Nekonomicon - Period Comparison of Onyx(1) IR and Commodity Graphics

recondas
Moderator
Who joined June 6, 2004, 5:55 p.m.
and authored 3378 notes

Wrote the following at Dec. 24, 2009, 9:54 a.m...

Ran across this web page <with a date of August 2000> that does some limited comparisons of the triangles per second and fill rate performance of an IR equipped Onyx( 1 ) and several period PCs with commodity graphics cards running linux. While the author acknowledges the tests weren't meant to be anything other than a superficial comparison < "The exact numbers given by tenmillion are almost as meaningless as a CPU's MIPS rating, but they do give some idea of the relative performance of the different systems, and are also good for seeing the effect of different options (triangle size, texturing, lighting, etc). >, I found it interesting.

http://www.evl.uic.edu/pape/CAVE/linux/speed/

A few more semi-related pages by the same author: http://www.evl.uic.edu/pape/CAVE/

I mentioned a similar comparison by the same author in this nekochan thread: viewtopic.php?f=11&t=15636&

If you haven't already, be sure to check out the links to running Quake2 in an SGI Cave system: viewtopic.php?f=11&t=15636&p=122514&#p122514

_________________
***********************************************************************
Welcome to ARMLand - 0/0x0d00
running...(sherwood-root 0607201829)
* InfiniteReality/Reality Software, IRIX 6.5 Release *
***********************************************************************

shadowless
Addict
Who joined March 11, 2009, 8:17 p.m.
and authored 308 notes

Wrote the following at Dec. 24, 2009, 10:27 a.m...

Interesting articles, thanks for sharing

tomo
Enthusiast
Who joined Jan. 18, 2008, 3:53 a.m.
and authored 222 notes

Wrote the following at Dec. 25, 2009, 6:29 a.m...

Nice numbers for old SGI's.
O2 was only 180MHz R5k and it catch modern graphics with mutch faster processors.

Just one think - can someone compile tenmilion test? I'd like to run it on my O2 witch is on the other side of speed spectrum to see difference. Thanks

_________________
:O2:

R7000/600 576MB Ram CDRW 18+9Gb HDD
http://www.tomosgi.co.cc

SAQ
Old Salt
Who joined July 19, 2006, 8:37 a.m.
and authored 3874 notes

Wrote the following at Dec. 25, 2009, 10:16 p.m...

tomo wrote:

Nice numbers for old SGI's.
O2 was only 180MHz R5k and it catch modern graphics with mutch faster processors.

Not to mention that it looks like some if not most of the IRISes were memory-starved (96MB for IRIX 6.5? 256 is better...)

_________________
Damn the torpedoes, full speed ahead!

:Indigo:

vishnu
Enthusiast
Who joined March 18, 2007, 3:25 p.m.
and authored 220 notes

Wrote the following at Jan. 8, 2010, 10:43 p.m...

The original source code to tenmillion.c is here:

http://www.oss.sgi.com/projects/performer/mail/info-performer/perf-96-11/0181.html

I got it to compile and run on my Linux box by changing the call to gettimeofday() to include a timezone struct and by changing the return type of main() to int. I'll try it on my SGI as soon as I get some caffeine into my system tomorrow morning... 8-)

If anyone's interested, to compile it (on Linux) I used:

gcc -Wall -L/usr/X11/lib -lGL -lGLU tenmillion.c -o tenmillion.c.exe

The -Wall switch is optional, but interesting. I always use it when I compile other people's code... :mrgreen:

canavan
Addict
Who joined July 21, 2003, 6:19 a.m.
and authored 373 notes

Wrote the following at Jan. 9, 2010, 9:10 a.m...

