guess even the phoronix test suite might have more reasonable results.

So far I haven't seen a single report that the latest Bulldozer got even near an i7 2600.

Btw: According to hardinfo my Atom cpu is faster than some Athlon64 or Core 2 Duo. :-)

Offline

#17 2011-11-20 20:00:25

Vamp898
Member
From: 東京
Registered: 2009-01-03
Posts: 891
Website

Re: Do you have to leave Archlinux for the AMD bulldozer

Vor sure i disabled CPU-Freq completely or set to "performance" before i run the Benchmarks.

I also run every Benchmarks about 5-6 Times and the changes was about 1-2% from run-to-run

Also compiling of glibc and gcc goes _much_ faster on the Bulldozer than on the Core i7 2600 and the compiling takes quite long so you can mesure it quite good

Anyway, i bought the i7 2600 just for testing and did more than 50 test over the whole weekend.

I didn´t found any test where the i7 2600 was fast. The i7 lost on the whole line. The only benchmark where the i7 is faster, was the Raytracing Benchmark.

i did 10 boots with bootchartd, 5 times with the i7, 5 times with bulldozer. Bulldozer won every boot.

i compiled the glibc and gcc, every of them 5 times on every cpu, bulldozer won on every compiling

Bulldozer won on every 5 apache benchmark runs

Bulldozer won on every KDE start (10 KDE starts per Processor)

Bulldozer won on every test i run.

And the Harddisk Speed didn´t changed, neither the memory speed. It was the identical computer except the Processor/Board.

So i dont know in which situations the i7 2600 is faster than the bulldozer, but i didn´t found any. Not only one (except raytracing benchmark)

I did raw-converting of the images of my Canon EOS 60D

i did video converting

i converted RAW to MP3

The Bulldozer was faster on really everything. So what if you think the i7 2600 is faster on something, tell me on what, i will test it.

Last edited by Vamp898 (2011-11-20 20:16:06)

Offline

#18 2011-11-23 22:25:19

korkadapa
Member
Registered: 2008-08-27
Posts: 32

Re: Do you have to leave Archlinux for the AMD bulldozer

Vamp898 wrote:

Vor sure i disabled CPU-Freq completely or set to "performance" before i run the Benchmarks.

I also run every Benchmarks about 5-6 Times and the changes was about 1-2% from run-to-run

Also compiling of glibc and gcc goes _much_ faster on the Bulldozer than on the Core i7 2600 and the compiling takes quite long so you can mesure it quite good

Anyway, i bought the i7 2600 just for testing and did more than 50 test over the whole weekend.

I didn´t found any test where the i7 2600 was fast. The i7 lost on the whole line. The only benchmark where the i7 is faster, was the Raytracing Benchmark.

i did 10 boots with bootchartd, 5 times with the i7, 5 times with bulldozer. Bulldozer won every boot.

i compiled the glibc and gcc, every of them 5 times on every cpu, bulldozer won on every compiling

Bulldozer won on every 5 apache benchmark runs

Bulldozer won on every KDE start (10 KDE starts per Processor)

Bulldozer won on every test i run.

And the Harddisk Speed didn´t changed, neither the memory speed. It was the identical computer except the Processor/Board.

So i dont know in which situations the i7 2600 is faster than the bulldozer, but i didn´t found any. Not only one (except raytracing benchmark)

I did raw-converting of the images of my Canon EOS 60D

i did video converting

i converted RAW to MP3

The Bulldozer was faster on really everything. So what if you think the i7 2600 is faster on something, tell me on what, i will test it.

How much faster was the glibc compile times compared to the i7? I'm very tempted to upgrade my old Intel Q6600 to a bulldozer, but all the negative reviews scare me away from it a bit. The biggest reason for this upgrade is to speed up compiling actually (android takes about two hours to compile on this setup...).

If you feel that you have time for it, and haven't sent your i7 back yet, I would be very interested to see what compile times you get on a bigger project. Say webkit for example. But of course, only if you have time for it. smile

Offline

#19 2011-11-24 11:51:46

sakisds
Member
From: Athens, Greece
Registered: 2011-10-03
Posts: 105

Re: Do you have to leave Archlinux for the AMD bulldozer

korkadapa wrote:

How much faster was the glibc compile times compared to the i7? I'm very tempted to upgrade my old Intel Q6600 to a bulldozer, but all the negative reviews scare me away from it a bit. The biggest reason for this upgrade is to speed up compiling actually (android takes about two hours to compile on this setup...).

