Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

dual g4 'trounced' by piii

P

Pants

macrumors regular
Original poster
Aug 21, 2001
194
9
ouch! :(

A fairly written article, and although the g4 wasnt exactly 'trounced' by the p3, it is a little embarrassing.

http://www.heise.de/ct/english/02/05/182/

"Thus according to the SPEC results the Mac is so far less well-suited for scientific applications and is no 'Supercomputer,' not by a long shot."


wherefore art thou g5?
 
Only one processor was used in this test, which is either stated in this article in a connected piece--I can't recall where I found that detail. So it's a single G4 that's being benchmarked. Also, none of the G4's AltiVec power was put to the test here--power that _is_ used in many modern OS X applications (and the OS itself, to a large degree).

This bench tells me little.



blakespot
 
Prioities

I am an engineer, so at work I use a box that was built for running CAD software. It just happens to be a PIII with WinNT. I have had this machine about 8 months and I think it is a 1 GHz machine, with 512 mb ram, but I am not sure about the Proc. It really doesn't matter so much to me. The rate limiting factor on my work flow is still my typing skills and my brain. If i needed more horse power, we have a machine dedicated to Finite Element Analysis (FEA) that the engineers can use. When I come home, I want to relax, turn on my internet radion via itunes, surf the web, write reports/spreadsheets for school (in the Mac OS X- Office Suite) , and on weekends, I might toy with iMovie. I love synincg my iPod with a custom mix that I made for my next trip to the gym (oh and all day at work, my iPod sits next to me so I can stave off engineering boredom for another 3min and 20 sec). When I go somewhere with my wife, family or friends, I have the option of taking my digital camera along and snapping pictures, then getting home and plugging in the camera and letting iPhoto organize them. I really dont hate pc's, they are so utilitarian though (like a hammer)... it is hard to have any emotion about them. But, I really dont enjoy windows (and in fairness, I havent tried XP).

Why did I post this here? Because for me, processor speed is an issue, but b no means the most important issue. I care more about how much I like the OS and what value-added items the comp brings to me. I switched to Mac OS about a year ago, because I wanted to try something that after a friend showed me (his G3 laptop) that I thought was way cool. I love my G4, it does what I want. I am even seriously debating getting a laptop (after 3/24... i just gotta know what they have in mind). I want to take my fun OS on the road... I want to show others how fun and useful OS X with the iApps can be.

So... in summary, for all you legitimate pros out there, I feel for you. It must be very difficult to wait for upgrades, but for me and many other customers it isnt so much of an issue. What counts to me is the experience and value of the comp. Getting people to switch does mean fighting a MegaHertz Myth somewhat, but that is changing. Many people stopped upgrading their comps proc because they did not see a need. Microsoft artificially tries to generate a need by releasing cookie cutter clone updates like Windows ME or Win 98. Essentially, these contained more drivers and non-value added features, in my opinion. With Win 2000 and XP their may be some value and cause some to upgrade. I really think we as apple fans should not fight a PC vs MAc war with our famillies, coworkers and friends, but rather show people what is so useful and cool about the Mac OS.

<stop>
 
G4 1Ghz = 23% faster than PIII 1Ghz

Did you read the article?
The article is very informative.

First, the article shows that the 1GHz G4 under OS X runs about 23% faster than a 1Ghz PIII running Windows (note: The Pentium 3 generally has better performance than the Pentium 4 at the same clock speed.)

Next, the article shows that the G4 is lacking a good Fortran compiler. In summary, if you primarily write your programs in Fortran, you should use Intel hardware. (That's like saying, "so if you only use your car for towing boats, a big SUV is your best bet for performance", or alternatively, "if you live in the 1970s and early 80's - when Fortran was popular - the Pentium is surely a better processor because of its superior Fortran compiler performance".

Of course, that doesn't mean that a 1Ghz G4 will compete head-to-head with a 2Ghz Pentium 4 for 90% of general computing tasks. That's why Apple/Motorola/IBM need to:

1) Ramp up the processor speed of the G4
2) Get the G5 out the door
 
Originally posted by blakespot
Only one processor was used in this test, which is either stated in this article in a connected piece--I can't recall where I found that detail. So it's a single G4 that's being benchmarked. Also, none of the G4's AltiVec power was put to the test here--power that _is_ used in many modern OS X applications (and the OS itself, to a large degree).

This bench tells me little.



blakespot
Click to expand...

absofts compiler is altivec aware and optimises for it - it is also multi processor aware. Most academic/research programs are still written in fortran - most codes for seismic analysis are fortran based for example - and ths type of signal analysis is extremely processor intensive, and as such are an good comparison of processor performance (especially now since gcc comes with osX).