Compiled with

Code:

  c99 -n32 -mips4 -OPT:Olimit=0:roundoff=3 -TARG:platform=IP27:proc=r10000 -Ofast  tenmillion.c  -o tenmillion -lGL -lGLU -lfastm  -lm -lX11
 

zmttoxics
Addict
Who joined Aug. 31, 2009, 9:11 a.m.
and authored 494 notes

Wrote the following at Jan. 9, 2010, 10:02 a.m...

vishnu wrote:

The original source code to tenmillion.c is here:

http://www.oss.sgi.com/projects/performer/mail/info-performer/perf-96-11/0181.html

I got it to compile and run on my Linux box by changing the call to gettimeofday() to include a timezone struct and by changing the return type of main() to int. I'll try it on my SGI as soon as I get some caffeine into my system tomorrow morning... 8-)

If anyone's interested, to compile it (on Linux) I used:

gcc -Wall -L/usr/X11/lib -lGL -lGLU tenmillion.c -o tenmillion.c.exe

The -Wall switch is optional, but interesting. I always use it when I compile other people's code... :mrgreen:

".c.exe". Haha. Have you been programming in windows for too long? -o tenmillion would have been fine, and probably wouldn't have caused me to have a morning chuckle.

_________________
Stuff.

vishnu
Enthusiast
Who joined March 18, 2007, 3:25 p.m.
and authored 220 notes

Wrote the following at Jan. 9, 2010, 11:08 a.m...

Yeah, I'm a Visual C++ goit from way back. :oops:

But that doesn't alter the fact that consistent filename extensions are a good idea. I've always wondered Unix uses them for it's shared objects but not executables. Just to be different from Microsoft? I find it a tremendous convenience, for example if I want a list of all the programs I've compiled in $HOME all I have to do is `find ~ -name \*exe` ...

ShadeOfBlue
Moderator
Who joined Nov. 25, 2003, 1:09 p.m.
and authored 568 notes

Wrote the following at Jan. 9, 2010, 12:11 p.m...

vishnu wrote:

if I want a list of all the programs I've compiled in $HOME all I have to do is `find ~ -name \*exe` ...

Code:

find ~ -exec file {} \; | fgrep executable

There is no excuse for putting .exe at the end of UNIX executable names

vishnu
Enthusiast
Who joined March 18, 2007, 3:25 p.m.
and authored 220 notes

Wrote the following at Jan. 9, 2010, 1:44 p.m...

Oh right! I suppose next you're going to be telling me I shouldn't be renaming all my .so's to dll's... :twisted:

ShadeOfBlue
Moderator
Who joined Nov. 25, 2003, 1:09 p.m.
and authored 568 notes

Wrote the following at Jan. 9, 2010, 2:13 p.m...

canavan
Addict
Who joined July 21, 2003, 6:19 a.m.
and authored 373 notes

Wrote the following at Jan. 10, 2010, 8:37 a.m...

Quote:

find ~ -exec file {} \; | fgrep executable

I'd prefer find ~ -perm -1

ShadeOfBlue
Moderator
Who joined Nov. 25, 2003, 1:09 p.m.
and authored 568 notes

Wrote the following at Jan. 10, 2010, 12:13 p.m...

canavan wrote:

Quote:

find ~ -exec file {} \; | fgrep executable

I'd prefer find ~ -perm -1

This will also return directories and other files which happen to have the executable bit set (e.g. shared libraries, scripts, ...).

dexter1
Moderator
Who joined Feb. 20, 2003, 7:57 a.m.
and authored 2057 notes

Wrote the following at Jan. 11, 2010, 2:39 a.m...

ShadeOfBlue wrote:

canavan wrote:

Quote:

find ~ -exec file {} \; | fgrep executable

I'd prefer find ~ -perm -1

This will also return directories and other files which happen to have the executable bit set (e.g. shared libraries, scripts, ...).

Code:

find . -type f -perm -1

_________________
:Crimson:

European nekoware mirror, updated twice a day: http://www.mechanics.citg.tudelft.nl/~everdij/nekoware
ftp://mech001.citg.tudelft.nl rsync mech001.citg.tudelft.nl::nekoware

Megatron-UK
Big Fan
Who joined Nov. 10, 2007, 4:31 a.m.
and authored 152 notes

Wrote the following at Jan. 11, 2010, 6:12 a.m...

