C-Ray FP/CPU Benchmark Test Results

SGI hardware problems, solutions, tips, hacks, etc.
Forum rules
Any posts concerning pirated software or offering to buy/sell/trade commercial software are subject to removal.
User avatar
nekonoko
Site Admin
Site Admin
Posts: 8145
Joined: Thu Jan 23, 2003 1:31 am
Location: Pleasanton, California
Contact:

Re: C-Ray FP/CPU Benchmark Test Results

Unread postby nekonoko » Sat Jun 21, 2008 3:00 am

mapesdhs wrote:I noticed that 8 x 600 system on eBay. Would have bid/bought, but couldn't afford it atm. :\


Nope, that wasn't mine - I sold it here on Nekochan (which is why it wound up split in half) :)

You've done all the C-Ray tests, but how about doing the Blender test?


I'll see what I can do. I don't know much about Blender which is why I've been dragging my feet; if you can give me a step-by-step on how to run the tests that would help tremendously.
Twitter: @neko_no_ko
IRIX Release 4.0.5 IP12 Version 06151813 System V
Copyright 1987-1992 Silicon Graphics, Inc.
All Rights Reserved.

User avatar
joerg
Donor
Donor
Posts: 2229
Joined: Thu Jan 08, 2004 6:57 am
Location: In an origin rack - Germany
Contact:

Re: C-Ray FP/CPU Benchmark Test Results

Unread postby joerg » Sat Jun 21, 2008 3:11 am

On O3400 32x600r14K (6.5.30, cc 7.4.4m)

Code: Select all

$ cat scene | ./c-ray-mt -t 128 > foo.ppm
c-ray-mt v1.1
Rendering took: 0 seconds (95 milliseconds)


When playing around with the number of threads that was my best result. What me take wonder is when runing the test 3 times you get 4 different results ;). It would be fine to have something bigger which runs a few seconds or better minutes.

Code: Select all

cat scene |./c-ray-mt  -t 200 > foo.ppm
c-ray-mt v1.1
Rendering took: 0 seconds (89 milliseconds)


Running the same thing a couple of times returns

Code: Select all

for ((i=0;$i<64;i++));do cat scene |./c-ray-mt  -t 200 > /dev/null;done

Fastest:
Rendering took: 0 seconds (62 milliseconds)
c-ray-mt v1.1

Slowest:
Rendering took: 0 seconds (238 milliseconds)
c-ray-mt v1.1

regards
Joerg

User avatar
mapesdhs
Posts: 2516
Joined: Mon Nov 10, 2003 4:17 pm
Location: Edinburgh, Scotland
Contact:

Re: C-Ray FP/CPU Benchmark Test Results

Unread postby mapesdhs » Sat Jun 21, 2008 5:12 am

joerg writes:
> On O3400 32x600r14K (6.5.30, cc 7.4.4m)

See above posts. The simple scene test is too short; on multi-CPU systems, the coms overhead is a big factor.

The full test suite involves 4 tests - you need to run the sphfract tests and the high-res scene test aswell.
Am off to see Hulk 2, will check back later, see if you have news. ;)

Are you recompiling the source btw? I always do.

Ian.

schleusel
Posts: 495
Joined: Mon Oct 20, 2003 6:49 am
Location: NRW, Germany
Contact:

Re: C-Ray FP/CPU Benchmark Test Results

Unread postby schleusel » Sat Jun 21, 2008 6:33 am

Some more boring intel results :)

ThinkPad T61, Core2Duo T7100 1.8GHz/2MB L2, 4GB DDR2/667
Debian Lenny, Kernel 2.6.24 x86_64, gcc 4.2.4 (-O3 -ffast-math -march=nocona), host name: morticia

Code: Select all

[timo@morticia ~/src/c-ray/c-ray-1.1 ]cat scene | ./c-ray-mt -t 32 > foo.ppm
c-ray-mt v1.1
Rendering took: 0 seconds (414 milliseconds)

[timo@morticia ~/src/c-ray/c-ray-1.1 ]cat sphfract | ./c-ray-mt -t 32 > foo.ppm
c-ray-mt v1.1
Rendering took: 7 seconds (7684 milliseconds)

