Sorry folks, my previous question was a dumb one, when I asked where I
could find comparisons of of these....of course there hasn't been time
to do reviews and get results to press/website yet.

I meant it in the future sense. I don't know what publications/web
sites to trust for objective tests. This is what I was asking.....where
should I watch for the straight scoop on which is faster, and which will
be faster in 6 months, a year...?

Thanks, sorry for my miswording.
Jim

On Sun, 29 Jun 2003 12:02:26 -0400, Jim Kroger
<jimkkREMOVEME[at]umich.edu> chose to bless us with the following wisdom:

>Sorry folks, my previous question was a dumb one, when I asked where I
>could find comparisons of of these....of course there hasn't been time
>to do reviews and get results to press/website yet.
>
>I meant it in the future sense. I don't know what publications/web
>sites to trust for objective tests. This is what I was asking.....where
>should I watch for the straight scoop on which is faster, and which will
>be faster in 6 months, a year...?
>
>Thanks, sorry for my miswording.
>Jim

The honest answer is that you will never be able to find what you
want. There are simply too many fudge factors that can be introduced.



--

"Whoever is advising them [Democrats] on gun control
should be shot."

Blaine Rummel, spokesman for the Coalition to
Stop Gun Violence.

On Sun, 29 Jun 2003 12:02:26 -0400, Jim Kroger
<jimkkREMOVEME[at]umich.edu> wrote:

>Sorry folks, my previous question was a dumb one, when I asked where I
>could find comparisons of of these....of course there hasn't been time
>to do reviews and get results to press/website yet.
>
>I meant it in the future sense. I don't know what publications/web
>sites to trust for objective tests. This is what I was asking
>should I watch for the straight scoop on which is faster, and which will
>be faster in 6 months, a year...?
>
>Thanks, sorry for my miswording.
>Jim

Barefeats.com usually has pretty interesting reviews. They recently
did a review that essentially stated the only way the Mac G5 was
faster in Q3 would be if the PC was a Xeon with lower FSB than the
current P4s that normal people buy, if all of its' options were maxxed
out to the very highest, and the Mac's options were put only on
'Normal' (which isn't nearly as stressful) - AND if the PC had a
slower last-gen graphics card too. Here's the review:
http://www.barefeats.com/#quick

Update: Here's the link info with RAM/speed info:

So in the spirit of fairness, we borrowed brand
new P4 3GHz test unit with the same 800MHz FSB
and 400MHz DDR memory as Apple's Pentium test
unit. The only thing we couldn't match was the
graphics card. Our Radeon 9700 is has a slower
clock speed than the Radeon 9800 used by Apple.

Thanks, and here is another link:


http://www.haxial.com/spls-soapbox/apple-powermac-G5/

Jim Kroger <jimkkREMOVEME[at]umich.edu> wrote:
>Thanks, and here is another link:
>
>http://www.haxial.com/spls-soapbox/apple-powermac-G5/

That article is neither trustworthy nor is it even a comparison, thus
making it rather off-topic :).

-- Erick

In article <hv4ufvke5s8s7qav1j201pl3g7fdvijpht[at]4ax.com>,
Mayor of R'lyeh <ev515o[at]hotmail.com> wrote:

> On Sun, 29 Jun 2003 12:02:26 -0400, Jim Kroger
> <jimkkREMOVEME[at]umich.edu> chose to bless us with the following wisdom:
>
> >Sorry folks, my previous question was a dumb one, when I asked where I
> >could find comparisons of of these....of course there hasn't been time
> >to do reviews and get results to press/website yet.
> >
> >I meant it in the future sense. I don't know what publications/web
> >sites to trust for objective tests. This is what I was asking.....where
> >should I watch for the straight scoop on which is faster, and which will
> >be faster in 6 months, a year...?
> >
> >Thanks, sorry for my miswording.
> >Jim
>
> The honest answer is that you will never be able to find what you
> want. There are simply too many fudge factors that can be introduced.



I was afraid of that. But what do you think of this:


Get a <computer A>. Get a <computer B>.

Install Matlab on both computers. Start matlab on both. On both, type:

bench

compare the results.You get several kinds of benchmarks (floating, etc.).

True, Matlab has probably been optimized more for PCs because there are
more PC customers, and the PC version was probably compiled on a
compiler that was more successfully optimal for the PC chip than is
possible for a G5 chip........and though this could change this is
reality right now, and if the G5 is REALLY faster, SIGNIFICANTLY faster,
this reality won't prevent that from becoming apparent.

So I think this is an OK way to go. People over in the matlab newsgroup
are waiting for somebody to get their hands on a G5.

It's esp. a good test for me, since Matlab is the only thing I do for
which these issues really matter, except on a good day when I'm typing
in MS Word REALLY fast....

