PDA

View Full Version : 32-bit vs. 64-bit Benchmarks


joshuapurcell
12-20-2006, 11:08 PM
I found a link to this PDF on the Ubuntu forums, and I thought it was right up our alley here at Phoronix. It would be cool to have possibly a more detailed, verifiable benchmark performed along these same lines. Here' the link:
http://art-blog.no-ip.info/files/amd64vsi386.pdf

I'm going to install Ubuntu 64-bit once I have a weekend to mess with it. Post if this link stops working anytime soon... I'll put the file up somewhere and change the link.

Michael
12-21-2006, 07:43 AM
I'll put the file up somewhere and change the link.

The link died, at least temporarily, when I was in the middle of downloading it... Looks like I'll just have to run my own tests ;)

joshuapurcell
12-21-2006, 02:50 PM
The link died, at least temporarily, when I was in the middle of downloading it... Looks like I'll just have to run my own tests ;)
That's what I was hoping :D... I'm up at work right now, but I'll get this file uploaded to some location and provide a link hopefully later tonight so you can see what this file looks like.

Michael
12-21-2006, 06:15 PM
That's what I was hoping :D... I'm up at work right now, but I'll get this file uploaded to some location and provide a link hopefully later tonight so you can see what this file looks like.

I'll see that such a i386 vs. x86_64 comparison gets completed before the New Year... Since it's the holidays, Ubuntu may be used for a change instead of Fedora.

Michael
12-22-2006, 09:09 AM
I'll see that such a i386 vs. x86_64 comparison gets completed before the New Year... Since it's the holidays, Ubuntu may be used for a change instead of Fedora.

Pardon my last statement, I may prepare for 32-bit and 64-bit tests on both Fedora and Ubuntu.

Michael
12-29-2006, 08:25 AM
While 64-bit support is now considered common for both Intel and AMD processors, many Linux (as well as Windows) users are uncertain whether to use a 32-bit or 64-bit operating system with there being advantages for both paths. With this being the last Phoronix article for 2006, we decided to take this opportunity to look at the common question of whether to use 32-bit or 64-bit software. In this article, we will be comparing the i386 and x86_64 performance with Ubuntu 6.10 Edgy Eft and Ubuntu 7.04 Feisty Fawn Herd 1 to see how the numbers truly stack up.

http://www.phoronix.com/vr.php?view=8325

If anyone is interested in any other benchmarks just let me know.

Fragadelic
12-29-2006, 11:26 AM
Not much benefit in running 64-bit at the moment which is the same thing I found myself. The only difference I found was encoding video was a bit faster on 64-but but everything else was pretty much the same like your article.

This leads me to believe that most code is not optimized enough for 64-bit yet.

Hopefully this will eventually change.

[Knuckles]
12-29-2006, 03:28 PM
Thanks for the benchmarking, but could you try to repeat some of the tests from the pdf that have a greater "gain %"?
That way we would get an idea if the pdf results are just "strange" -- e.g. maybe the fact that the testing was done using livecd's somehow influenced the results, or if the apps chosen are more prone to speeding up on x86_64.

feight
12-29-2006, 05:46 PM
Appreciate the benchmark testing. I had no idea the differences were so slight even with 64bit enabled programs. I knew the advantage was mostly in number crunching because of the increased ram usage but I think I will stop fiddling with x86_64 for a bit as I'm not a coder or a tester.

It's probably for the best; it generally just causes me grief because of the codec/flash forced architectures and croots. These benchmarks are all the more valid to me because Ubuntu is my distro.

joshuapurcell
12-30-2006, 02:48 PM
Way to go on the benchmarks, thanks for posting this. I'll be going with the Edgy 64-bit version once I get some time to go through the install. Here's a little more information on why I'll be going with a 64-bit OS (copied from another post I made a couple of weeks ago):

I've made some notes on what different OSes see as far as memory goes on my system. I should say the BIOS always says 4193216K unless I disable 'Memory Hole Remapping' (which it then reports 3327MB some reason). The following numbers are all taken from within the OS while the BIOS says 4GB:
Edgy 64-bit: 4047512K
Edgy 32-bit: 3369216K
WinXP 32-bit: 3406252K

Edgy64 shows the most memory at around 3.8GB (still not the full 4GB for some reason), but Edgy32 doesn't even see as much as WinXP! I didn't have WinXP 64-bit to test. I tried installing Feisty64 from the alternate CD, but I had problems getting the partition setup in the same way that I had Edgy (with LVM and software RAID being used)... I'll test out Feisty64 most likely with the next release (Herd 2). I also want to see what other Linux distributions with different kernels detect at some point.

artik
12-31-2006, 05:04 AM
I have seen links from this forum to my article (amd64vsi386.pdf).

The main point of benchmarks that I had done was to test wide range of software - not few specific programs but verify common tasks that computers are used for (multimedia,servers,games,math,3D) and see what are the advantages, in certain cases there are but sometimes no.