[timo@morticia ~/src/c-ray/c-ray-1.1 ]cat sphfract | ./c-ray-mt -t 32 -s 1024x768 -r 8 > foo.ppm
c-ray-mt v1.1
Rendering took: 100 seconds (100433 milliseconds)

[timo@morticia ~/src/c-ray/c-ray-1.1 ]cat scene | ./c-ray-mt -t 32 -s 7500x3500 > foo.ppm
c-ray-mt v1.1
Rendering took: 24 seconds (24011 milliseconds)


*edit*

recompiled with ICC 10.1 (-fast).. more of a difference than i expected :)

Code: Select all

[timo@morticia ~/src/c-ray/c-ray-1.1 ]cat scene | ./c-ray-mt.icc -t 32 > foo.ppm
c-ray-mt v1.1
Rendering took: 0 seconds (276 milliseconds)

[timo@morticia ~/src/c-ray/c-ray-1.1 ]cat sphfract | ./c-ray-mt.icc -t 32 > foo.ppm
c-ray-mt v1.1
Rendering took: 4 seconds (4779 milliseconds)

[timo@morticia ~/src/c-ray/c-ray-1.1 ]cat sphfract | ./c-ray-mt.icc -t 32 -s 1024x768 -r 8 > foo.ppm
c-ray-mt v1.1
Rendering took: 62 seconds (62434 milliseconds)

[timo@morticia ~/src/c-ray/c-ray-1.1 ]cat scene | ./c-ray-mt.icc -t 32 -s 7500x3500 > foo.ppm
c-ray-mt v1.1
Rendering took: 15 seconds (15798 milliseconds)

User avatar
joerg
Donor
Donor
Posts: 2229
Joined: Thu Jan 08, 2004 6:57 am
Location: In an origin rack - Germany
Contact:

Re: C-Ray FP/CPU Benchmark Test Results

Unread postby joerg » Sat Jun 21, 2008 10:24 am

mapesdhs wrote:joerg writes:
> On O3400 32x600r14K (6.5.30, cc 7.4.4m)

See above posts. The simple scene test is too short; on multi-CPU systems, the coms overhead is a big factor.

The full test suite involves 4 tests - you need to run the sphfract tests and the high-res scene test aswell.


Code: Select all

cat scene|./c-ray-mt -t 200
c-ray-mt v1.1
Rendering took: 0 seconds (63 milliseconds)

cat sphfract|./c-ray-mt -t 200
c-ray-mt v1.1
Rendering took: 2 seconds (1548 milliseconds)

cat sphfract|./c-ray-mt -s 1024x768 -r 8  -t 200
c-ray-mt v1.1
Rendering took: 20 seconds (19811 milliseconds)


cat scene|./c-ray-mt -s 7500x3500  -t 200
c-ray-mt v1.1
Rendering took: 3 seconds (3161 milliseconds)


Are you recompiling the source btw?

Sure,

Code: Select all

cc -O3 -mips4 -TARG:platform=ip35:proc=r14000  -Ofast=ip35  c-ray-mt.c -o c-ray-mt -lm -lpthread


Code: Select all

for ((i=0;$i<64;i++));do cat sphfract|./c-ray-mt -t 200 > /dev/null 2>>sphfract.txt;done
cat sphfrac.txt | php -r 'while(!feof(STDIN)){$c.=fgets(STDIN);};preg_match_all("#\((\d+)#", $c, $res);sort($res[1]);echo "Shortest run: ".$res[1][0]." ms\n";


regards
Joerg

User avatar
mapesdhs
Posts: 2516
Joined: Mon Nov 10, 2003 4:17 pm
Location: Edinburgh, Scotland
Contact:

Re: C-Ray FP/CPU Benchmark Test Results

Unread postby mapesdhs » Sun Jun 22, 2008 1:58 pm

Thanks! All new results added.

joerg, did you see no benefit from compiling with -IPA and/or -LNO? I did, but only sometimes. Depended on the system.

Btw, interesting to see just how much communications overhead is involved for the shorter scene test on your system.
Compared to linear scaling over a single R14K/600, your O3400 is only 2.8% slower than ideal for the longer high-res
scene test (the one test that involves some main RAM access), 7.8% slower for complex sphfract, 10% slower for normal
sphfract, but a much larger 25% slower than ideal for the short scene test.