Thanks for the reply...
Jim

In article <bdnojm$c81$1[at]morgoth.sfu.ca>,
erick[at]sfu.ca (Erick Bryce Wong) wrote:

> Jim Kroger <jimkkREMOVEME[at]umich.edu> wrote:
> >Thanks, and here is another link:
> >
> >http://www.haxial.com/spls-soapbox/apple-powermac-G5/
>
> That article is neither trustworthy nor is it even a comparison, thus
> making it rather off-topic :).
>
> -- Erick


Well, the author does seem a bit adolescent, but it seems he had dug up
some facts which are echoed in other web sites.

Jim

In article <bdnojm$c81$1[at]morgoth.sfu.ca>,
erick[at]sfu.ca (Erick Bryce Wong) wrote:

> Jim Kroger <jimkkREMOVEME[at]umich.edu> wrote:
> >Thanks, and here is another link:
> >
> >http://www.haxial.com/spls-soapbox/apple-powermac-G5/
>
> That article is neither trustworthy nor is it even a comparison, thus
> making it rather off-topic :).
>
> -- Erick

I don't think it's possible that there could be any trustworthy
comparisons because it's rather like comparing Diesel and gasoline
engine horsepower ratings (as someone else pointed out). They don't
jibe. It's probably not possible to devise a test that that measures the
throughput of the computer that doesn't depend on the accuracy (by which
I mean absolute "sameness") or the acceptibility as being absolutely
equal by the various camps involved in the software used to test the
benchmarks.

In other words NO tests can be devised so that everybody, in both camps
will just accept the results with no one challenging the results on the
basis of the efficiency of the compiler used, the software's
optimization for the platforms under test, etc. So. If one is interested
in speed, one just picks the computer which runs fastest with the
software that one most needs to be as speedy as possible. That's the
ultimate test and in the end, the only one which really matters.

--
George Graves

Jim Kroger <jimkkREMOVEME[at]umich.edu> wrote:
>I was afraid of that. But what do you think of this:
>
>Get a <computer A>. Get a <computer B>.
>
>Install Matlab on both computers. Start matlab on both. On both, type:
>
> bench
>
>compare the results.You get several kinds of benchmarks (floating, etc.).
>
>True, Matlab has probably been optimized more for PCs because there are
>more PC customers, and the PC version was probably compiled on a
>compiler that was more successfully optimal for the PC chip than is
>possible for a G5 chip........and though this could change this is
>reality right now, and if the G5 is REALLY faster, SIGNIFICANTLY faster,
>this reality won't prevent that from becoming apparent.

I think that a confounding factor will be the JIT compiler in Matlab
6.5. Since it is unlikely that Mathworks wrote the compiler with the
G5 architecture in mind, the Matlab performance will be at a greater
disadvantage than normal apps, at least until an updated release is
available (and given the non-Aqua nature of the current Mac OS X port,
and the complete lack of any non-x86 Linux version, it is not clear
how committed Mathworks is to the PowerPC architecture in general).

-- Erick

In article
<jimkkREMOVEME-A15738.19291929062003[at]visonmassif.rs.itd.umich.edu>,
Jim Kroger <jimkkREMOVEME[at]umich.edu> wrote:

> In article <hv4ufvke5s8s7qav1j201pl3g7fdvijpht[at]4ax.com>,
> Mayor of R'lyeh <ev515o[at]hotmail.com> wrote:
>
> > On Sun, 29 Jun 2003 12:02:26 -0400, Jim Kroger
> > <jimkkREMOVEME[at]umich.edu> chose to bless us with the following wisdom:
> >
> > >Sorry folks, my previous question was a dumb one, when I asked where I
> > >could find comparisons of of these....of course there hasn't been time
> > >to do reviews and get results to press/website yet.
> > >
> > >I meant it in the future sense. I don't know what publications/web
> > >sites to trust for objective tests. This is what I was asking.....where
> > >should I watch for the straight scoop on which is faster, and which will
> > >be faster in 6 months, a year...?
> > >
> > >Thanks, sorry for my miswording.
> > >Jim
> >
> > The honest answer is that you will never be able to find what you
> > want. There are simply too many fudge factors that can be introduced.
>
>
>
> I was afraid of that. But what do you think of this:
>
>
> Get a <computer A>. Get a <computer B>.
>
> Install Matlab on both computers. Start matlab on both. On both, type:
>
> bench
>
> compare the results.You get several kinds of benchmarks (floating, etc.).
>
> True, Matlab has probably been optimized more for PCs because there are
> more PC customers, and the PC version was probably compiled on a
> compiler that was more successfully optimal for the PC chip than is
> possible for a G5 chip........and though this could change this is
> reality right now, and if the G5 is REALLY faster, SIGNIFICANTLY faster,
> this reality won't prevent that from becoming apparent.

I think that in either case, it won't really tell you a thing. One might
be better for integer performance and the other for floating point, and
the mix of floating point and integer test run on Matlab's 'Bench' test
might still favor one platform over the other even though ultimately the
"slower" one is really the fastest computer. The two platforms are just
too unlike one another for anything to be infered from the results of
such a test (IMHO, of course).