I think one of the problems with most of benchmarks that they use "standard benchmark tools" - like FPU speed or RAM speed that compare different HW (the speed of RAM/FPU will not change on same HW). What you need to compare is "same software" compiled for different platforms.

So IMHO tests like "RAM speed" just irrelevant in case of 64 vs 32 comparison.

Best,
Artyom

miknix
12-31-2006, 11:12 AM
Clearly, Ubuntu it's not the perfect distro for 64bit computing.
I think a more optimized distro should be used to benchmark 64bit VS 32bit, like Gentoo.
My Gentoo AMD64 laptop gives me far away better results in data compression, data encoding and source compiling!

Unreal Tournament has some specific maps just for 64bits platforms. Play them in x86 and feel the difference!

When talking about my developed apps that needs critical performance in execution time, using ACML in my AMD64 gives me more than 40% of speed increase.

joshuapurcell
01-01-2007, 02:53 PM
Clearly, Ubuntu it's not the perfect distro for 64bit computing.
I think a more optimized distro should be used to benchmark 64bit VS 32bit, like Gentoo.
My Gentoo AMD64 laptop gives me far away better results in data compression, data encoding and source compiling!

Unreal Tournament has some specific maps just for 64bits platforms. Play them in x86 and feel the difference!

When talking about my developed apps that needs critical performance in execution time, using ACML in my AMD64 gives me more than 40% of speed increase.From my previous post I showed that Edgy64 could see 4047512K, which is the most I have currently been able to get an OS to see. Do you also have 4GB of memory, and if so does Gentoo64 see more memory? I haven't tested this out yet.

miknix
01-01-2007, 03:16 PM
No, sorry.

joshuapurcell
01-01-2007, 03:51 PM
Thanks for the reply anyway. I'll install Gentoo64 and see how much memory it finds and make a post once it's done.Clearly, Ubuntu it's not the perfect distro for 64bit computing.
I think a more optimized distro should be used to benchmark 64bit VS 32bit, like Gentoo.
My Gentoo AMD64 laptop gives me far away better results in data compression, data encoding and source compiling!I hear this argument every once in a while, and I'm fully prepared to accept that Gentoo may be better than Ubuntu in some aspects. I'd like to see the results of your tests though if you have them, because without some tests or other types of proof it just comes down to the same old "which distribution you like better" argument, and we can find that discussion anywhere. In any area (but especially when it involves gaming), do you have some test results that show Gentoo64 is faster than Ubuntu64? I expect some things to be faster in Gentoo64 because most packages are compiled for the computer the OS is running on (if I'm not mistaken). But I want to know if there is actually a speed increase where it matters to me, and if any speed increase makes it actually worth it to move from Ubuntu and all the benefits that come with using this distribution.

miknix
01-01-2007, 04:41 PM
In any area (but especially when it involves gaming), do you have some test results that show Gentoo64 is faster than Ubuntu64?

I'll try to benchmark my laptop soon as I can. My laptop graphics card burned out because the VGA cooler din't start. (ACPI BUG)
So I don't have any amd64 processor at hand right now :(


I expect some things to be faster in Gentoo64 because most packages are compiled for the computer the OS is running on (if I'm not mistaken). But I want to know if there is actually a speed increase where it matters to me, and if any speed increase makes it actually worth it to move from Ubuntu and all the benefits that come with using this distribution.
Yes, Gentoo builds all packages from source. But we can archieve major optimizations by choosing compiler specific flags for each package.

Meanwhile, a thread about this benchmark in gentoo forums was opened. I think you would like to take a look
http://forums.gentoo.org/viewtopic-t-349691.html

If you would like to try gentoo, AMD recommends (http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/32035.pdf)
these CFLAGS="-O3 -ffast-math -funroll-all-loops -fpeel-loops -ftracer -funswitch-loops -funit-at-a-time"

Another interesting link: http://enterprise.linux.com/article.pl?sid=05/06/09/1413209&tid=121
Look at these benchmarks and read the comments :)
These results, despite being gentoo based, are not optimal just because the packages were compiled with march=pentium4. march=athlon64 should be used instead.
This is what happens to most of binary packages based distros. They have generic (safe) x86_64 support.

joshuapurcell
01-01-2007, 04:58 PM
mknix: Thanks for the very helpful information. This type of stuff is exactly what I was interested in to help me decide between Gentoo64 and Ubuntu64. I'll continue looking for some performance information and comparisons for the two distributions and post when I find some more information.

MU_Engineer
01-04-2007, 09:31 PM
From my previous post I showed that Edgy64 could see 4047512K, which is the most I have currently been able to get an OS to see. Do you also have 4GB of memory, and if so does Gentoo64 see more memory? I haven't tested this out yet.