Output from my work machine (dual core Athlon X2 5600 + Nvidia Quadro 3450):

Code:

  ./ten light texture zbuffer area=850
  
  running on xxxxx (Quadro FX 3450/4000 SDI/PCI/SSE2) Mon Jan 11 13:11:03 2010
  
  visual: 0x29
  
  850 pixel triangles, 39 triangles/strip, 1 strips
  
  vertex data: n3s t2s v3f
  
  flat shading, zbuffer on, light on, texture on
  
  window size: 848 x 848
  
  test time: 1.947 secs.
  
  time per frame: 0.196 msecs.
  
  geometry rate: 1594743 triangles/sec
  
  fill rate: 1355.532 Mpixels/sec

_________________
:Indigo2IMP:

[ R10k-195/MaxImpact +TRAMs/896Mb/73Gb-15krpm/G160/6.5.22 ]
:Octane2:

[ Dual R12k-400/V8/2Gb/73Gb-15krpm/6.5.29 ]

ShadeOfBlue
Moderator
Who joined Nov. 25, 2003, 1:09 p.m.
and authored 568 notes

Wrote the following at Jan. 11, 2010, 8:38 a.m...

dexter1 wrote:

ShadeOfBlue wrote:

canavan wrote:

Quote:

find ~ -exec file {} \; | fgrep executable

I'd prefer find ~ -perm -1

This will also return directories and other files which happen to have the executable bit set (e.g. shared libraries, scripts, ...).

Code:

find . -type f -perm -1

This will still return shared libraries and other files with the exec bit set

I considered all these options before posting my version with 'file', which returns only actual executables (and possibly shell scripts, depending on the version of the 'file' program used).

sybrfreq
Old Salt
Who joined Aug. 21, 2007, 10:12 p.m.
and authored 1647 notes

Wrote the following at Jan. 12, 2010, 3:13 p.m...

I finally got around to running on my own Onyx which is similar but has only one RM6-16 instead of two RM6-64s.

default options:

Code:

  $ ./tenmillion.bin
  
  running on Aldebaran (IRS/M/1/16/4) Wed Jan 13 13:50:29 2010
  
  visual: 0x4e
  
  50 pixel triangles, 90 triangles/strip, 1 strips
  
  vertex data: v3f
  
  flat shading, zbuffer off, light off, texture off
  
  test time: 1.998 secs.
  
  geometry rate: 2377705 triangles/sec
  
  fill rate: 118.885 Mpixels/sec
  
  $ ./tenmillion.bin texture light smooth
  
  running on Aldebaran (IRS/M/1/16/4) Wed Jan 13 13:56:57 2010
  
  visual: 0x4e
  
  50 pixel triangles, 90 triangles/strip, 1 strips
  
  vertex data: n3s t2s v3f
  
  smooth shading, zbuffer off, light on, texture on
  
  test time: 1.995 secs.
  
  geometry rate: 2010655 triangles/sec
  
  fill rate: 100.533 Mpixels/sec

to match the settings used in the benchmark page (the extra RM really makes a difference!).
12,100,1000 pixels, without and with special effects:

Code:

  $ ./tenmillion.bin area=12
  
  running on Aldebaran (IRS/M/1/16/4) Wed Jan 13 14:26:19 2010
  
  visual: 0x4e
  
  12 pixel triangles, 90 triangles/strip, 2 strips
  
  vertex data: v3f
  
  flat shading, zbuffer off, light off, texture off
  
  test time: 2.000 secs.
  
  geometry rate: 6014936 triangles/sec
  
  fill rate: 72.179 Mpixels/sec
  
  $ ./tenmillion.bin area=100
  
  running on Aldebaran (IRS/M/1/16/4) Wed Jan 13 14:26:32 2010
  
  visual: 0x4e
  
  100 pixel triangles, 90 triangles/strip, 1 strips
  
  vertex data: v3f
  
  flat shading, zbuffer off, light off, texture off
  
  test time: 2.000 secs.
  
  geometry rate: 1433856 triangles/sec
  
  fill rate: 143.386 Mpixels/sec
  
  $ ./tenmillion.bin area=1000
  
  running on Aldebaran (IRS/M/1/16/4) Wed Jan 13 14:26:44 2010
  
  visual: 0x4e
  
  1000 pixel triangles, 36 triangles/strip, 1 strips
  
  vertex data: v3f
  
  flat shading, zbuffer off, light off, texture off
  
  test time: 1.999 secs.
  