This suggests the overall throughput of a system like O3400 would be better demonstrated by a more real-world test,
ie. a more complex scene render, or multiple frames. Thus, definitely a good idea to try the Blender test; full details on how
to run the test are on my site, but briefly: download Blender 2.44 and the test scene file, install Blender in /usr/local/bin,
make sure the Blender directory is added to your path, turn off all unnecessary background daemons (I shut down mediad,
httpd, nsd, ipaliases, nfs, timed, etc.), if the system has no gfx then rlogin from elsewhere (gbit link is best of course), run
up Blender, load the scene file, change the number of threads to 8, press F12 to begin the render. Blender's renderer is a
little odd - you won't see the multiple threads running until part or all of the way through the processing of the first
subsection of the scene. Once finished, note the full description of the elapsed time and PM it to me (don't post here, that
would be off-topic). Since your system has 32 CPUs, its effective throughput would obviously be 4X as fast as the final
result, ie. if rendering multiple frames for a complete animation.

Ian.

mozi
Posts: 56
Joined: Tue Jun 21, 2005 2:46 pm
Location: Canterbury, UK

Re: C-Ray FP/CPU Benchmark Test Results

Unread postby mozi » Tue Jun 24, 2008 3:01 am

This is the system "hinv"

System Configuration: Sun Microsystems sun4u SUNW,Sun-Blade-1000 (2 X UltraSPARC-III+)
System clock frequency: 150 MHZ
Memory size: 8GB

==================================== CPUs ====================================
E$ CPU CPU
CPU Freq Size Implementation Mask Status Location
--- -------- ---------- --------------------- ----- ------ --------
0 1200 MHz 8MB SUNW,UltraSPARC-III+ 11.0 on-line +-board/cpu0
1 1200 MHz 8MB SUNW,UltraSPARC-III+ 11.0 on-line +-board/cpu1

================================= IO Devices =================================
Bus Freq Slot + Name +
Type MHz Status Path Model
------ ---- ---------- ---------------------------- --------------------
pci 33 +s/system-board ebus/ns87317-ecpp (parallel)
okay /pci@8,700000/ebus@5/parallel@1,300278

pci 33 +s/system-board ebus/se (serial)
okay /pci@8,700000/ebus@5/serial@1,400000

pci 33 +s/system-board pci108e,1101 (network) SUNW,pci-eri
okay /pci@8,700000/network@5,1

pci 33 +s/system-board pciclass,0c0010 (firewire)
okay /pci@8,700000/firewire@5,2

pci 33 +s/system-board scsi-pci1000,f (scsi-2)
okay /pci@8,700000/scsi@6

pci 33 +s/system-board scsi-pci1000,f (scsi-2)
okay /pci@8,700000/scsi@6,1

pci 66 +s/system-board SUNW,qlc-pci1077,2200 (scsi-+
okay /pci@8,600000/SUNW,qlc@4

pci 66 +s/system-board SUNW,XVR-600 (display) SUNW,375-3153
okay /pci@8,600000/SUNW,XVR-600@1


============================ Memory Configuration ============================
Segment Table:
-----------------------------------------------------------------------
Base Address Size Interleave Factor Contains
-----------------------------------------------------------------------
0x0 8GB 4 BankIDs 0,1,2,3

Bank Table:
-----------------------------------------------------------
Physical Location
ID ControllerID GroupID Size Interleave Way
-----------------------------------------------------------
0 0 0 2GB 0
1 0 1 2GB 1
2 0 0 2GB 2
3 0 1 2GB 3

Memory Module Groups:
--------------------------------------------------
ControllerID GroupID Labels Status
--------------------------------------------------
0 0 chassis/system-board/J0100
0 0 chassis/system-board/J0202
0 0 chassis/system-board/J0304
0 0 chassis/system-board/J0406
0 1 chassis/system-board/J0101
0 1 chassis/system-board/J0203
0 1 chassis/system-board/J0305
0 1 chassis/system-board/J0407