You will *never* see the "full" 4194304K as the kernel will reserve some of the RAM for itself. My 64-bit Gentoo box with 2GB RAM sees 2011MB. My 32-bit laptop sees about 1510MB out of 1.5GB. Every distro will see roughly the same amount of RAM as long as it is capable. A 32-bit distro will not be able to see much more than 3GB RAM as there is a 3GB/1GB user/kernel split set up in most 32-bit distros. If you recompile the kernel with PAE (64GB High Memory Support) you should see ~4GB. I did this on a 32-bit box running FC5 and I see about 4GB RAM now.

joshupurcell: Almost all Linuxes that run the same applications will run more or less the same speed. I run Gentoo and like it a lot, but to be brutally honest, it runs no faster than any other KDE distribution with unneeded daemons shut off. Gentoo is about configurability (and perhaps learning) but not speed, except in very odd cases (such as being able to tune a very specific number-crunching math app.) Gentoo is also a wonderful platform to develop on as you have a full toolkit and headers already there once you have the system up. If there is any speed to be gained from Gentoo, it's by cutting out stuff you don't need so there's less of a load on the system. The Gentoo devs are very good, but they're not magic. We are all bound by the limits of our hardware and the design of the common set of applications we all run like OpenOffice, gcc, KDE, and Gnome.

cakey
01-04-2007, 10:31 PM
Did you use the 64 bit ut2004 binary?

joshuapurcell
01-05-2007, 10:23 AM
You will *never* see the "full" 4194304K as the kernel will reserve some of the RAM for itself. My 64-bit Gentoo box with 2GB RAM sees 2011MB. My 32-bit laptop sees about 1510MB out of 1.5GB. Every distro will see roughly the same amount of RAM as long as it is capable. A 32-bit distro will not be able to see much more than 3GB RAM as there is a 3GB/1GB user/kernel split set up in most 32-bit distros. If you recompile the kernel with PAE (64GB High Memory Support) you should see ~4GB. I did this on a 32-bit box running FC5 and I see about 4GB RAM now.I'm not expecting any Linux distribution to see all of the 4GB, but since there is so much difference between the three OS's that I've tried so far I wanted to see if there were any similar differences between two 64-bit Linux distributions. I would like to say that since both Edgy64 and Gentoo64 are both using the Linux kernel that would be enough to assume they will both see the same amount of memory. But I'm not sure about that since I have no idea what relevant options were on/off when compiling the kernels for these distributions, and I know the two distributions are using different versions of the kernel.

The 32-bit memory limitations are something I'm aware of, but I am surprised at the information you gave regarding PAE and 4GB of memory under Fedora Core5. I'll have to look into possibly recompiling my current kernel with PAE enabled before doing a reinstall and seeing if that helps out.

davidecapod
01-13-2007, 01:19 PM
Hi guys, I want to buy a new PC but I have a doubt in buying AMD and to go with a 64bit distro or Intel and stick with 32bit...
I know that Core2Duo beats Athlon64X2, but is this the same if the second uses 64bit?
I am obviously talking about same price level, for example
E6400 vs 4600+ or E6600 vs 5000+

I have not found any benchmark around of this kind...

Thanks in advance
Davide

Michael
01-13-2007, 01:28 PM
Both Intel and AMD have great processors, but whether you use a 32-bit or 64-bit distribution, the Intel Core 2 Duo will likely take the lead. What exactly will you be using this system for?

davidecapod
01-15-2007, 04:05 AM
Both Intel and AMD have great processors, but whether you use a 32-bit or 64-bit distribution, the Intel Core 2 Duo will likely take the lead. What exactly will you be using this system for?

Normal desktop use, plus some gaming, movie transcoding, and development. So the Core2duo is definitely the best buy, even under a 64bit environment?

1c3d0g
01-15-2007, 10:05 AM
Of course it is. The Core 2 Duo's are some of the best CPU's Intel has ever released. Just read the online reviews, even AMD-biased websites have acknowledged that Intel has the performance crown for now.

joshuapurcell
01-15-2007, 12:51 PM
Not counting my first two computers (an Apple II GS and then some Intel box), I've bought AMD chips for my computers every time I've had the opportunity. Given that, I'll admit that everything I've read points to Intel having the leading chip at this point with its Core 2 Duo. At the same time, the processor market is like a pendulum: this quarter Intel is the favorite while AMD was the favorite for a long time a little while back. I would look at the online support you find specific to the motherboards supporting the different chips and the communities like Phoronix to see what works best for you. If you find a group that seems to have lots of answers to questions you have and they are all running systems based on a specific hardware, then it may pay off to go with something similar to what they are running if this is one of your first systems to build.

Synergy6
02-03-2007, 05:34 PM
Normal desktop use, plus some gaming, movie transcoding, and development. So the Core2duo is definitely the best buy, even under a 64bit environment?

Yes.
Synergy6