--
George Graves

In article
<jimkkREMOVEME-E6FFD8.19351529062003[at]visonmassif.rs.itd.umich.edu>,
Jim Kroger <jimkkREMOVEME[at]umich.edu> wrote:

> In article <bdnojm$c81$1[at]morgoth.sfu.ca>,
> erick[at]sfu.ca (Erick Bryce Wong) wrote:
>
> > Jim Kroger <jimkkREMOVEME[at]umich.edu> wrote:
> > >Thanks, and here is another link:
> > >
> > >http://www.haxial.com/spls-soapbox/apple-powermac-G5/
> >
> > That article is neither trustworthy nor is it even a comparison, thus
> > making it rather off-topic :).
> >
> > -- Erick
>
>
> Well, the author does seem a bit adolescent, but it seems he had dug up
> some facts which are echoed in other web sites.
He dug up something but i would not call them facts. :) He is obviously
an AMD fanboy. He directly quotes AMD benchmarks and then goes a head to
compare but to a mythical Single 2 Ghz G5.

He does not have any original info on his site.

> Jim

--
Ari Ukkonen

In article <bdnv4c$kc9$1[at]morgoth.sfu.ca>,
erick[at]sfu.ca (Erick Bryce Wong) wrote:

(and given the non-Aqua nature of the current Mac OS X port,
> and the complete lack of any non-x86 Linux version, it is not clear
> how committed Mathworks is to the PowerPC architecture in general).


Erick,

I've been asking various Linux vs OSX and Mac vs PC questions for a
couple months trying to decide which way to go when I set up my lab.

You may have just said something that trumps everything. Matlab is what
my computer(s) must do, as well as possible, so do you mind if I explore
this a little further?

I did not know Matlab's OSX release did not support Aqua. Do you have
any further information/thoughts on whether Mathworks will release a
full-blown Mac version, and whether the current aqua-less version
suffers for being so?

When I am concerned about getting the speediest machine, I mean
speediest at big (multi-day) Matlab jobs.


Thanks much,
Jim


(ps - hope nobody minds if I for obvious reasons post this to the matlab
group as well)

In article
<gmgraves-FE10E9.16531129062003[at]newssvr15-ext.news.prodigy.com>,
George Graves <gmgraves[at]pacbell.net> wrote:

> In article <bdnojm$c81$1[at]morgoth.sfu.ca>,
> erick[at]sfu.ca (Erick Bryce Wong) wrote:
>
> > Jim Kroger <jimkkREMOVEME[at]umich.edu> wrote:
> > >Thanks, and here is another link:
> > >
> > >http://www.haxial.com/spls-soapbox/apple-powermac-G5/
> >
> > That article is neither trustworthy nor is it even a comparison, thus
> > making it rather off-topic :).
> >
> > -- Erick
>
> I don't think it's possible that there could be any trustworthy
> comparisons because it's rather like comparing Diesel and gasoline
> engine horsepower ratings (as someone else pointed out). They don't
> jibe. It's probably not possible to devise a test that that measures the
> throughput of the computer that doesn't depend on the accuracy (by which
> I mean absolute "sameness") or the acceptibility as being absolutely
> equal by the various camps involved in the software used to test the
> benchmarks.
>
> In other words NO tests can be devised so that everybody, in both camps
> will just accept the results with no one challenging the results on the
> basis of the efficiency of the compiler used, the software's
> optimization for the platforms under test, etc. So. If one is interested
> in speed, one just picks the computer which runs fastest with the
> software that one most needs to be as speedy as possible. That's the
> ultimate test and in the end, the only one which really matters.



I agree, and then for me it's Matlab, who provide a benchmark program
in bench.

Jim

In article
<jimkkREMOVEME-E6FFD8.19351529062003[at]visonmassif.rs.itd.umich.edu>,
Jim Kroger <jimkkREMOVEME[at]umich.edu> wrote:

> In article <bdnojm$c81$1[at]morgoth.sfu.ca>,
> erick[at]sfu.ca (Erick Bryce Wong) wrote:
>
> > Jim Kroger <jimkkREMOVEME[at]umich.edu> wrote:
> > >Thanks, and here is another link:
> > >
> > >http://www.haxial.com/spls-soapbox/apple-powermac-G5/
> >
> > That article is neither trustworthy nor is it even a comparison, thus
> > making it rather off-topic :).
> >
> > -- Erick
>
>
> Well, the author does seem a bit adolescent, but it seems he had dug up
> some facts which are echoed in other web sites.
>
> Jim

Show some of these "facts" and show us where they are echoed other than
sites that are referring to his...

--
Alan Baker
Vancouver, British Columbia
"If you raise the ceiling 4 feet, move the fireplace from that wall
to that wall, you'll still only get the full stereophonic effect
if you sit in the bottom of that cupboard."

