Fractal benchmark

IRIX/Nekoware development, porting and related topics.
Forum rules
Any posts concerning pirated software or offering to buy/sell/trade commercial software are subject to removal.
User avatar
Bill622
Posts: 45
Joined: Fri Feb 20, 2009 6:47 am
Location: Sherbrooke, QC
Contact:

Re: Fractal benchmark

Unread postby Bill622 » Tue Oct 06, 2009 5:18 pm

Any chance to have an executable version for my O2?

Code: Select all

# ./fractbatch: Exec format error. Wrong Architecture.
# file fractbatch
fractbatch:     ELF 64-bit MSB mips-4 dynamic executable (not stripped) MIPS - version 1
#

I'm just curious...

:)
Strangly, girls talks to me when I walk with my O2?!
Image R12k @ 300MHz, 384Mb Ram

User avatar
henrycault
Posts: 131
Joined: Tue Dec 09, 2008 10:30 am
Location: Sherbrooke, Quebec
Contact:

Re: Fractal benchmark

Unread postby henrycault » Sun Oct 11, 2009 2:40 pm

run2.sh on my 2 best machines

Code: Select all

Octane2 2x360mhz

real   1h45m15.03s
user   3h6m42.59s
sys   0m40.77s

Origin 2400 8x250mhz

real   39m43.19s
user   4h16m6.97s
sys   0m24.51s


Anyone can explain to me what is user time? and why it's actually higher on my Origin2400 ?

Oh and another strange behavior, it seems i lose a router link when second test starts... Doesn't seem to affect the system, but it always shuts this link down each time it starts the second test of the run2.sh
front_link.jpg
front_link.jpg (37.56 KiB) Viewed 1068 times
:Octane: 270Mhz SI 384Mb ram :Octane2: 2x360Mhz V6 1.5Gb ram
:O2000: Death by flooding... :(
Amiga A3000D / Full ECS, Kickstart 3.1, 2Mb CHIP/24Mb FAST with 2+18Gb SCSI HD
Amiga A1200 starting to work on this one.

User avatar
bri3d
Posts: 669
Joined: Sat Jun 28, 2008 11:08 am
Location: Boulder, CO

Re: Fractal benchmark

Unread postby bri3d » Sun Oct 11, 2009 8:30 pm

Quick crash course:
"real" time is also commonly called "wallclock" - it's how much actual, measurable time the program spent from when you invoked it to when it finished. Generally measured using a real-time clock in the system, if it's available.
"user" time is an estimate of how much CPU time was spent in user code (i.e. the program itself). It's usually measured by counting instruction cycles.
"system" time is an estimate of how much CPU time was spent in system code (i.e. the kernel and syscalls). Again usually measured by counting instruction cycles.

This is a grossly simplified explanation, but user time is higher on your Origin 2400 because it has slower CPUs - it had to do about 4 hours of processing "work" to match your Octane's 3 hours. It could just slice it up into 8 pieces, so it took less real time.

User avatar
henrycault
Posts: 131
Joined: Tue Dec 09, 2008 10:30 am
Location: Sherbrooke, Quebec
Contact:

Re: Fractal benchmark

Unread postby henrycault » Mon Oct 12, 2009 3:35 pm

Oh! explained like that i can now fully understand the results.

So it also means, that if i had another module (16 cpus total), it would then take about 20mins of real time instead of 40mins, but would still sum up to 4h of user time.
:Octane: 270Mhz SI 384Mb ram :Octane2: 2x360Mhz V6 1.5Gb ram
:O2000: Death by flooding... :(
Amiga A3000D / Full ECS, Kickstart 3.1, 2Mb CHIP/24Mb FAST with 2+18Gb SCSI HD
Amiga A1200 starting to work on this one.

User avatar
Martin Steen
Posts: 159
Joined: Tue Sep 25, 2007 2:00 am
Location: Braunschweig (Germany)
Contact:

Re: Fractal benchmark

Unread postby Martin Steen » Mon Oct 12, 2009 4:12 pm

Bill622 wrote:Any chance to have an executable version for my O2?

Code: Select all

# ./fractbatch: Exec format error. Wrong Architecture.
# file fractbatch
fractbatch:     ELF 64-bit MSB mips-4 dynamic executable (not stripped) MIPS - version 1
#

I'm just curious...

:)


Is the O2 not a 64-bit-machine?
It's 10 years ago that I worked on a O2.

Here is another executable for you, compiled with compiler-switches -n32 -mips3:
http://www.martin-steen.de/sgi/n32/fractbatch.gz

Best regards,
Martin

User avatar
ShadeOfBlue
Moderator
Moderator
Posts: 799
Joined: Tue Nov 25, 2003 12:09 pm
Location: Europe

Re: Fractal benchmark

Unread postby ShadeOfBlue » Mon Oct 12, 2009 11:42 pm

Martin Steen wrote:Is the O2 not a 64-bit-machine?