=============================== usb Devices ===============================

Name Port#
------------ -----
mouse 1
mouse 2
keyboard 3



This is the test result with firefox and top running

bash-3.00$ cat scene | ./c-ray-mt -t 1 -s 1024x768 > foo.ppm
c-ray-mt v1.1
Rendering took: 3 seconds (3837 milliseconds)

bash-3.00$ cat scene | ./c-ray-mt -t 32 -s 1024x768 > foo.ppm
c-ray-mt v1.1
Rendering took: 1 seconds (1987 milliseconds)

bash-3.00$ cat scene | ./c-ray-mt -t 32 > foo.ppm
c-ray-mt v1.1
Rendering took: 1 seconds (1214 milliseconds)

bash-3.00$ cat scene | ./c-ray-mt -t 32 -s 7500x3500 > foo.ppm
c-ray-mt v1.1
Rendering took: 72 seconds (72558 milliseconds)

bash-3.00$ cat scene | ./c-ray-mt -t 32 > foo.ppm
c-ray-mt v1.1
Rendering took: 1 seconds (1219 milliseconds)
Sun Blade 2000 dual 1.2GHz 8GB XVR-600 XVR-500 2x73GB

User avatar
mapesdhs
Posts: 2516
Joined: Mon Nov 10, 2003 4:17 pm
Location: Edinburgh, Scotland
Contact:

Re: C-Ray FP/CPU Benchmark Test Results

Unread postby mapesdhs » Tue Jun 24, 2008 3:11 am

mozi writes:
> This is the test result with firefox and top running

Hmm, any chance you could run the tests without Firefox/top running? (and turn off any unnecessary background
daemons). It can make quite a difference on some systems.

Note that for such a system, I wouldn't worry too much about submitting single-threaded results.

Don't forget to try other numbers of threads besides 32 btw. Systems vary as to what is optimal.

Also note that the tests do not include a 1024x768 version of the 'scene' file. The relevant test
for that resolution is for the 'sphfract' file with '-r 8' also included to increase the aliasing quality.

Lastly, what are the compiler/OS details?

Ian.

User avatar
recondas
Moderator
Moderator
Posts: 5441
Joined: Sun Jun 06, 2004 5:55 pm
Location: NC - USA

Re: C-Ray FP/CPU Benchmark Test Results

Unread postby recondas » Sun Jul 12, 2009 7:07 am

Ian has the blender test results for my Onyx 300 on his Blender Benchmark Results page. The Onyx 300 has since been upgraded from four to sixteen processors, so I re-ran the test. The test was run with the Blender 2.44 provided by Ian, following his lead for multi-processor systems, I increased the tiles to 16x16. The IRIX revision is 6.5.30, an updated hinv is here. The test results with typical desktop configuration background processes, was 00:00:49:66 (tiles increased to 16x16).
***********************************************************************
Welcome to ARMLand - 0/0x0d00
running...(sherwood-root 0607201829)
* InfiniteReality/Reality Software, IRIX 6.5 Release *
***********************************************************************

User avatar
bjornl
Posts: 342
Joined: Tue May 09, 2006 11:55 am
Location: Sweden

Re: C-Ray FP/CPU Benchmark Test Results

Unread postby bjornl » Tue Aug 11, 2009 1:50 pm

Finally did a complete test on the 32 cpu Origin 2000.
Running the precompiled c-ray-mt from Ian's homepage on IRIX 6.5.30. Removed some daemons running in the background first.
hinv viewtopic.php?f=14&t=16720995

cat scene | ./c-ray-mt -t 1 -s 800x600 > foo.ppm

Code: Select all

Threads
1 Rendering took: 4 seconds (4063 milliseconds)
2 Rendering took: 4 seconds (4066 milliseconds)
4 Rendering took: 3 seconds (3078 milliseconds)
8 Rendering took: 1 seconds (1826 milliseconds)
16 Rendering took: 0 seconds (742 milliseconds)
32 Rendering took: 0 seconds (476 milliseconds)
64 Rendering took: 0 seconds (380 milliseconds)
128 Rendering took: 0 seconds (246 milliseconds)
256 Rendering took: 0 seconds (481 milliseconds)
512 Rendering took: 0 seconds (300 milliseconds)
1024 Rendering took: 0 seconds (379 milliseconds) more threads than scanlines specified, reducing number of threads to 600.
Min at 128 threads as 'expected'. Strange increase at 256 threads.