In article <alangbaker-8AFEF2.19284629062003[at]news.telus.net>,
Alan Baker <alangbaker[at]telus.net> wrote:

> In article
> <jimkkREMOVEME-E6FFD8.19351529062003[at]visonmassif.rs.itd.umich.edu>,
> Jim Kroger <jimkkREMOVEME[at]umich.edu> wrote:
>
> > In article <bdnojm$c81$1[at]morgoth.sfu.ca>,
> > erick[at]sfu.ca (Erick Bryce Wong) wrote:
> >
> > > Jim Kroger <jimkkREMOVEME[at]umich.edu> wrote:
> > > >Thanks, and here is another link:
> > > >
> > > >http://www.haxial.com/spls-soapbox/apple-powermac-G5/
> > >
> > > That article is neither trustworthy nor is it even a comparison, thus
> > > making it rather off-topic :).
> > >
> > > -- Erick
> >
> >
> > Well, the author does seem a bit adolescent, but it seems he had dug up
> > some facts which are echoed in other web sites.
> >
> > Jim
>
> Show some of these "facts" and show us where they are echoed other than
> sites that are referring to his...



Nah.....

On Sun, 29 Jun 2003 19:29:19 -0400, Jim Kroger
<jimkkREMOVEME[at]umich.edu> chose to bless us with the following wisdom:

>In article <hv4ufvke5s8s7qav1j201pl3g7fdvijpht[at]4ax.com>,
> Mayor of R'lyeh <ev515o[at]hotmail.com> wrote:
>
>> On Sun, 29 Jun 2003 12:02:26 -0400, Jim Kroger
>> <jimkkREMOVEME[at]umich.edu> chose to bless us with the following wisdom:
>>
>> >Sorry folks, my previous question was a dumb one, when I asked where I
>> >could find comparisons of of these....of course there hasn't been time
>> >to do reviews and get results to press/website yet.
>> >
>> >I meant it in the future sense. I don't know what publications/web
>> >sites to trust for objective tests. This is what I was asking.....where
>> >should I watch for the straight scoop on which is faster, and which will
>> >be faster in 6 months, a year...?
>> >
>> >Thanks, sorry for my miswording.
>> >Jim
>>
>> The honest answer is that you will never be able to find what you
>> want. There are simply too many fudge factors that can be introduced.
>
>
>
>I was afraid of that. But what do you think of this:
>
>
>Get a <computer A>. Get a <computer B>.
>
>Install Matlab on both computers. Start matlab on both. On both, type:
>
> bench
>
>compare the results.You get several kinds of benchmarks (floating, etc.).
>
>True, Matlab has probably been optimized more for PCs because there are
>more PC customers, and the PC version was probably compiled on a
>compiler that was more successfully optimal for the PC chip than is
>possible for a G5 chip........and though this could change this is
>reality right now, and if the G5 is REALLY faster, SIGNIFICANTLY faster,
>this reality won't prevent that from becoming apparent.
>
>So I think this is an OK way to go. People over in the matlab newsgroup
>are waiting for somebody to get their hands on a G5.
>
>It's esp. a good test for me, since Matlab is the only thing I do for
>which these issues really matter, except on a good day when I'm typing
>in MS Word REALLY fast....
>
>Thanks for the reply...
>Jim

If Matlab is the app you use the most then it will provide the best
benchmark for you definitely. Its pretty much a YMMV kind of thing.


--

"Whoever is advising them [Democrats] on gun control
should be shot."

Blaine Rummel, spokesman for the Coalition to
Stop Gun Violence.

The only way to get an even somewhat legitimate comparison is to run
some application you want to run fast on both machines. Apache might
serve pages much faster on box 'a' while box 'b' may raytrace much
faster. And even more so, it will come down to what compiler &
options were used on a specific binary in a lot of cases.

The 970 is a very good CPU though.

dave

Oops! Jim Kroger <jimkkREMOVEME[at]umich.edu> was seen spray-painting on a wall:
> It's esp. a good test for me, since Matlab is the only thing I do for
> which these issues really matter, except on a good day when I'm typing
> in MS Word REALLY fast....

Which means that's likely a good benchmark for you.

For someone else, the better benchmark might involve:
- Running a set of transactions through a database;
- Running an X benchmark;
- Some other _relevant_ task, in the eyes of the beholder...
--
(reverse (concatenate 'string "gro.gultn" "[at]" "enworbbc"))
http://www.ntlug.org/~cbbrowne/spiritual.html
"Problem solving under linux has never been the circus that it is
under AIX." -- Pete Ehlke in comp.unix.aix

Jim Kroger wrote:
> I've been asking various Linux vs OSX and Mac vs PC questions for a
> couple months trying to decide which way to go when I set up my lab.

In a way that has only started entertaining flame wars but generated
no useful information, I might point out.

> You may have just said something that trumps everything. Matlab is what
> my computer(s) must do, as well as possible, so do you mind if I explore
> this a little further?

Yes, if it's in c.s.mac.advocacy or c.s.linux.misc. c.soft-sys.matlab
is a good place. Apple's sci tech list is better.

> I did not know Matlab's OSX release did not support Aqua.

Yes you did, because I told you. It is also clearly stated on
MathWorks web page, where I sent you. This is a piece of Unix
software, identical in all respects to the same software on any other
Unix(like) system. You know that too.

> Do you have
> any further information/thoughts on whether Mathworks will release a
> full-blown Mac version,

They have already. It is fully native and uses OS X's built in java
and X11.