  geometry rate: 198800 triangles/sec
  
  fill rate: 198.800 Mpixels/sec
  
  $ ./tenmillion.bin texture light smooth area=12
  
  running on Aldebaran (IRS/M/1/16/4) Wed Jan 13 14:27:46 2010
  
  visual: 0x4e
  
  12 pixel triangles, 90 triangles/strip, 2 strips
  
  vertex data: n3s t2s v3f
  
  smooth shading, zbuffer off, light on, texture on
  
  test time: 1.987 secs.
  
  geometry rate: 2935215 triangles/sec
  
  fill rate: 35.223 Mpixels/sec
  
  $ ./tenmillion.bin texture light smooth area=100
  
  running on Aldebaran (IRS/M/1/16/4) Wed Jan 13 14:27:58 2010
  
  visual: 0x4e
  
  100 pixel triangles, 90 triangles/strip, 1 strips
  
  vertex data: n3s t2s v3f
  
  smooth shading, zbuffer off, light on, texture on
  
  test time: 1.988 secs.
  
  geometry rate: 1231437 triangles/sec
  
  fill rate: 123.144 Mpixels/sec
  
  $ ./tenmillion.bin texture light smooth area=1000
  
  running on Aldebaran (IRS/M/1/16/4) Wed Jan 13 14:28:11 2010
  
  visual: 0x4e
  
  1000 pixel triangles, 36 triangles/strip, 1 strips
  
  vertex data: n3s t2s v3f
  
  smooth shading, zbuffer off, light on, texture on
  
  test time: 2.012 secs.
  
  geometry rate: 171486 triangles/sec
  
  fill rate: 171.486 Mpixels/sec

Code:

  $ /usr/gfx/gfxinfo
  
  Graphics board 0 is "KONAS" graphics.
  
  Managed (":0.0") 1280x1024
  
  Display has 2 channels
  
  4 GEs (of 4), occmask = 0x0f
  
  4MB external BEF ram, 32bit path
  
  1 RM6 board (of 1) 1/0/0/0
  
  Texture Memory: 16MB/-/-/-
  
  Medium pixel depth
  
  32K cmap
  
  Sirius video option detected
  
  Channel 0:
  
  Origin = (0,0)
  
  Video Output: 1280 pixels, 1024 lines, 60.00Hz (1280x1024_60.vfo)
  
  Channel 1:
  
  Channel's Source Size: 1280 pixels, 1024 lines; Origin = (0,0)
  
  Video Output: 640 pixels, 486 lines, 29.97Hz (640x486_30i.vfo)

_________________

(Aldebaran)

(Chaos)

(Machop)
:hp xw9300: (Aggrocrag) :hp dv8000: (Attack)

recondas
Moderator
Who joined June 6, 2004, 5:55 p.m.
and authored 3378 notes

Wrote the following at Jan. 13, 2010, 6:07 a.m...

As a comparison, here's the same tests on an Onyx InfinitePerformance with V12 graphics and 8 700MHz processors <the test runs on a single processor>. The tests were run without stopping any normal background processes or typical desktop applications <FireFox was running so I could cut-n-paste sybrfreq's test commands>. An hinv -vm and gfxinfo for the system are here .

Default options:

Code:

  12% ./tenmillion.bin
  
  running on electraglide (VPRO/B/128) Wed Jan 13 07:54:17 2010
  
  visual: 0x49
  
  50 pixel triangles, 90 triangles/strip, 1 strips
  
  vertex data: v3f
  
  flat shading, zbuffer off, light off, texture off
  
  test time: 1.962 secs.
  
  geometry rate: 5695627 triangles/sec
  
  fill rate: 284.781 Mpixels/sec
  
  electraglide 13% ./tenmillion.bin texture light smooth
  
  running on electraglide (VPRO/B/128) Wed Jan 13 07:55:04 2010
  
  visual: 0x49
  
  50 pixel triangles, 90 triangles/strip, 1 strips
  
  vertex data: n3s t2s v3f
  
  smooth shading, zbuffer off, light on, texture on
  
  test time: 2.003 secs.
  
  geometry rate: 3334624 triangles/sec
  
  fill rate: 166.731 Mpixels/sec

The tenmillion benchmark tests from the article tables 1 and 2>:

