Gentoo Forums
Gentoo Forums
Gentoo Forums
Quick Search: in
AMD64 64bit vs 32 bit benchmarks on the web
View unanswered posts
View posts from last 24 hours

 
Reply to topic    Gentoo Forums Forum Index Gentoo on AMD64
View previous topic :: View next topic  
Author Message
Gestalt73
n00b
n00b


Joined: 02 Jun 2004
Posts: 29

PostPosted: Mon Jul 12, 2004 4:58 pm    Post subject: AMD64 64bit vs 32 bit benchmarks on the web Reply with quote

I thought I'd pass these along. I stumbled across them this morning on Slashdot. I'd been looking for comparisons between 32bit and 64bit performance:

http://anandtech.com/linux/showdoc.aspx?i=2114
http://www.thejemreport.com/modules.php?op=modload&name=News&file=article&sid=126
http://www.thejemreport.com/modules.php?op=modload&name=News&file=article&sid=117

Does anyone have any others to share?

Alan
Back to top
View user's profile Send private message
headache
Apprentice
Apprentice


Joined: 26 May 2004
Posts: 226

PostPosted: Mon Jul 12, 2004 9:48 pm    Post subject: Reply with quote

After I read the Anandtech benchmarks of 64-bit Linux i thought I was going to replicate one of the benchmarks, namely the POVRAY test.
I did 3 different runs with different CFLAGS. The first run was done with the CFLAGS I normally use to emerge programs with. The second one was just a change from O2 to O3. The last run was with CFLAGS that I extracted from the Acovea thread in the Chat forum.

Between each run, I unmerged and re-emerged povray with the new CFLAGS. I only ran one single test with each set of CFLAGS.

I ran this on an AMD64 3200+ with 1GB of RAM and using gcc 3.4.1 and glibc-2.3.4.20040619.

CFLAGS="-march=k8 -O2 -pipe -funroll-loops -fpeel-loops"
Code:
Statistics for benchmark.pov, Resolution 384 x 384
----------------------------------------------------------------------------
Pixels:          147840   Samples:          555728   Smpls/Pxl: 3.76
Rays:           1824355   Saved:             23589   Max Level: 12/12
----------------------------------------------------------------------------
Ray->Shape Intersection          Tests       Succeeded  Percentage
----------------------------------------------------------------------------
Box                           77295978         9373568     12.13
Cone/Cylinder                 75749548         6531688      8.62
CSG Intersection             160974278        55366504     34.39
CSG Merge                       771681           34694      4.50
Fractal                        1815380          104693      5.77
Height Field                   3759713          105232      2.80
Height Field Box               3759713          711817     18.93
Height Field Triangle          3398511          108397      3.19
Height Field Block             5930579         1758119     29.64
Height Field Cell             23668666         1866748      7.89
Isosurface                    11694978          719265      6.15
Isosurface Container          12186396        11695534     95.97
Isosurface Cache                158312           41002     25.90
Mesh                          12429024           63730      0.51
Plane                         85687243         1280724      1.49
Sphere                       261647690       153783921     58.78
Superellipsoid                  693720           44335      6.39
Torus                          2943724          405892     13.79
Torus Bound                    2943724          474488     16.12
True Type Font                  842539           79753      9.47
Clipping Object                2554850         1520146     59.50
Bounding Box                 499515090       146369020     29.30
Vista Buffer                  21820603        12553293     57.53
----------------------------------------------------------------------------
Isosurface roots:         11689683
Function VM calls:       169849997
----------------------------------------------------------------------------
Roots tested:               474488   eliminated:              272644
Calls to Noise:          236731216   Calls to DNoise:     2417540818
----------------------------------------------------------------------------
Media Intervals:          36542508   Media Samples:        329623399 (9.02)
Shadow Ray Tests:        119370042   Succeeded:             49124947
Reflected Rays:             222617   Total Internal:             442
Refracted Rays:             144971
Transmitted Rays:           604939
Number of photons shot:          74025
Surface photons stored:          64084
Priority queue insert:         6937092
Priority queue remove:         1278966
Gather function called:         658399
----------------------------------------------------------------------------
Smallest Alloc:                 17 bytes   Largest:          2720016
Peak memory used:          7887262 bytes
----------------------------------------------------------------------------
Time For Parse:    0 hours  0 minutes   1.0 seconds (1 seconds)
Time For Photon:   0 hours  0 minutes  36.0 seconds (36 seconds)
Time For Trace:    0 hours 22 minutes  55.0 seconds (1375 seconds)
    Total Time:    0 hours 23 minutes  32.0 seconds (1412 seconds)