It has a 32-bit kernel, since there's really no reason to run it in 64-bit mode due to the 1GB memory limit (it can, however, still use 64-bit instructions -- it just lacks the kernel support and libraries that use a 64-bit address space).

It is more efficient to run MIPS4 N32 code on all SGI machines that have an R5k or better CPU inside (and MIPS3 N32 for R4k). Compile it as 64-bit only if your app uses more than 2GB of memory, otherwise it will just cause a slow down and use more memory.

User avatar
Martin Steen
Posts: 159
Joined: Tue Sep 25, 2007 2:00 am
Location: Braunschweig (Germany)
Contact:

Re: Fractal benchmark

Unread postby Martin Steen » Tue Oct 13, 2009 3:04 am

ShadeOfBlue wrote:
Martin Steen wrote:Is the O2 not a 64-bit-machine?

It has a 32-bit kernel, since there's really no reason to run it in 64-bit mode due to the 1GB memory limit (it can, however, still use 64-bit instructions -- it just lacks the kernel support and libraries that use a 64-bit address space).

It is more efficient to run MIPS4 N32 code on all SGI machines that have an R5k or better CPU inside (and MIPS3 N32 for R4k). Compile it as 64-bit only if your app uses more than 2GB of memory, otherwise it will just cause a slow down and use more memory.


So the O2 has a 64-bit CPU, but it cannot run 64-bit code because the kernel and the libraries are
32-bit? That's good to know, if one is going to use posix memory mapping (mmap) with big files.
(on my job I have to deal with satellite images that can be 200GB large).

Best regards, Martin

User avatar
Martin Steen
Posts: 159
Joined: Tue Sep 25, 2007 2:00 am
Location: Braunschweig (Germany)
Contact:

Re: Fractal benchmark

Unread postby Martin Steen » Tue Oct 13, 2009 3:10 am

henrycault wrote:run2.sh on my 2 best machines

Oh and another strange behavior, it seems i lose a router link when second test starts... Doesn't seem to affect the system, but it always shuts this link down each time it starts the second test of the run2.sh
front_link.jpg


That is indeed very strange. There is no network-code in the program.

User avatar
ShadeOfBlue
Moderator
Moderator
Posts: 799
Joined: Tue Nov 25, 2003 12:09 pm
Location: Europe

Re: Fractal benchmark

Unread postby ShadeOfBlue » Tue Oct 13, 2009 4:00 am

Martin Steen wrote:So the O2 has a 64-bit CPU, but it cannot run 64-bit code because the kernel and the libraries are 32-bit?

Yes, although it can still use 64-bit instructions and operate on 64-bit data (e.g. dadd, dsub, dmult, ...) normally in N32 mode :)
Essentially, the 64-bit ABI only enlarges the 'long' and 'pointer' types to 64-bits and thus enables the program to use more than 2GB of RAM.

This document explains the exact differences between N32 and O32 and N64: http://techpubs.sgi.com/library/tpl/cgi-bin/browse.cgi?coll=0650&db=bks&cmd=toc&pth=/SGI_Developer/Mpro_n32_ABI.

That's good to know, if one is going to use posix memory mapping (mmap) with big files.
(on my job I have to deal with satellite images that can be 200GB large).

An Onyx/Origin 3800 system would be more suited to that kind of work if you want to fit the entire dataset within main memory (though an O2 with 1GB of RAM [the maximum for an O2] could map textures up to ~800MB at a time).

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

Re: Fractal benchmark

Unread postby recondas » Tue Oct 13, 2009 4:59 am

henrycault wrote:Oh and another strange behavior, it seems i lose a router link when second test starts... Doesn't seem to affect the system, but it always shuts this link down each time it starts the second test of the run2.sh
After looking at your photo, I don't think you're losing the link - the green LED on the left indicates a successful link, the yellow LED on the right indicates activity <or a fault>. That particular pair of LEDs are associated with node 1 - so it might be possible that node 1 isn't used during the the run2 test.

linkstat or xbstat might give you some additional detail on the router/xbow activity during the test.
***********************************************************************
Welcome to ARMLand - 0/0x0d00
running...(sherwood-root 0607201829)
* InfiniteReality/Reality Software, IRIX 6.5 Release *
***********************************************************************

User avatar
Bill622
Posts: 45
Joined: Fri Feb 20, 2009 6:47 am
Location: Sherbrooke, QC
Contact:

Re: Fractal benchmark

Unread postby Bill622 » Tue Oct 13, 2009 6:18 am

Martin Steen wrote:Is the O2 not a 64-bit-machine?
It's 10 years ago that I worked on a O2.