> and whether the current aqua-less version
> suffers for being so?

Only the user interface is not a pretty as an Aqua interface. I never
use the gui anyway. Use emacs and Terminal or Xterm.

Now, for what you really want to know, from the SciTech list, where
you should be asking these questions:

Resent-From: Surajit Nundy <nundys[at]neuro.duke.edu>
From: Surajit Nundy <nundys[at]neuro.duke.edu>
Date: Mon Jun 30, 2003 8:09:16 AM US/Pacific
Resent-To: scitech[at]lists.apple.com
To: beta <beta[at]mathworks.com>
Subject: G5 benchmarks on MATLAB


I attended WWDC (conference at which Apple unveiled the G5 computer)
and I ran bench (MATLAB's cross-platform benchmarking function) on it
using MATLAB 6.5 R13. The results are below.

The G5 numbers for bench were:


LU FFT ODE Sparse 2-D
3-D
0.3097 1.1282 0.4329 0.5004 0.9657 0.7893

For comparison, the Mathworks supplies the following benchmark numbers
on other machines.

LU FFT ODE Sparse 2-D
3-D
Intel Pentium3 700MHz Laptop 2.0933 2.3500 1.2020 1.8230 1.7290
2.5540
Windows 2000, ATI Mobility
Intel Pentium4 2.0GHz 0.8440 1.6410 0.5620 0.7970 0.9220
0.5940
Windows 2000, Matrox G450
AMD Athlon 600Mhz 1.6819 3.0545 1.9673 2.5462 2.8626
3.8693
Linux, Pentium 3 BLAS, Software OpenGL
AMD Athlon 1800+ Dual 0.6702 1.3188 0.6752 0.9850 0.8665
0.4710
Linux, Matrox G450
HP PA-8700 750MHz 0.5040 2.2458 0.7684 1.3686 1.3578
2.4139
HP-UX, Visualize FX10
Compaq Alpha 21264B 834Mhz 0.8588 0.9543 1.0236 1.2184 1.3656
1.0363
Tru64, PowerStorm 350, Software OpenGL
IBM Power3 450Mhz 0.5956 1.7016 1.1413 1.4862 2.0769
3.0919
AIX, GXT4000P
SGI R12000 400Mhz 1.4594 3.4102 2.7477 2.0578 5.3070
2.0957
IRIX64, V10 graphics
Sun UltraSPARC-II 750Mhz 1.0215 2.7104 1.5564 2.1607 1.8234
3.5861
Solaris, PGX64
Macintosh G4 1.0GHz Dual 0.9688 1.7966 0.7920 1.4270 1.8071
3.1606
OS X, Graphics uses X-windows, OroborOSX and Mesa, GeForce4 MX

Christopher Browne wrote:
For someone else, the better benchmark might involve:

> - Running a set of transactions through a database;
> - Running an X benchmark;
> - Some other _relevant_ task, in the eyes of the beholder...

Exactly so......arguing over SPEC figures for the overwhelming majority
of users is just so much mental masturbation. The new Macs are good news
for Mac users, period. If they greatly enhance the user experience for Mac
users, Apple will succeed.........nothing anybody posts on c.s.m.a will change
that market reality.


--
Joseph Crowe

erick[at]sfu.ca (Erick Bryce Wong) writes:

[...]

> Mathworks has had somewhat spotty Mac support in recent history (but
> keep in mind I'm coming from the standpoint of a Mac user who takes
> easy offense at any injustice :). They discontinued Matlab for
> Classic Mac OS a while back, only to announce their return to Mac OS
> X a year later. I'd expected a nice Aqua port, but instead they
> delivered an X11-based port requiring a third-party (bundled) X11
> server.

http://www.mathworks.com/products/system.shtml/Macintosh


--
Edward Dodge

/Confabulation Consulting/

esalathe[at]cascade.org (Eric Salathe) writes:


[...]

> I attended WWDC (conference at which Apple unveiled the G5 computer)
> and I ran bench (MATLAB's cross-platform benchmarking function) on it
> using MATLAB 6.5 R13. The results are below.
>
> The G5 numbers for bench were:
>
>
> LU FFT ODE Sparse 2-D 3-D 0.3097 1.1282 0.4329 0.5004 0.9657 0.7893
>
> For comparison, the Mathworks supplies the following benchmark numbers
> on other machines.
>
> LU FFT ODE Sparse 2-D 3-D Intel
> Pentium3 700MHz Laptop 2.0933 2.3500 1.2020 1.8230 1.7290 2.5540
> Windows 2000, ATI Mobility Intel Pentium4 2.0GHz 0.8440 1.6410 0.5620
> 0.7970 0.9220 0.5940 Windows 2000, Matrox G450 AMD Athlon 600Mhz
> 1.6819 3.0545 1.9673 2.5462 2.8626 3.8693 Linux, Pentium 3 BLAS,
> Software OpenGL AMD Athlon 1800+ Dual 0.6702 1.3188 0.6752 0.9850
> 0.8665 0.4710 Linux, Matrox G450 HP PA-8700 750MHz 0.5040 2.2458
> 0.7684 1.3686 1.3578 2.4139 HP-UX, Visualize FX10 Compaq Alpha 21264B
> 834Mhz 0.8588 0.9543 1.0236 1.2184 1.3656 1.0363 Tru64, PowerStorm
> 350, Software OpenGL IBM Power3 450Mhz 0.5956 1.7016 1.1413 1.4862
> 2.0769 3.0919 AIX, GXT4000P SGI R12000 400Mhz 1.4594 3.4102 2.7477
> 2.0578 5.3070 2.0957 IRIX64, V10 graphics Sun UltraSPARC-II 750Mhz
> 1.0215 2.7104 1.5564 2.1607 1.8234 3.5861 Solaris, PGX64 Macintosh G4
> 1.0GHz Dual 0.9688 1.7966 0.7920 1.4270 1.8071 3.1606 OS X, Graphics
> uses X-windows, OroborOSX and Mesa, GeForce4 MX

