Ubuntu 64-bit More Competitive Against Mac OS X

[Apopas]

Yes, I just thought we'll get Ext 4 as a bonus .

Hehehe, I'm sure we'll see ext4 benchmarks when Fedora 11 is out.

[kraftman]

ext4 would be very interesting indeed but the current test should take place with ext3 as long as it begun with ext3. If Michael had used ext4 for 64 bit then we couldn't make a comparison between 32 and 64 since we would have two different factors (the bits and the filesystem).

Yes, I just thought we'll get Ext 4 as a bonus .

[Apopas]

Thanks VERY much for next Ext3 benchmark... Who's interested in Ext4?

ext4 would be very interesting indeed but the current test should take place with ext3 as long as it begun with ext3. If Michael had used ext4 for 64 bit then we couldn't make a comparison between 32 and 64 since we would have two different factors (the bits and the filesystem).

[kraftman]

So i guess it is proven now what i said about OpenSSL test in

http://www.phoronix.com/scan.php?pag...nchmarks&num=4

The 2.6.29 test definitely was run with 64 bit!!! Compare to:

http://www.phoronix.com/scan.php?pag...sx_64bit&num=6

The speed improvement is only possible using a 64 bit build. Different kernel do not matter for that. Maybe for 3d performance + intel benchmarks. Or maybe add 2.6.29/30 tests.

What's going on here?! :O

[Apopas]

I am a Linux user too, but I have to admit that the 3d games benchmark is WAY MORE important than the openssl signs/second...
Unless we don't want Linux as a gaming platform.I know I want.

Hopefully, the nasty Intel driver regression will be fixed soon and Linux on Intel graphics will outperform MAC.
And since SQQLite regression is already fixed, soon we will be able to say that Ubuntu/Linux outperforms MAC in most aspects. Unless Snow Leopard can improve enough to keep up.

Yup but you aren't going afterall to buy a mac-mini or an Intel VGA for a game station. Even the 16 FPS of MacOSX make the game unplayable. The only option is ATI or NVIDIA VGA and there Linux goes very well.

Anyway, the results were not of a big surprise, Linux 64 was, is and probably will be faster than any other OS. Only one test seems strange to me. The GnuPG performance.

Last year vs today the test was:
63.58 vs 91.86 vs 67.57 (OSX, 32bit, 64bit)
63.92 vs 67.22 vs 85.86

Look! back then OSX was similar with 64bit Ubuntu and 32bit sucked.
Today they changed position. 64bit sucks while 32bit approached OSX
Someone could give me a logical explanation plz?

[L4Linux]

Lies, damned lies, and the statistics:

Seven of the 29 tests changed their winner/loser status for Ubuntu by switching from 32 to 64-bit. Six tests became winners for Ubuntu 64 and 1 became a loser.

Overall,

12 wins for Mac OS X,
4 wins for Ubuntu 32, and
13 wins for Ubuntu 64.

Ubuntu 64 wins.


I think if the graphics and SQLite performance were fixed then there really wouldn't be much of a performance gap for most desktop users to care about.

I am a Linux user too, but I have to admit that the 3d games benchmark is WAY MORE important than the openssl signs/second...
Unless we don't want Linux as a gaming platform.I know I want.

Hopefully, the nasty Intel driver regression will be fixed soon and Linux on Intel graphics will outperform MAC.
And since SQQLite regression is already fixed, soon we will be able to say that Ubuntu/Linux outperforms MAC in most aspects. Unless Snow Leopard can improve enough to keep up.

[Kano]

So i guess it is proven now what i said about OpenSSL test in

http://www.phoronix.com/scan.php?pag...nchmarks&num=4

The 2.6.29 test definitely was run with 64 bit!!! Compare to:

http://www.phoronix.com/scan.php?pag...sx_64bit&num=6

The speed improvement is only possible using a 64 bit build. Different kernel do not matter for that. Maybe for 3d performance + intel benchmarks. Or maybe add 2.6.29/30 tests.

[kraftman]

cool but the current osx version is 10.5.7

It would be cool to see Ubuntu x86_64 using Ext 4 (possible?) vs OS X 10.5.7 and both using same version of GCC.

[El_Angelo]

and to think that *ANY* other distro beats ubuntu hands down in performance...

[chithanh]

The major difference (which these benchs show) is the use of SSE by default in x86_64, which is way faster than x87 floating point unit.

Actually, 64-bit code even uses SSE2 as all x86_64 CPUs are SSE2 capable. To test your theory, one could compare a source based distro (eg. Gentoo) which was compiled with such optimization against a non-optimized distro.

That being said, encryption and video encoding tests also benefit from the additional x86_64 registers, so I expect that 32 bit code will still perform worse.