CFLAGS="-march=k8 -O3 -pipe -funroll-loops -fpeel-loops"
Code:
Statistics for benchmark.pov, Resolution 384 x 384
----------------------------------------------------------------------------
Pixels:          147840   Samples:          555728   Smpls/Pxl: 3.76
Rays:           1824355   Saved:             23589   Max Level: 12/12
----------------------------------------------------------------------------
Ray->Shape Intersection          Tests       Succeeded  Percentage
----------------------------------------------------------------------------
Box                           77295978         9373568     12.13
Cone/Cylinder                 75749548         6531688      8.62
CSG Intersection             160974278        55366504     34.39
CSG Merge                       771681           34694      4.50
Fractal                        1815380          104693      5.77
Height Field                   3759713          105232      2.80
Height Field Box               3759713          711817     18.93
Height Field Triangle          3398511          108397      3.19
Height Field Block             5930579         1758119     29.64
Height Field Cell             23668666         1866748      7.89
Isosurface                    11694978          719265      6.15
Isosurface Container          12186396        11695534     95.97
Isosurface Cache                158312           41002     25.90
Mesh                          12429024           63730      0.51
Plane                         85687243         1280724      1.49
Sphere                       261647690       153783921     58.78
Superellipsoid                  693720           44335      6.39
Torus                          2943724          405892     13.79
Torus Bound                    2943724          474488     16.12
True Type Font                  842539           79753      9.47
Clipping Object                2554850         1520146     59.50
Bounding Box                 499515090       146369020     29.30
Vista Buffer                  21820603        12553293     57.53
----------------------------------------------------------------------------
Isosurface roots:         11689683
Function VM calls:       169849997
----------------------------------------------------------------------------
Roots tested:               474488   eliminated:              272644
Calls to Noise:          236731216   Calls to DNoise:     2417540818
----------------------------------------------------------------------------
Media Intervals:          36542508   Media Samples:        329623399 (9.02)
Shadow Ray Tests:        119370042   Succeeded:             49124947
Reflected Rays:             222617   Total Internal:             442
Refracted Rays:             144971
Transmitted Rays:           604939
Number of photons shot:          74025
Surface photons stored:          64084
Priority queue insert:         6937092
Priority queue remove:         1278966
Gather function called:         658399
----------------------------------------------------------------------------
Smallest Alloc:                 17 bytes   Largest:          2720016
Peak memory used:          7887262 bytes
----------------------------------------------------------------------------
Time For Parse:    0 hours  0 minutes   1.0 seconds (1 seconds)
Time For Photon:   0 hours  0 minutes  37.0 seconds (37 seconds)
Time For Trace:    0 hours 22 minutes  56.0 seconds (1376 seconds)
    Total Time:    0 hours 23 minutes  34.0 seconds (1414 seconds)