cat sphfract | ./c-ray-mt -t 1 -s 800x600 > foo.ppm

Code: Select all

Threads
1 Rendering took: 113 seconds (113284 milliseconds)
2 Rendering took: 62 seconds (62533 milliseconds)
4 Rendering took: 34 seconds (34320 milliseconds)
8 Rendering took: 17 seconds (17758 milliseconds)
16 Rendering took: 8 seconds (8689 milliseconds)
32 Rendering took: 4 seconds (4555 milliseconds)
64 Rendering took: 4 seconds (4212 milliseconds)
128 Rendering took: 3 seconds (3945 milliseconds)
256 Rendering took: 4 seconds (4047 milliseconds)
512 Rendering took: 5 seconds (5212 milliseconds)
1024 Rendering took: 3 seconds (3898 milliseconds) more threads than scanlines specified, reducing number of threads to 600
Min at 600 threads.

cat scene | ./c-ray-mt -t 1 -s 7500x3500 > foo.ppm

Code: Select all

Threads
1 Rendering took: 243 seconds (243453 milliseconds)
2 Rendering took: 141 seconds (141704 milliseconds)
4 Rendering took: 73 seconds (73677 milliseconds)
8 Rendering took: 39 seconds (39163 milliseconds)
16 Rendering took: 22 seconds (22230 milliseconds)
32 Rendering took: 11 seconds (11619 milliseconds)
64 Rendering took: 9 seconds (9872 milliseconds)
128 Rendering took: 9 seconds (9511 milliseconds)
256 Rendering took: 9 seconds (9367 milliseconds)
512 Rendering took: 9 seconds (9537 milliseconds)
1024 failed to spawn thread: Error 0
Min at 256 threads, but what happened here. Repeated test with 1024 threads gives same result.

cat sphfract | ./c-ray-mt -t 1 -s 1024x768 -r 8 > foo.ppm

Code: Select all

Threads
1 Rendering took: 1484 seconds (1484891 milliseconds)
2 Rendering took: 762 seconds (762014 milliseconds)
4 Rendering took: 407 seconds (407361 milliseconds)
8 Rendering took: 224 seconds (224860 milliseconds)
16 Rendering took: 109 seconds (109506 milliseconds)
32 Rendering took: 55 seconds (55870 milliseconds)
64 Rendering took: 49 seconds (49601 milliseconds)
128 Rendering took: 49 seconds (49542 milliseconds)
256 Rendering took: 49 seconds (49321 milliseconds)
512 Rendering took: 48 seconds (48505 milliseconds)
1024 Rendering took: 47 seconds (47854 milliseconds) more threads than scanlines specified, reducing number of threads to 768
Min at 768 threads

User avatar
mapesdhs
Posts: 2516
Joined: Mon Nov 10, 2003 4:17 pm
Location: Edinburgh, Scotland
Contact:

Re: C-Ray FP/CPU Benchmark Test Results

Unread postby mapesdhs » Tue Aug 11, 2009 3:43 pm

bjornl writes:
> Finally did a complete test on the 32 cpu Origin 2000.

Thanks!!

For a while now I've been wanting to switch the page around so that the simple scene test is not used for
the 'main' table, but which of the other tests should I use instead? Test 2, 3 or 4?

Your results certainly show the overhead processing for the short scene is a signifcant factor, since the scaling
efficiency is only 50% compared to the Tezro Quad-1GHz, whereas for test 2 it's much better and not far off linear
for test 3. In that sense, test 3 is probably the best to use, but it's a much longer test - might discourage people
from bothering to run it. Opinions?


> 1024 failed to spawn thread: Error 0
> Min at 256 threads, but what happened here. Repeated test with 1024 threads gives same result.

I forget offhand, either a limit in the program, or the kernel thread limit.


Note there's no need to run all the inbetween numbers of threads, ie. 2, 4, 8, etc. The interesting data points are those
for 1 thread and then however many gives the best result. For any multi-CPU system, I find the fastest multithreaded result
by starting at 32 threads and then working upwards (so far I've never found less than 32 threads to be better; grud knows
how bri3d obtained his dual-1GHz O350 result).