EMACS Auto-Fill can be a terrible, terrible thing:

LU FFT ODE Sparse 2-D 3-D Benchmark (assumed) sorted by speed:

A) 0.3097 1.1282 0.4329 0.5004 0.9657 0.7893
F) 0.5040 2.2458 0.7684 1.3686 1.3578 2.4139
H) 0.5956 1.7016 1.1413 1.4862 2.0769 3.0919
E) 0.6702 1.3188 0.6752 0.9850 0.8665 0.4710
C) 0.8440 1.6410 0.5620 0.7970 0.9220 0.5940
G) 0.8588 0.9543 1.0236 1.2184 1.3656 1.0363
K) 0.9688 1.7966 0.7920 1.4270 1.8071 3.1606
J) 1.0215 2.7104 1.5564 2.1607 1.8234 3.5861
I) 1.4594 3.4102 2.7477 2.0578 5.3070 2.0957
D) 1.6819 3.0545 1.9673 2.5462 2.8626 3.8693
B) 2.0933 2.3500 1.2020 1.8230 1.7290 2.5540

A) G5 G) HP-UX
Visualize FX10
B) Intel P3 700MHz Laptop Compaq Alpha 21264B 834Mhz

C) Windows 2000 H) Tru64 PowerStorm 350
ATI Mobility Software OpenGL
Intel P4 2.0GHz IBM Power3 450Mhz

D) Windows 2000 I) AIX GXT4000P
Matrox G450 SGI R12000 400Mhz
AMD Athlon 600Mhz
J) IRIX64
E) Linux, Pentium 3 BLAS V10 graphics
Software OpenGL Sun UltraSPARC-II 750Mhz
AMD Athlon 1800+ Dual
K) Solaris PGX64
F) Linux Macintosh G4 1.0GHz Dual
Matrox G450 GeForce4 MX
HP PA-8700 750MHz OS X
Graphics uses X-windows
OroborOSX and Mesa

I don't know what happened with K)... looks like a Score has been
added or removed here... 3 names for only two scores.

Also, what speed G5 are we talking about? And is the app DP-aware?

--
Edward Dodge

/Confabulation Consulting/

I overstated:
> > often the SPEC scores are with compilers that no one
> > uses except to run SPEC scores (eg icc/ifc -- all scientific code on
> > linux clusters uses portland, lahey, or gcc).

James objected:
> ALL? You've taken a poll?

Sorry, writing too fast. I have not really delved beyond the models
used in my own field. All the code I'm familiar with, global climate
models, mesoscale weather models, hydrologic models, developed at NCAR
or the University of Washington, use these compilers. Intel compilers
may well be used in other disciplines, such as computational fluid
dynamics or weapons design, which have very different sorts of models.

Eric Salathe

Edward Dodge wrote:
>
> EMACS Auto-Fill can be a terrible, terrible thing:
>
> LU FFT ODE Sparse 2-D 3-D Benchmark (assumed) sorted by speed:
>
> A) 0.3097 1.1282 0.4329 0.5004 0.9657 0.7893
> F) 0.5040 2.2458 0.7684 1.3686 1.3578 2.4139
> H) 0.5956 1.7016 1.1413 1.4862 2.0769 3.0919
> E) 0.6702 1.3188 0.6752 0.9850 0.8665 0.4710
> C) 0.8440 1.6410 0.5620 0.7970 0.9220 0.5940
> G) 0.8588 0.9543 1.0236 1.2184 1.3656 1.0363
> K) 0.9688 1.7966 0.7920 1.4270 1.8071 3.1606
> J) 1.0215 2.7104 1.5564 2.1607 1.8234 3.5861
> I) 1.4594 3.4102 2.7477 2.0578 5.3070 2.0957
> D) 1.6819 3.0545 1.9673 2.5462 2.8626 3.8693
> B) 2.0933 2.3500 1.2020 1.8230 1.7290 2.5540
>
> A) G5 G) HP-UX
> Visualize FX10
> B) Intel P3 700MHz Laptop Compaq Alpha 21264B 834Mhz
>
> C) Windows 2000 H) Tru64 PowerStorm 350
> ATI Mobility Software OpenGL
> Intel P4 2.0GHz IBM Power3 450Mhz
>
> D) Windows 2000 I) AIX GXT4000P
> Matrox G450 SGI R12000 400Mhz
> AMD Athlon 600Mhz
> J) IRIX64
> E) Linux, Pentium 3 BLAS V10 graphics
> Software OpenGL Sun UltraSPARC-II 750Mhz
> AMD Athlon 1800+ Dual
> K) Solaris PGX64
> F) Linux Macintosh G4 1.0GHz Dual
> Matrox G450 GeForce4 MX
> HP PA-8700 750MHz OS X
> Graphics uses X-windows
> OroborOSX and Mesa

Those descriptions look screwy to me, starting with "F" for sure. A
Matrox card and Linux on a PA-RISC box? Maybe, but not likely! I think
the HP PA-8700 description goes with the HP-UX, Visualize FX10 lines
from "G". In "H" you've got an IBM chip running Tru64 (an Alpha O/S)
while "I" indicates IBM's AIX on SGI hardware...those are clearly not
right. I think the descriptions go like this:

A) G5 G) Compaq Alpha 21264B 834Mhz
Tru64
B) Intel P3 700MHz Laptop PowerStorm 350
Windows 2000 Software OpenGL
ATI Mobility
H) IBM Power3 450Mhz
C) Intel P4 2.0GHz AIX
Windows 2000 GXT4000P
Matrox G450
I) SGI R12000 400Mhz
D) AMD Athlon 600Mhz IRIX64
Linux, Pentium 3 BLAS V10 graphics
Software OpenGL
J) Sun UltraSPARC-II 750Mhz
E) AMD Athlon 1800+ Dual Solaris
Linux PGX 64
Matrox G450
K) Macintosh G4 1.0GHz Dual
F) HP PA-8700 750MHz GeForce4 MX
HP-UX OS X
Visualize FX10 Graphics uses X-windows
OroborOSX and Mesa

That at least matches CPUs with operating systems and graphics cards
that make sense for each platform.

The way I read the original article (now snipped from the post), reading
accross it was: CPU, six columns of results, then notes about O/S and
graphics and whatever else. Some weird wrapping happened in there, for
sure. But then the table presented above seems to have included the
notes about O/S and graphics with the wrong CPU. (Unless, of course,
IBM has ported AIX to SGI hardware for "I" and the Sun CPU in "J" was
running with an SGI O/S and graphics card. ;)

Since my above post seems to have gotten garbled for some folks,
here's the benchmark results again. Note that these are second hand,
posted to the Apple SciTech list. For further discussion, go there.
Note that this is Matlab compiled on the G4 run on the G5. Also,
Matlab computations only use one cpu. For those not on the Matlab
group, the first four columns are time to complete standardized
computations; the final two are times for rendering a sequence of
figures. Since the test G5 did not hav a finalized graphics setup,
these final columns should improve.

LU FFT ODE Sparse 2-D
3-D
a) Apple G5 2 GHz 0.3097 1.1282 0.4329 0.5004 0.9657
0.7893
b) Intel PIII 700 MHz 2.0933 2.3500 1.2020 1.8230 1.7290
2.5540
c) Intel P4 2.0 GHz 0.8440 1.6410 0.5620 0.7970 0.9220
0.5940
d) AMD Athlon 600 MHz 1.6819 3.0545 1.9673 2.5462 2.8626
3.8693
e) AMD Athlon 1800+ 0.6702 1.3188 0.6752 0.9850 0.8665
0.4710
f) HP PA-8700 750MHz 0.5040 2.2458 0.7684 1.3686 1.3578
2.4139
g) Compaq Alpha 834Mhz 0.8588 0.9543 1.0236 1.2184 1.3656
1.0363
h) IBM Power3 450Mhz 0.5956 1.7016 1.1413 1.4862 2.0769
3.0919
i) SGI R12000 400Mhz 1.4594 3.4102 2.7477 2.0578 5.3070
2.0957
j) Sun UltraSPARC-II 750Mhz 1.0215 2.7104 1.5564 2.1607 1.8234
3.5861
k) Macintosh G4 1.0GHz Dual 0.9688 1.7966 0.7920 1.4270 1.8071
3.1606

System details:

a) Apple G5 dual 2 GHz OS X 10.2.7 2 GB RAM Unknown graphics card with
buggy drivers
b) Intel Pentium3 700MHz Laptop Windows 2000, ATI Mobility
c) Intel Pentium4 2.0GHz Windows 2000, Matrox G450
d) AMD Athlon 600Mhz Linux, Pentium 3 BLAS, Software OpenGL
d) AMD Athlon 1800+ Dual Linux, Matrox G450
f) HP PA-8700 750MHz HP-UX, Visualize FX10
g) Compaq Alpha 21264B 834Mhz Tru64, PowerStorm 350, Software OpenGL
h) IBM Power3 450Mhz AIX, GXT4000P
i) SGI R12000 400Mhz IRIX64, V10 graphics
j) Sun UltraSPARC-II 750Mhz Solaris, PGX64
k) Macintosh G4 1.0GHz Dual OS X, Graphics uses X-windows, OroborOSX
and Mesa, GeForce4 MX

Take 2....