If you feel that you have time for it, and haven't sent your i7 back yet, I would be very interested to see what compile times you get on a bigger project. Say webkit for example. But of course, only if you have time for it. smile

I just tried compiling the kernel using Arch's stock settings on a i5 2500K and it took me 10 minutes and 2 seconds. I guess the i7 will be faster since it has hyperthreading. But I hear a lot of people picking the i5 over the i7 for budget reasons since the difference isn't that dramatic.

Offline

#20 2011-11-24 12:02:33

korkadapa
Member
Registered: 2008-08-27
Posts: 32

Re: Do you have to leave Archlinux for the AMD bulldozer

sakisds wrote:
korkadapa wrote:

How much faster was the glibc compile times compared to the i7? I'm very tempted to upgrade my old Intel Q6600 to a bulldozer, but all the negative reviews scare me away from it a bit. The biggest reason for this upgrade is to speed up compiling actually (android takes about two hours to compile on this setup...).

If you feel that you have time for it, and haven't sent your i7 back yet, I would be very interested to see what compile times you get on a bigger project. Say webkit for example. But of course, only if you have time for it. smile

I just tried compiling the kernel using Arch's stock settings on a i5 2500K and it took me 10 minutes and 2 seconds. I guess the i7 will be faster since it has hyperthreading. But I hear a lot of people picking the i5 over the i7 for budget reasons since the difference isn't that dramatic.

Great. If we could get someone with a bulldozer to do the same test it would give us an decent measurement of it's performance in real situations. I'll try it on my Q6600 when I get home too.

Offline

#21 2011-11-24 16:45:01

Lone_Wolf
Member
From: Netherlands, Europe
Registered: 2005-10-04
Posts: 11,866

Re: Do you have to leave Archlinux for the AMD bulldozer

It might have to do with the compiler used for the benchmarks.

http://www.agner.org/optimize/blog/read.php?i=49#49


Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.


(A works at time B)  && (time C > time B ) ≠  (A works at time C)

Offline

#22 2011-11-24 17:41:05

bwat47
Member
Registered: 2009-10-07
Posts: 638

Re: Do you have to leave Archlinux for the AMD bulldozer

sakisds wrote:
korkadapa wrote:

How much faster was the glibc compile times compared to the i7? I'm very tempted to upgrade my old Intel Q6600 to a bulldozer, but all the negative reviews scare me away from it a bit. The biggest reason for this upgrade is to speed up compiling actually (android takes about two hours to compile on this setup...).

If you feel that you have time for it, and haven't sent your i7 back yet, I would be very interested to see what compile times you get on a bigger project. Say webkit for example. But of course, only if you have time for it. smile

I just tried compiling the kernel using Arch's stock settings on a i5 2500K and it took me 10 minutes and 2 seconds. I guess the i7 will be faster since it has hyperthreading. But I hear a lot of people picking the i5 over the i7 for budget reasons since the difference isn't that dramatic.

Yeah, for most uses the 2500 will be just as fast. For ~200 dollars that processor is some of the best bang for your buck you can get.

Offline

#23 2011-11-24 18:56:58

korkadapa
Member
Registered: 2008-08-27
Posts: 32

Re: Do you have to leave Archlinux for the AMD bulldozer

==> Finished making: linux 3.1.1-1 (Thu Nov 24 19:53:36 UTC 2011)

real	54m15.641s
user	47m47.343s
sys	5m33.228s

Seems like I really need a new CPU. This was made on a intel Q6600, 8GB RAM and an SSD.

Offline

#24 2011-11-24 21:05:30

.:B:.
Forum Fellow
Registered: 2006-11-26
Posts: 5,819
Website

Re: Do you have to leave Archlinux for the AMD bulldozer

There's tricks like using ccache, might speed up compilation quite a bit - if you haven't tried that yet. I'd say a Q6600 is still quite beefy?

Last edited by .:B:. (2011-11-25 06:43:56)


Got Leenucks? :: Arch: Power in simplicity :: Get Counted! Registered Linux User #392717 :: Blog thingy

Offline

#25 2011-11-24 21:25:10

korkadapa
Member
Registered: 2008-08-27
Posts: 32

Re: Do you have to leave Archlinux for the AMD bulldozer

Yes, I'm aware of ccache. Compiling large projects is a pain anyways, especially android takes forever to compile. That's why I'm interested in the bulldozer, it seems to offer great multithread performance while still having a lower price than the 2600k. But if I upgrade I would like to see a great speed increase, if the increase will be barely noticeable I might just as well wait a year or two for my upgrade. It seems though, from my simple test here, that the latest generation CPUs offer a great speed increase, given that sakisds didn't use stuff like ccach.

Offline

Board footer

Powered by FluxBB