Ian.

User avatar
jan-jaap
Donor
Donor
Posts: 4935
Joined: Thu Jun 17, 2004 11:35 am
Location: Wijchen, The Netherlands
Contact:

Re: C-Ray FP/CPU Benchmark Test Results

Unread postby jan-jaap » Wed Mar 24, 2010 4:03 pm

New results, from this oldskool numbercruncher with 12* R10000 @ 194MHz (2MB) and 2GB RAM (2-way interleaved). Running IRIX 6.2 and MIPSpro 7.3, both fully patched.

Code: Select all

$ cat scene | ./c-ray-mt -t 60 > foo.ppm
c-ray-mt v1.1
Rendering took: 0 seconds (545 milliseconds)

Code: Select all

$ cat sphfract | ./c-ray-mt -t 60 > foo.ppm
c-ray-mt v1.1
Rendering took: 12 seconds (12544 milliseconds)
:PI: :Indigo: :Indigo: :Indy: :Indy: :Indy: :Indigo2: :Indigo2: :Indigo2IMP: :Octane: :Octane2: :O2: :O2+: Image :Fuel: :Tezro: :4D70G: :Skywriter: :PWRSeries: :Crimson: :ChallengeL: :Onyx: :O200: :Onyx2: :O3x02L:
To accentuate the special identity of the IRIS 4D/70, Silicon Graphics' designers selected a new color palette. The machine's coating blends dark grey, raspberry and beige colors into a pleasing harmony. (IRIS 4D/70 Superworkstation Technical Report)

User avatar
mapesdhs
Posts: 2516
Joined: Mon Nov 10, 2003 4:17 pm
Location: Edinburgh, Scotland
Contact:

Re: C-Ray FP/CPU Benchmark Test Results

Unread postby mapesdhs » Wed Mar 24, 2010 4:49 pm

jan-jaap writes:

New results, from this oldskool numbercruncher with
> 12* R10000 @ 194MHz (2MB) and 2GB RAM (2-way interleaved). Running IRIX 6.2 and MIPSpro 7.3, both fully patched.

Thanks!! 8)

What was the compile line you used to build the c-ray-mt binary?

Also, don't forget the other two tests! They would be:

Code: Select all

cat sphfract | ./c-ray-mt -t 60 -s 1024x768 -r 8 > foo.ppm
cat scene | ./c-ray-mt -t 60 -s 7500x3500 > foo.ppm


Data added!

Here's a question for one and all. The default 'main' test is clearly very short and not so good for testing systems with
lots of CPUs. One of the other tests would be better, but which one? I was thinking the 3rd test, sphfract at 1024x768
with oversampling (lasts long enough for modern PC comparisons, and other non-SGI systems), but on the other hand
it does take a long time to run on older systems. Your thoughts?

Keep meaning to get the other three tests' data formatted into proper tables, but never seem to have the time...

Ian.

User avatar
jan-jaap
Donor
Donor
Posts: 4935
Joined: Thu Jun 17, 2004 11:35 am
Location: Wijchen, The Netherlands
Contact:

Re: C-Ray FP/CPU Benchmark Test Results

Unread postby jan-jaap » Thu Mar 25, 2010 2:25 pm

mapesdhs wrote:What was the compile line you used to build the c-ray-mt binary?

Nothing fancy, I just changed the optimization from IP30 to IP25:

Code: Select all

CC = cc -n32
CFLAGS = -O3 -mips4 -IPA -LNO -TARG:platform=ip25 -Ofast=ip25

I will have to look again at the list of patches installed to this system. Apparently, 'inst' will happily install patchSG0001872 (linker updates for IDO 7.1) even though I had the newer IDF 1.2 for 6.2 installed (MIPSpro 7.3). This results in many internal linker errors for any non-trivial or highly optimized application. God knows which other incorrect patches have been applied by default.

mapesdhs wrote:Also, don't forget the other two tests!