Since my above post seems to have gotten garbled for some folks,
here's the benchmark results again. Note that these are second hand,
posted to the Apple SciTech list. For further discussion, go there.
Note that this is Matlab compiled on the G4 run on the G5. Also,
Matlab computations only use one cpu. For those not on the Matlab
group, the first four columns are time to complete standardized
computations; the final two are times for rendering a sequence of
figures. Since the test G5 did not hav a finalized graphics setup,
these final columns should improve.

LU FFT ODE Sparse 2-D 3-D
Apple G5 2 GHz 0.3097 1.1282 0.4329 0.5004 0.9657 0.7893
Intel PIII 700 MHz 2.0933 2.3500 1.2020 1.8230 1.7290 2.5540
Intel P4 2.0 GHz 0.8440 1.6410 0.5620 0.7970 0.9220 0.5940
AMD Athlon 600 MHz 1.6819 3.0545 1.9673 2.5462 2.8626 3.8693
AMD Athlon 1800+ 0.6702 1.3188 0.6752 0.9850 0.8665 0.4710
HP PA-8700 750MHz 0.5040 2.2458 0.7684 1.3686 1.3578 2.4139
Compaq Alpha 834Mhz 0.8588 0.9543 1.0236 1.2184 1.3656 1.0363
IBM Power3 450Mhz 0.5956 1.7016 1.1413 1.4862 2.0769 3.0919
SGI R12000 400Mhz 1.4594 3.4102 2.7477 2.0578 5.3070 2.0957
Sun UltSPARC-II 750Mhz 1.0215 2.7104 1.5564 2.1607 1.8234 3.5861
Mac G4 1.0GHz Dual 0.9688 1.7966 0.7920 1.4270 1.8071 3.1606

System details:

a) Apple G5 dual 2 GHz OS X 10.2.7 2 GB RAM Unknown graphics card with
buggy drivers
b) Intel Pentium3 700MHz Laptop Windows 2000, ATI Mobility
c) Intel Pentium4 2.0GHz Windows 2000, Matrox G450
d) AMD Athlon 600Mhz Linux, Pentium 3 BLAS, Software OpenGL
d) AMD Athlon 1800+ Dual Linux, Matrox G450
f) HP PA-8700 750MHz HP-UX, Visualize FX10
g) Compaq Alpha 21264B 834Mhz Tru64, PowerStorm 350, Software OpenGL
h) IBM Power3 450Mhz AIX, GXT4000P
i) SGI R12000 400Mhz IRIX64, V10 graphics
j) Sun UltraSPARC-II 750Mhz Solaris, PGX64
k) Macintosh G4 1.0GHz Dual OS X, Graphics uses X-windows, OroborOSX
and Mesa, GeForce4 MX

"Eric Salathe" <esalathe[at]cascade.org> wrote in message
news:e7d80033.0307011040.38fe8ad1[at]posting.google.com...
> I overstated:
> > > often the SPEC scores are with compilers that no one
> > > uses except to run SPEC scores (eg icc/ifc -- all scientific code on
> > > linux clusters uses portland, lahey, or gcc).
>
> James objected:
> > ALL? You've taken a poll?
>
> Sorry, writing too fast. I have not really delved beyond the models
> used in my own field. All the code I'm familiar with, global climate
> models, mesoscale weather models, hydrologic models, developed at NCAR
> or the University of Washington, use these compilers. Intel compilers
> may well be used in other disciplines, such as computational fluid
> dynamics or weapons design, which have very different sorts of models.

Actually, one of the best (on Windows, at least) Fortran (the language you
should be working in :) )
compilers is Compaq Visual Fortran. That product is not Intel's and uses
their very good Fortran compiler.
I use CVF at the present time.

JCS

This may be a moot point since the P4 is soon to become an obsolete
processor.

http://www.tomshardware.com/cpu/20030623/index.html

Tom

In article <qqsagv4pp929rib90clsa2thp42p2npfrf[at]4ax.com>,
- ILUVJazz - <ILVJazz[at]NYOB.com> wrote:

> This may be a moot point since the P4 is soon to become an obsolete
> processor.
>
> http://www.tomshardware.com/cpu/20030623/index.html
>

That's true. It will become obsolete in August when the G5s start
shipping.

Thanks for pointing that out.

esalathe[at]cascade.org (Eric Salathe) wrote in message news:<e7d80033.0306301533.62f7169[at]posting.google.com>...


> I did not know Matlab's OSX release did not support Aqua.

Yes you did, because I told you. It is also clearly stated on
MathWorks web page, where I sent you. This is a piece of Unix
software, identical in all respects to the same software on any other
Unix(like) system. You know that too.



-------------

Eric, use your imagination to formulate a reply from me to you on this
one. It seems to work very well.

Jim