Here is another executable for you, compiled with compiler-switches -n32 -mips3:
http://www.martin-steen.de/sgi/n32/fractbatch.gz

Best regards,
Martin


Thanks for the new compiled version. Here is my O2 work time for run1.sh:

Code: Select all

real    3m2.04s
user    2m58.60s
sys     0m0.34s


And this is for run2.sh:

Code: Select all

real    3h13m53.64s
user    3h10m24.12s
sys     0m21.15s
Strangly, girls talks to me when I walk with my O2?!
Image R12k @ 300MHz, 384Mb Ram

User avatar
edefault
Posts: 625
Joined: Fri Nov 28, 2003 5:02 am
Location: Germany
Contact:

Re: Fractal benchmark

Unread postby edefault » Thu Oct 29, 2009 5:49 am

Oskar45 wrote:Anyway, for fun I did a <time ./testrun.sh> [with the default sizes/iterations] and get on my Fuel 700Mhz:

Code: Select all

real    0m58.05s
user    0m46.82s
sys     0m0.19s


quite fast, compared to my 600MHz R14000 IP35 Fuel

Code: Select all

real    2m11.07s
user    1m45.13s
sys     0m0.57s


or have I missed something here eventually?
Valueing life is not weakness; disregarding it is not strength. -Mirage-
Image

User avatar
japes
Donor
Donor
Posts: 1006
Joined: Thu Nov 08, 2007 4:35 pm
Location: Lynnwood, WA

Re: Fractal benchmark

Unread postby japes » Thu Oct 29, 2009 2:31 pm

So, I'm a little late to this party, but I was playing around yesterday with it.

I grabbed the initial build and the later N32 build. I don't know if you made any changes between them, but I ran both on a 500 MHz Fuel. I changed the text file so I'm only creating the _8 fractal, I liked it best.

Code: Select all

time ./fractbatch 1600 1024 $Iter $Threads jp.txt
time ./fractbatch32 1600 1024 $Iter $Threads jp-32.txt


N64 fractbench time wrote:real 33m19.41s
user 32m41.94s
sys 0m2.18s


N32 fractbench time wrote:real 28m16.96s
user 27m44.94s
sys 0m1.82s


I thought it was interesting to see, sure enough, 17% longer run time for N64 code.

ShadeOfBlue wrote:It is more efficient to run MIPS4 N32 code on all SGI machines that have an R5k or better CPU inside (and MIPS3 N32 for R4k). Compile it as 64-bit only if your app uses more than 2GB of memory, otherwise it will just cause a slow down and use more memory.
Last edited by japes on Sun Nov 01, 2009 1:33 am, edited 2 times in total.
:O3000: :Fuel: :Tezro: :Tezro: :Octane2: :Octane: :Octane: :Indigo: :Indigo: :Indigo: :Indigo: :O2: :1600SW: :O2: :1600SW: :1600SW: :Indigo2: :Indigo2: :Indigo2: :Indigo2: :Indigo2IMP: :Indy: :Indy: :Indy: :Indy: :O3x0: :O3x02L: :O3x02L:

User avatar
ShadeOfBlue
Moderator
Moderator
Posts: 799
Joined: Tue Nov 25, 2003 12:09 pm
Location: Europe

Re: Fractal benchmark

Unread postby ShadeOfBlue » Sat Oct 31, 2009 5:34 am

japes wrote:I thought it was interesting to see, sure enough, 17% longer run time for N64 code.

No surprise there :)

64-bit code has longer instructions, so it will effectively halve the available L1 and L2 cache and also require more memory fetches. The only benefit of having 64-bit code is, as I said before, the larger address space.

People who come from an x86 background often think that 64-bit code is faster everywhere, but this is really not the case. The only reason why x86_64 code is faster is that they've increased the number of available registers and made some other changes.

hilti
Posts: 3
Joined: Wed Nov 25, 2009 12:53 pm
Contact:

Re: Fractal benchmark

Unread postby hilti » Fri Nov 27, 2009 10:07 am

Hi!

A little bit late, but here are my results from an Indy 4400/200 (128 MB RAM) running IRIX 6.5.22

run1.sh

Code: Select all

real 30m.15.40s
user 24m.19.45s
sys 0m8.99s


Unfortunately tga2jpeg doesn't work on my system.

Cheers
Hilti
Started in October 2009. My collection so far.
:Indy: :Indy: :Octane2: and an Indy Presenter


Return to “SGI: Development”

Who is online

Users browsing this forum: linkdex [Bot] and 1 guest