There:

Code: Select all

$ cat sphfract | ./c-ray-mt -t 60 -s 1024x768 -r 8 > foo.ppm
c-ray-mt v1.1
Rendering took: 162 seconds (162363 milliseconds)

$ cat scene | ./c-ray-mt -t 60 -s 7500x3500 > foo.ppm
c-ray-mt v1.1
Rendering took: 26 seconds (26805 milliseconds)


mapesdhs wrote:Your thoughts?

* The first 'scene' test doesn't do much beyond a brief spike on a multi CPU machine, and that's not typical for most numerical jobs.
* Personally, I don't mind a few minutes. But I'm not going to let this machine run for hours, heating up the room just to produce a benchmark number.
* In many cases, dumping the PPM file seems to last longer than the calculation. It's only a guess, but it looks like it's dumping huge files in word sized chunks or something inefficient like that.
:PI: :Indigo: :Indigo: :Indy: :Indy: :Indy: :Indigo2: :Indigo2: :Indigo2IMP: :Octane: :Octane2: :O2: :O2+: Image :Fuel: :Tezro: :4D70G: :Skywriter: :PWRSeries: :Crimson: :ChallengeL: :Onyx: :O200: :Onyx2: :O3x02L:
To accentuate the special identity of the IRIS 4D/70, Silicon Graphics' designers selected a new color palette. The machine's coating blends dark grey, raspberry and beige colors into a pleasing harmony. (IRIS 4D/70 Superworkstation Technical Report)

User avatar
mapesdhs
Posts: 2516
Joined: Mon Nov 10, 2003 4:17 pm
Location: Edinburgh, Scotland
Contact:

Re: C-Ray FP/CPU Benchmark Test Results

Unread postby mapesdhs » Thu Mar 25, 2010 6:12 pm

jan-jaap writes:
> Nothing fancy, I just changed the optimization from IP30 to IP25:

Thanks!


> I will have to look again at the list of patches installed to this system. Apparently, 'inst' will happily install
> patchSG0001872 (linker updates for IDO 7.1) even though I had the newer IDF 1.2 for 6.2 installed (MIPSpro 7.3).

Yeah, 6.2 inst was hell for dealing with patches. I hit numerous headaches while running a student Indy lab in the
late 1990s. The only solution is to clone the disk before installing a patch, and never remove the patch history
files so one can remove patches later if required.


> $ cat sphfract | ./c-ray-mt -t 60 -s 1024x768 -r 8 > foo.ppm
> Rendering took: 162 seconds (162363 milliseconds)
>
> cat scene | ./c-ray-mt -t 60 -s 7500x3500 > foo.ppm
> Rendering took: 26 seconds (26805 milliseconds)

Thanks!! Hey, those are pretty good! Clock for clock, that's quicker than one core of a 2.4GHz Core2Quad, and for
Test 4 the result even bests one core of a 4.1GHz Core2Duo.


> * The first 'scene' test doesn't do much beyond a brief spike on a multi CPU machine, and that's not typical for most numerical jobs.

Absolutely. When I was first given the test data, the numbers were being run on much slower machines. I discovered only later just
how short it was executing on faster systems. I had to wait a while for the more complex data file.


> * Personally, I don't mind a few minutes. But I'm not going to let this machine run for hours, heating up the room just to produce a
> benchmark number.

I'm inclined to get with Test 3, because it seems about right for spanning the widest possible range of systems. On a system like
your's, it takes just 2 or 3 mins which is perfect for a benchmark, but it still lasts long enough on modern systems to be comparable.
Takes a fair while for any older system of course, especially R4K, doubly so if there's no L2.

I think the best thing to do is to continue with the same 4 tests, but reorder them so that the data for Test 3 is presented first (and I'll
rename the tests so Test 3 becomes Test 1). I'll list the original test second, giving an immediate table of info for those testing slower
systems but who do not want to run the long test.


> * In many cases, dumping the PPM file seems to last longer than the calculation. It's only a guess, but it looks like it's dumping
> huge files in word sized chunks or something inefficient like that.

Ah, I need to change the script. The program author told me the output can be safely redirected to /dev/null or something - the
benchmark time does not include the time spent writing out the image.

Thanks for the feedback!!

Ian.


Return to “SGI: Hardware”

Who is online

Users browsing this forum: No registered users and 1 guest