Code:

  electraglide 14% ./tenmillion.bin area=12
  
  running on electraglide (VPRO/B/128) Wed Jan 13 07:55:33 2010
  
  visual: 0x49
  
  12 pixel triangles, 90 triangles/strip, 2 strips
  
  vertex data: v3f
  
  flat shading, zbuffer off, light off, texture off
  
  test time: 2.000 secs.
  
  geometry rate: 13747075 triangles/sec
  
  fill rate: 164.965 Mpixels/sec
  
  electraglide 15% ./tenmillion.bin area=100
  
  running on electraglide (VPRO/B/128) Wed Jan 13 07:55:58 2010
  
  visual: 0x49
  
  100 pixel triangles, 90 triangles/strip, 1 strips
  
  vertex data: v3f
  
  flat shading, zbuffer off, light off, texture off
  
  test time: 2.001 secs.
  
  geometry rate: 3083382 triangles/sec
  
  fill rate: 308.338 Mpixels/sec
  
  electraglide 16% ./tenmillion.bin area=1000
  
  running on electraglide (VPRO/B/128) Wed Jan 13 07:56:12 2010
  
  visual: 0x49
  
  1000 pixel triangles, 36 triangles/strip, 1 strips
  
  vertex data: v3f
  
  flat shading, zbuffer off, light off, texture off
  
  test time: 1.998 secs.
  
  geometry rate: 336738 triangles/sec
  
  fill rate: 336.738 Mpixels/sec
  
  electraglide 17% ./tenmillion.bin texture light smooth area=12
  
  running on electraglide (VPRO/B/128) Wed Jan 13 07:56:24 2010
  
  visual: 0x49
  
  12 pixel triangles, 90 triangles/strip, 2 strips
  
  vertex data: n3s t2s v3f
  
  smooth shading, zbuffer off, light on, texture on
  
  test time: 2.003 secs.
  
  geometry rate: 6130902 triangles/sec
  
  fill rate: 73.571 Mpixels/sec
  
  electraglide 18% ./tenmillion.bin texture light smooth area=100
  
  running on electraglide (VPRO/B/128) Wed Jan 13 07:56:42 2010
  
  visual: 0x49
  
  100 pixel triangles, 90 triangles/strip, 1 strips
  
  vertex data: n3s t2s v3f
  
  smooth shading, zbuffer off, light on, texture on
  
  test time: 1.999 secs.
  
  geometry rate: 1939448 triangles/sec
  
  fill rate: 193.945 Mpixels/sec
  
  electraglide 19% ./tenmillion.bin texture light smooth area=1000
  
  running on electraglide (VPRO/B/128) Wed Jan 13 07:57:19 2010
  
  visual: 0x49
  
  1000 pixel triangles, 36 triangles/strip, 1 strips
  
  vertex data: n3s t2s v3f
  
  smooth shading, zbuffer off, light on, texture on
  
  test time: 1.999 secs.
  
  geometry rate: 206274 triangles/sec
  
  fill rate: 206.274 Mpixels/sec

_________________
***********************************************************************
Welcome to ARMLand - 0/0x0d00
running...(sherwood-root 0607201829)
* InfiniteReality/Reality Software, IRIX 6.5 Release *
***********************************************************************

tomo
Enthusiast
Who joined Jan. 18, 2008, 3:53 a.m.
and authored 222 notes

Wrote the following at Feb. 18, 2010, 12:39 a.m...

canavan wrote:

Compiled with

Code:

   c99 -n32 -mips4 -OPT:Olimit=0:roundoff=3 -TARG:platform=IP27:proc=r10000 -Ofast  tenmillion.c  -o tenmillion -lGL -lGLU -lfastm  -lm -lX11
  

Thanks many times - I'v not noticed since now that you compiled it!

I'l add results ASAP.