CFLAGS="-march=k8 -O1 -mieee-fp -funsafe-math-optimizations -freorder-blocks -fstrength-reduce -fstrict-aliasing -fprefetch-loop-arrays -funroll-all-loops -finline-functions -ffast-math"
Code:
Statistics for benchmark.pov, Resolution 384 x 384
----------------------------------------------------------------------------
Pixels:          147840   Samples:          555728   Smpls/Pxl: 3.76
Rays:           1824355   Saved:             23589   Max Level: 12/12
----------------------------------------------------------------------------
Ray->Shape Intersection          Tests       Succeeded  Percentage
----------------------------------------------------------------------------
Box                           77295978         9373568     12.13
Cone/Cylinder                 75749548         6531688      8.62
CSG Intersection             160974278        55366504     34.39
CSG Merge                       771681           34694      4.50
Fractal                        1815380          104693      5.77
Height Field                   3759713          105232      2.80
Height Field Box               3759713          711817     18.93
Height Field Triangle          3398511          108397      3.19
Height Field Block             5930579         1758119     29.64
Height Field Cell             23668666         1866748      7.89
Isosurface                    11694978          719265      6.15
Isosurface Container          12186396        11695534     95.97
Isosurface Cache                163658           43108     26.34
Mesh                          12429024           63730      0.51
Plane                         85687243         1280724      1.49
Sphere                       261647690       153783921     58.78
Superellipsoid                  693720           44335      6.39
Torus                          2943724          405892     13.79
Torus Bound                    2943724          474488     16.12
True Type Font                  842539           79753      9.47
Clipping Object                2554850         1520146     59.50
Bounding Box                 499515090       146369020     29.30
Vista Buffer                  21820603        12553293     57.53
----------------------------------------------------------------------------
Isosurface roots:         11689683
Function VM calls:       169845487
----------------------------------------------------------------------------
Roots tested:               474488   eliminated:              272644
Calls to Noise:          236697236   Calls to DNoise:     2417539018
----------------------------------------------------------------------------
Media Intervals:          36542508   Media Samples:        329623399 (9.02)
Shadow Ray Tests:        119370042   Succeeded:             49124947
Reflected Rays:             222617   Total Internal:             442
Refracted Rays:             144971
Transmitted Rays:           604939
Number of photons shot:          74025
Surface photons stored:          64084
Priority queue insert:         6937092
Priority queue remove:         1278966
Gather function called:         658399
----------------------------------------------------------------------------
Smallest Alloc:                 17 bytes   Largest:          2720016
Peak memory used:          7887262 bytes
----------------------------------------------------------------------------
Time For Parse:    0 hours  0 minutes   2.0 seconds (2 seconds)
Time For Photon:   0 hours  0 minutes  36.0 seconds (36 seconds)
Time For Trace:    0 hours 23 minutes  27.0 seconds (1407 seconds)
    Total Time:    0 hours 24 minutes   5.0 seconds (1445 seconds)


As you can see, the first set of CFLAGS yielded the best result. Quite interesting.
When comparing my best run, 1412 seconds with the result in the benchmark with an AMD64 3500+ CPU 1399 seconds, I'd say my box did quite well! That was using SuSE 9.1. When you compare it against FC2, 1700 seconds, it's quite obvious that a properly compiled program can make or break your performance.
_________________
"I'd rather have a President who does it to a woman than one who does it to his country" -- Shirley Maclaine
Liberal (adj.): Free from bigotry; open to progress; tolerant of others.
Back to top
View user's profile Send private message
toofastforyahuh
Apprentice
Apprentice


Joined: 18 May 2004
Posts: 158

PostPosted: Tue Jul 13, 2004 10:54 am    Post subject: Re: AMD64 64bit vs 32 bit benchmarks on the web Reply with quote

Gestalt73 wrote:

Does anyone have any others to share?
Alan


Yes. xmame has been a study for a few of us for awhile now. After some hiccups handsomepete helped me with I finally sat down and did some 32-bit versus 64-bit gcc 3.3.3 and then compared it to 64-bit gcc 3.4.0.