And irrepsective of one or two processors, the floating point performance aint exactly great eh? the p3 is around 60% faster- and this folks, is an aged p3!!(and fp performance is usually a better indicator of real world performance than integer based benchmarks) ok, I have no idea how 'good' or otherwise the compiler is, (doesnt gcc come with a fortran to c convertor?) but if this benchmark tells you nothing, Im not sure theres a benchmark that does! Im a little shocked by these results - no saddened.
 
Originally posted by Pants


absofts compiler is altivec aware and optimises for it - it is also multi processor aware. Most academic/research programs are still written in fortran - most codes for seismic analysis are fortran based for example - and ths type of signal analysis is extremely processor intensive, and as such are an good comparison of processor performance (especially now since gcc comes with osX).

And irrepsective of one or two processors, the floating point performance aint exactly great eh? the p3 is around 60% faster- and this folks, is an aged p3!!(and fp performance is usually a better indicator of real world performance than integer based benchmarks) ok, I have no idea how 'good' or otherwise the compiler is, (doesnt gcc come with a fortran to c convertor?) but if this benchmark tells you nothing, Im not sure theres a benchmark that does! Im a little shocked by these results - no saddened.
Click to expand...

You say that "Most academic/research programs are still written in fortran". Do you have a reference for this?
 
Someone posted the same thing at Macnn, and I'll say the same thing.

You mean I wasted $3k on a suck ass dual gig?

k
 
Ha!

Originally posted by Pants

(and fp performance is usually a better indicator of real world performance than integer based benchmarks)
Click to expand...

Real world performance for what exactly? There are many, many tasks that require absolutely no fp calculations. Just as there are many tasks that DO require fp calculations.

That statement is ill-informed and lacks any quantifiers so as to make it meaningful. Also, the SPEC suite is still in its infancy for this platform. As the article says, there isn't a great Fortran compiler and gcc ususally requires tweaks (or they use a more capable compiler) in order to make it run well.

That being said, the fp speed of the G4 *does* lag behind that of pentiums. And it is an area that needs a remedy. Also, I believe that Motorola need to bump speeds across the bored. Their chips are lagging a bit.

I still wouldn't take this bench as proof of the pentium's victory over PowerPC. The war hasn't been won yet...

Matthew
 
Looking at that...

...it seems that the performance hit is due to two things. Double precision floating point, and the Fortran compiler. Altivec loses half of its gain when it's doing double precision because it can only do two operations per pipeline instead of four (in other words, for every day single precision apps, the G4 will do *much* better). Hopefully the G5 will have 256 bit Altivec so it'll have good double precision performance. Also, the second processor was turned off for the tests so even with the problems of the bad compiler and code that Altivec couldn't handle very well the G4 still would have pulled ahead by a little bit if the second processor had been on (you kind of have to expect that though, a P3 running at the same clock speed is NOT competition for a G4 in most cases, a P4 running at 1.5 times the clockspeed might be).
 
*sigh*

Somebody already posted this info before the article was published on the Web. Now, benchmarks in general are fairly meaningless, but there are a number of reason why they are particularly meaningless in this case.

From the earlier thread, I'm given to understand that this "c't magazine" is well-respected for technical expertise in Germany. If that is the case, then in this instance they are both failing to demonstrate that expertise, and abusing their readers' faith in them.

To be charitable, they may not realize they're doing this. These people are obviously no experts in the area of performance analysis. It just isn't as easy as getting a SPEC CD, booting single-user and running. The fact that they used a similar version number of GCC in both cases means little -- the benchmark very likely says more about the quality of GCC's code generation on Intel versus PPC than it does about the performance of the processors themselves.

The author is absolutely correct in his observation that Motorola and Apple would both be well served if Motorola would release a reference compiler for the G4 at the same level of quality as the Intel Reference Compiler. If I recall correctly, as a matter of fact, GCC on Intel has been improved specifically by comparing its performance to that of the reference compiler.

Comparing performance fairly is very hard work. Comparing across architectures is nearly impossible. SPEC is a very specialized professional suite which attempts to provide meaningful results, but it was simply not designed to provide those results in a basic install-and-run scenario. Believe me, anyone who can do a good job of producing SPEC results can get job that pays a lot more than working for a technical magazine.
 
Originally posted by buffsldr


You say that "Most academic/research programs are still written in fortran". Do you have a reference for this?
Click to expand...

a cursory glance around here says that all our seismic analysis codes are done in fortran, for a variety of reasons - the most obvious being thats there already a large mass of 'peer reviewed' code already in existance.
heres a reference from http://www.faqs.org/faqs/fortran-faq/
"There is a vast body of existing FORTRAN code (much of which
is publically available and of high quality). Numerical codes
are particularly difficult to "vet", scientific establishments
usually do not have large otherwise idle programming staffs, etc.
so massive recoding into any new language is typically resisted
quite strongly.

c) Fortran tends to meet some of the needs of scientists better.
Most notably, it has built in support for: - variable dimension
array arguments in subroutines - a compiler-supported infix
exponentiation operator which is generic with respect to both
precision and type, *and* which is generally handled very
efficiently or the commonly occuring special case
floating-point**small-integer - complex arithmetic -
generic-precision intrinsic functions"



i also stand corrected - the dual processor option was turned off in this test.


as for fp performance not being important - ever played a 3d game? crunched real world numbers or done any 3-d modelling/video editing? no?
 
Originally posted by Pants

as for fp performance not being important - ever played a 3d game? crunched real world numbers or done any 3-d modelling/video editing? no?
Click to expand...

Sure I have. I've also browsed the internet, edited images, written code and compiled programs. Not much fp utilization there.

I'm just saying some tasks require it. Others don't. In my case, double fp's just don't come into play often. I play a few games every once in a while, but my processor gives me what I require for that. I just don't do scientific calculations.

Some people certainly do need double fp performance. I don't and reckin' that most average people don't either.

Also, Fortran was at one time ubiquitous in the world of scientific programming. And while there is still work done in Fortran and a bunch of existing code, from what I've heard it isn't as popular today. I went to an engineering school for CS and my department would teach Fortran to engineering students.

From the engineers that I have heard from in the "real world", there isn't a lot of Fortran being written. In fact, some of those engineers (chemical and mechanical) where programming and said that they weren't using Fortran. More anecdotal evidence than anything, but its the truth.

Also, at MTU, many of the engineering departments would "hire out" to the CS department to aid in programming work for their projects. In those cases, Fortran was often passed on. I worked for 2.5 years on a Chemical Engineering. Most of the complex estimation code was written in C. This was not an uncommon situation at the school.

Matthew
 
Saying that most academic and research software is written in FORTRAN is like saying that most business software is written in COBOL. It's not entirely false, but it does misrepresent the state of affairs.
 
Absoft...

I'm sure Absoft is a fine company and all, but it's probably a little ridiculous to expect their compiler to perform as well on the PowerPC as Intel's does on their own hardware.

Intel is known for their compiler expertise and they've got a slight advantage in having also made their processor. (ie, Bob, who writes Fortran compilers at Intel can collaborate with Jim, who's designing P4 revisions.)

Of course, if you're writing stuff in Fortran that doesn't help you much...

Doing compiler optimizations to take advantage of Altivec is an extremely difficult process. Intel has been hard at work trying to get their compilers to do this with the P4's SIMD instructions with limited success. For the most part, I doubt that Absoft's "altivec optimization" is anything more than attempting to use the Altivec units as a generic execution unit that gets used in parallel with the integer and fp units.
 
Geek speak.....

blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah blah

what's really important is real life comparsions. is the G4 faster in Photoshop, at encoding MPEG-2, rendering in Lightwave and so on...

In many (but of course not all) cases the answer is yes...
 
Re: Geek speak.....

Originally posted by crassusad44


what's really important is real life comparsions. is the G4 faster in Photoshop, at encoding MPEG-2, rendering in Lightwave and so on...

In many (but of course not all) cases the answer is yes...
Click to expand...

What about Cubase VST or ProTools. FPU intensive? I think I have read that they are, and a cheaper Dual Athlon system will outrun a Dual g4. I would think some of the DSP of altivec would help out for multitrack audio workstations, but perhaps it doesn't.
 
You beat one dead horse,I may as well beat another.

So these were just the CPU's in custom Mobos of EXACTLY the same speed and spec?

No?

Then it's not a comparisson of Chips.

Next case.
 
Isn't most Fractal Math FPU intensive? There is probably some low level Mandelbrot Set generataor written from the same code base, compiled for both platforms to assembly level available on the web, if not several. Give the two systems the same equation and see which gets further.

Or run Bryce on both platforms and run a render.

Altivec gives a considerable boost for many DSP tasks I'm surprised that floating point stuff takes a hit. Although it is probably completely due to the way the ops are fed into the CPU pipelines, which might be considered biased, even tailored to be so, by the right analysis.

Personally, I'm glad altivec is available, but its a shame it is in the hands of a badly managed company that doesn't know how to produce or market its own chips. Bet either IBM or AMD could get higher yields to market faster.
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back