Code:

  running on smetiacik (CRIME) Thu Feb 18 08:47:20 2010
  
  visual: 0x33
  
  50 pixel triangles, 90 triangles/strip, 1 strips
  
  vertex data: v3f
  
  flat shading, zbuffer off, light off, texture off
  
  test time: 2.008 secs.
  
  geometry rate: 1117160 triangles/sec
  
  fill rate: 55.858 Mpixels/sec
  
  -bash-4.0$ ./tenmilion.bin texture light smooth
  
  running on smetiacik (CRIME) Thu Feb 18 08:47:56 2010
  
  visual: 0x33
  
  50 pixel triangles, 90 triangles/strip, 1 strips
  
  vertex data: n3s t2s v3f
  
  smooth shading, zbuffer off, light on, texture on
  
  test time: 2.010 secs.
  
  geometry rate: 391204 triangles/sec
  
  fill rate: 19.560 Mpixels/sec
  
  -bash-4.0$ ./tenmilion.bin area=12
  
  running on smetiacik (CRIME) Thu Feb 18 08:48:39 2010
  
  visual: 0x33
  
  12 pixel triangles, 90 triangles/strip, 2 strips
  
  vertex data: v3f
  
  flat shading, zbuffer off, light off, texture off
  
  test time: 2.005 secs.
  
  geometry rate: 2185646 triangles/sec
  
  fill rate: 26.228 Mpixels/sec
  
  -bash-4.0$ ./tenmilion.bin area=100
  
  running on smetiacik (CRIME) Thu Feb 18 08:49:12 2010
  
  visual: 0x33
  
  100 pixel triangles, 90 triangles/strip, 1 strips
  
  vertex data: v3f
  
  flat shading, zbuffer off, light off, texture off
  
  test time: 1.981 secs.
  
  geometry rate: 618896 triangles/sec
  
  fill rate: 61.890 Mpixels/sec
  
  -bash-4.0$ ./tenmilion.bin area=1000
  
  running on smetiacik (CRIME) Thu Feb 18 08:49:42 2010
  
  visual: 0x33
  
  1000 pixel triangles, 36 triangles/strip, 1 strips
  
  vertex data: v3f
  
  flat shading, zbuffer off, light off, texture off
  
  test time: 1.971 secs.
  
  geometry rate: 66195 triangles/sec
  
  fill rate: 66.195 Mpixels/sec
  
  -bash-4.0$ ./tenmilion.bin texture light smooth area=12
  
  running on smetiacik (CRIME) Thu Feb 18 08:50:12 2010
  
  visual: 0x33
  
  12 pixel triangles, 90 triangles/strip, 2 strips
  
  vertex data: n3s t2s v3f
  
  smooth shading, zbuffer off, light on, texture on
  
  test time: 2.006 secs.
  
  geometry rate: 765094 triangles/sec
  
  fill rate: 9.181 Mpixels/sec
  
  -bash-4.0$ ./tenmilion.bin texture light smooth area=100
  
  running on smetiacik (CRIME) Thu Feb 18 08:50:48 2010
  
  visual: 0x33
  
  100 pixel triangles, 90 triangles/strip, 1 strips
  
  vertex data: n3s t2s v3f
  
  smooth shading, zbuffer off, light on, texture on
  
  test time: 1.991 secs.
  
  geometry rate: 205448 triangles/sec
  
  fill rate: 20.545 Mpixels/sec
  
  -bash-4.0$ ./tenmilion.bin texture light smooth area=1000
  
  running on smetiacik (CRIME) Thu Feb 18 08:51:04 2010
  
  visual: 0x33
  
  1000 pixel triangles, 36 triangles/strip, 1 strips
  
  vertex data: n3s t2s v3f
  
  smooth shading, zbuffer off, light on, texture on
  
  test time: 1.978 secs.
  
  geometry rate: 23734 triangles/sec
  
  fill rate: 23.734 Mpixels/sec

_________________
:O2:

R7000/600 576MB Ram CDRW 18+9Gb HDD
http://www.tomosgi.co.cc

MAKHNO
Stranger
Who joined Dec. 28, 2009, 5:05 p.m.
and authored 13 notes

Wrote the following at Feb. 22, 2010, 3:22 p.m...

Has anybody compiled it for Windows, to try and see the numbers on a current ati/nvidia?

_________________
:Indy:

No longer with me:
:Indigo:

SGI: Computer Graphics

Period Comparison of Onyx(1) IR and Commodity Graphics - Page 1