Newest results are here:
http://www.anthrofox.org/code/mame/xmame64_bench83.html

Older ones are here:
http://www.anthrofox.org/code/mame/xmame64_bench.html
Back to top
View user's profile Send private message
-ptah-
n00b
n00b


Joined: 29 Jun 2004
Posts: 3

PostPosted: Tue Jul 13, 2004 12:37 pm    Post subject: Reply with quote

Well i did some benchmarks with various filesystem,filesystem encryptions (dm-crypt) and keysizes all tested with bonnie++.
The test was done on a AMD64 3200+ (2ghz 1mb cache) with 1GB Ram. The drive was a Maxtor 80GB 7200rpm ( D740x-6L ). Always using GCC 3.3.2.

The tests were done on a Gentoo 32bit compile with

CFLAGS="-march=athlon-xp -O3 -pipe -fomit-frame-pointer"

and on a 64bit Gentoo compiled with:

CFLAGS="-O2 -pipe"

I know these cflags will probably affect the benchmark a lot but since the 64bit Linux is my main workstation i'm not willing to optimize to much since i simply don't have the time to deal with problems it might cause. But i have some free time coming up so i might be able to share some new test with the gcc 3.4 and 03 optimization.
Also the read benchmarks are missing since i didn't use enough files for a proper time reading. I will fix that when i do another test. But since a complete test takes around 10hrs i will be a few days at least until i can rerun the script...

So here is the 32bit test:

http://homepages.tu-darmstadt.de/~fritschi/gentoo/output_x86.html

and 64bit:

http://homepages.tu-darmstadt.de/~fritschi/gentoo/output-x86_64.html

If anyone is interested in the scripts i used:

I used a modified mkcfs from the dm-crypt wiki
http://homepages.tu-darmstadt.de/~fritschi/gentoo/mkcfs
and some scripts i wrote myself:
http://homepages.tu-darmstadt.de/~fritschi/gentoo/dm-crypt_benchmark-amd64.sh
http://homepages.tu-darmstadt.de/~fritschi/gentoo/dm-crypt_benchmark_x86.sh
Back to top
View user's profile Send private message
AlyM
n00b
n00b


Joined: 02 Sep 2003
Posts: 62
Location: Edinburgh, Scotland

PostPosted: Wed Jul 14, 2004 2:13 pm    Post subject: Re: AMD64 64bit vs 32 bit benchmarks on the web Reply with quote

toofastforyahuh wrote:

Newest results are here:
http://www.anthrofox.org/code/mame/xmame64_bench83.html


I thought gcc 3.4.x was meant to give a fair performance boost over 3.3.x on amd64, but quite a few of your benchmarks showed performance regressions. Especially galaxyfg with showing a 20% performance decrease with gcc 3.4.0.

Worth filing a performance regression with the gcc team, or is this just a blip in your benchmarking?
Back to top
View user's profile Send private message
Desti²
Tux's lil' helper
Tux's lil' helper


Joined: 06 Sep 2003
Posts: 127

PostPosted: Sat Jul 17, 2004 12:33 am    Post subject: Re: AMD64 64bit vs 32 bit benchmarks on the web Reply with quote

Gestalt73 wrote:
I thought I'd pass these along. I stumbled across them this morning on Slashdot. I'd been looking for comparisons between 32bit and 64bit performance:
[...]
Does anyone have any others to share?

Alan


# GIMP 2.0.1 - x86 und AMD64 im Vergleich
http://planet64bit.de/modules/news/article.php?storyid=115 8)
Back to top
View user's profile Send private message
Stolz
Moderator
Moderator


Joined: 19 Oct 2003
Posts: 3008
Location: Hong Kong

PostPosted: Fri Aug 05, 2005 7:37 am    Post subject: Re: AMD64 64bit vs 32 bit benchmarks on the web Reply with quote

Gestalt73 wrote:
...
Does anyone have any others to share?

Alan


Take a look at
https://forums.gentoo.org/viewtopic-t-349691.html
http://enterprise.linux.com/enterprise/05/06/09/1413209.shtml?tid=121
Back to top
View user's profile Send private message
Display posts from previous:   
Reply to topic    Gentoo Forums Forum Index Gentoo on AMD64 All times are GMT
Page 1 of 1

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum