No 3GHz Soon and No G5 Laptop

[AidenShaw]

Add in the Linpack performance where the G5 kills comparable x86 systems. Add in things like the gene matching program (I can't remember its name). It's pretty clear that SPEC is the outlier - probably due to the compiler issues that I've already mentioned.

Kills? The #4 cluster in last year's Top500 was only slightly slower than VATech's. It used 2900 3GHz Xeons, or about 30% more CPUs for about the same power. "30%" hardly is "kills".

Even IBM admits that LINPACK is a bit of a sweetheart for POWER systems, since it uses the MADD instruction in POWER quite a bit. In the Spanish BladeCenter article (http://www-306.ibm.com/chips/products/powerpc/newsletter/jun2004/ppc_process_at_work.html) it says:

In practice, only a small portion of peak capacity is achieved because a processor is rarely scheduled to do simultaneous “multiply and adds” in double precision. However, the LINPACK benchmark, which is often used to rank supercomputers (the Top500 Supercomuter list), makes extensive use of simultaneous multiply and add.

The BLAST gene sequencing program has been extensively optimized for AltiVec with the help of Apple. I agree completely that the G4/G5 systems are tops for running this one application. (Haven't I been saying benchmark the application that you use?)

 

[Mav451]

ROTFLMAO.

My 'I know everything about supercomputers' wants to rely on a benchmark that's probably more subject to compiler oddities than anything else and which isn't even capable of handling multiple processors.

Forget SPEC. It's a lousy benchmark-unless you happen to be using one of the SPEC applications every day.

You dismiss SPEC, but you allow the careful selection of "real world" applications that show the processor in a good light AND you still stick with Apple's theoretical FLOP #'s (which are far worse than the Mhz Myth in being misleading about performance). How many times has Apple used Photoshop to show it "crushing" Pentiums? With the G4 and the G5, it has done so. I don't expect to see anything different 10 years from now either -_-. This is what I call selective benchmarking, and it is rather deceptive and misleading if you base all of your consideration of a processor on these #'s alone.

If you dismiss one benchmark, you have to dismiss all of them. #'s, can and WILL always be manipulated to highlight strengths (while hiding weaknesses) to mislead consumers. This is why I never fully trust any benchmarks anymore--hell, I stopped trusting them since the Bapco incident where a benchmark skewed all the #'s to test Intel CPUS in only their strengths, while removing the tests from previous year's benchmark program that had ORIGINALLY shown AMD's strengths. Again, selective benchmarking is the problem here.

 

[DMann]

Marketing

MacCentral quotes Apple's Directory of Power Mac Product Marketing, Tom Boger regarding the 3GHz prediction from Steve Jobs last year. According to Boger, Apple will not meet the 3GHz promise:

Boger also states that users should not expect G5 PowerBooks this year.

Marketing, Marketing, Marketing.... just wait until WWDC

 

[DMann]

G5 powerbook

Ouch ... I was really looking forward to getting a G5 powerbook this year.

Another broken promise it seems? Not like it wasn't expected after all. Oh well, nice upgrades. Looking forward for new displays! ^_^

-OA

With a scaled down version of a 970 FX running at
1.6 Ghz, (90nm) it could happen

 

[wdlove]

Marketing, Marketing, Marketing.... just wait until WWDC

That is all that we can really do now. We are at the mercy of Steve Jobs.

 

[macdong]

Marketing, Marketing, Marketing.... just wait until WWDC

you seem rather confident
i still hope they'll release PowerBook G5 at WWDC.
i'd hate to go on a 2-month trip with an iBook...

 

[jragosta]

You dismiss SPEC, but you allow the careful selection of "real world" applications that show the processor in a good light AND you still stick with Apple's theoretical FLOP #'s (which are far worse than the Mhz Myth in being misleading about performance). How many times has Apple used Photoshop to show it "crushing" Pentiums? With the G4 and the G5, it has done so. I don't expect to see anything different 10 years from now either -_-. This is what I call selective benchmarking, and it is rather deceptive and misleading if you base all of your consideration of a processor on these #'s alone.

If you dismiss one benchmark, you have to dismiss all of them. #'s, can and WILL always be manipulated to highlight strengths (while hiding weaknesses) to mislead consumers. This is why I never fully trust any benchmarks anymore--hell, I stopped trusting them since the Bapco incident where a benchmark skewed all the #'s to test Intel CPUS in only their strengths, while removing the tests from previous year's benchmark program that had ORIGINALLY shown AMD's strengths. Again, selective benchmarking is the problem here.

You're wrong on virtually all counts.

I didn't dismiss SPEC at all. I said that it's useless for most users - because very few people run apps that behave like SPEC.

Apple's tests, OTOH, are shrink-wrapped commercial apps, most of them graphics or audio related - which is exactly the type of application that most PowerMac users are going to run.

You want benchmarks that are close to what you're doing with the computer. For the overwhelming majority of people, Apple's tests are much, much closer than SPEC is.

If you run applications in the SPEC suite or similar to those, feel free to use SPEC as your benchmark. Very few people do so.

 

[jragosta]

Marketing, Marketing, Marketing.... just wait until WWDC

I don't think so.

It's against the law to knowingly make material, false statements. Saying that there would be no PowerBook G5 this year and then announcing one just a few weeks later would definitely be false AND material.

Anyone who sold their Apple stock during the intervening time would have a very strong case with the SEC.

You can refuse to answer questions. But when you give an answer, it had better be accurate.

 

[Mav451]

You're wrong on virtually all counts.

I didn't dismiss SPEC at all. I said that it's useless for most users - because very few people run apps that behave like SPEC.

Apple's tests, OTOH, are shrink-wrapped commercial apps, most of them graphics or audio related - which is exactly the type of application that most PowerMac users are going to run.

You want benchmarks that are close to what you're doing with the computer. For the overwhelming majority of people, Apple's tests are much, much closer than SPEC is.

If you run applications in the SPEC suite or similar to those, feel free to use SPEC as your benchmark. Very few people do so.

Thanks for the flame, but I did pick up something useful.

"Shrink-wrapped commercial apps".

I think it really sums up the Apple business strategy, and I guess in that sense, it works (show only the optimized apps in a good light, benchmarks essentially in a distortion field). And where did i mention in my post that I support SPEC?

Again, thanks for the great terminology though.

 

[jragosta]

Thanks for the flame, but I did pick up something useful.

"Shrink-wrapped commercial apps".

I think it really sums up the Apple business strategy, and I guess in that sense, it works (show only the optimized apps in a good light, benchmarks essentially in a distortion field). And where did i mention in my post that I support SPEC?

Again, thanks for the great terminology though.

I don't think that there's ever been any doubt that Apple focuses on commercial applications. How is it 'a distortion field' to simply benchmark the apps that their customers are likely to use?

 

[pjkelnhofer]

I don't think that there's ever been any doubt that Apple focuses on commercial applications. How is it 'a distortion field' to simply benchmark the apps that their customers are likely to use?

I agree I think that Apple's benchmarks are based in real world scenarios. I am sure that they are selective when they choose which ones to publish, but if you are looking for completely honest and unbaised benchmarks from a manufacturer you are living in a fantasy world.

 

[macsrus]

There is no doubt that Apple always uses the same 4 or 5 commercial apps
as their benchmark... there is good reason for this.... because they r the only freaking apps that run on a mac

Lets face it Apple really only markets to other apple users.... PC users dont really switch because they cant run the apps they are used to running...
And there are not really any Macintosh equivalents....

I know I will get flamed for this post, but a walk into any software/computer store will prove me right... 1000s of PC apps.... few or less Mac apps on the shelf.

That being said I still like my MAC...

 

[jragosta]

There is no doubt that Apple always uses the same 4 or 5 commercial apps
as their benchmark... there is good reason for this.... because they r the only freaking apps that run on a mac

Lets face it Apple really only markets to other apple users.... PC users dont really switch because they cant run the apps they are used to running...
And there are not really any Macintosh equivalents....

I know I will get flamed for this post, but a walk into any software/computer store will prove me right... 1000s of PC apps.... few or less Mac apps on the shelf.

That being said I still like my MAC...

Umm, yeah.

150 different antivirus programs
325 Registry repair programs
426 utilities designed to address deficiencies in Windows
1104 programs designed to add features that are built into OSX
152 Word processors (how many do you need?)
And so on.

I've never found anything significant software deficiencies on the Mac. Granted, there aren't as many games, but if I wanted to play games, I'd get a Playstation.

Raw numbers aren't everything.

 

[macsrus]

Umm, yeah.

150 different antivirus programs
325 Registry repair programs
426 utilities designed to address deficiencies in Windows
1104 programs designed to add features that are built into OSX
152 Word processors (how many do you need?)
And so on.

I've never found anything significant software deficiencies on the Mac. Granted, there aren't as many games, but if I wanted to play games, I'd get a Playstation.

Raw numbers aren't everything.

Lets try.... Autocad... Matlab... Mapping Software.... Any Shrinkwrap Engineering Software.... Accounting Software... etc..I could go on for hours about the lack of real software for the MAC.....

This isnt really the MACs fault per se its just a fact of its small market share.

 

[Mav451]

Lets try.... Autocad... Matlab... Mapping Software.... Any Shrinkwrap Engineering Software.... Accounting Software... etc..I could go on for hours about the lack of real software for the MAC.....

This isnt really the MACs fault per se its just a fact of its small market share.

Ugh AutoCad and Matlab are two of the main pieces of software used in Engineering weeder courses (at least at Maryland) -_-.

 

[AidenShaw]

Maryland - Oh! That's close to the Mac cluster in Virginia, right?

Ugh AutoCad and Matlab are two of the main pieces of software used in Engineering weeder courses (at least at Maryland) -_-.

If they're that close to the #3 cluster (made with Macs - until it was shut down, torn apart and sold) it must be a great school!


Just kidding, but my previous sentence makes as much sense (and is as firmly grounded in logic) as a third of the arguments in this thread !

 

[Mav451]

If they're that close to the #3 cluster (made with Macs - until it was shut down, torn apart and sold) it must be a great school!


Just kidding, but my previous sentence makes as much sense (and is as firmly grounded in logic) as a third of the arguments in this thread !

Haha I've been posting in this thread for quite a while and you never once looked at my "Location"? 0_0

Anyway, yeah I'm pretty close to Vtech from my home (UMD campus is north east from my house), plus I know a few guys who go there. None of them are in engineering/comp sci, so I guess they don't really care about it--they failed to mention it even once last fall semester. Imagine that haha.

 

[AidenShaw]

maybe it's not on the radar screen?

so I guess they don't really care about it--they failed to mention it even once last fall semester. Imagine that haha.

Not surprising - outside of the Mac rumour sites that VA cluster isn't getting much traction.

It's been good for InfiniBand (I've seen a few places that decided on IB instead of Myrinet based on VATech), and IBM's using it as a reference for the PPC BladeCenter systems.

If Apple were smart, they'd "knife" OS X and push Linux for these clusters. It's a Linux community out there, and trying to force OS X on most of these big clusters will be a lost cause.

OS X will be OK for the small shops without any real staff, and for the content creation folks already using Macs who need a handful of systems for a "render farm". The multi-million dollar clusters, however, are the domain of Linux and UNIX.

 

[AidenShaw]

BTW - Next Top500 list out on Wednesday

http://www.isc2004.org/conference.p...23,+2004+(1st+day)&file=program/first_day.inc

Schedule on Wednesday, June 23, 2004 (1st day)

I. Opening Session

Chairman: Hans W. Meuer, Prometeus GmbH, Daisbach, Germany

10:00-12:00
Welcome Address
Hans-Wolfgang Arndt, University of Mannheim, Mannheim, Germany
Andreas Reuter, European Media Laboratory, Heidelberg, Germany

Highlights of the 23rd TOP500 List / Awards for the #1 System worldwide and the #1 System in Europe
Erich Strohmaier, CRD, Lawrence Berkeley National Laboratory, Berkeley, CA, USA
Jack Dongarra, University of Tennessee, Knoxville, TN, USA

 

[DrBoar]

There is no need to "knife" OS X and replace it with Linux/Unix. OS X is a UNIX and they could even do a more pure BSD version with no HFS and all that and call it "OS X enterprice" even if it is NetBSD in a X wrapper.

Regarding no 3 GHz and G5 PB anytime soon. Why would they say anything else? "sure the dual core G6 and the G5 PB is just months away but do buy G4 powerbooks to support Motorola that we all love and to boost 2004 Q2 for Apple as well"

 

[jragosta]

Not surprising - outside of the Mac rumour sites that VA cluster isn't getting much traction.

That's true - if you want to ignore all the awards that they've been given by so many computer organizations.

 

[pjkelnhofer]

That's true - if you want to ignore all the awards that they've been given by so many computer organizations.

Like this?
I think the main thing that stands out in everything the I have read about the Big Mac project is not how fast it turned out to be, but it's cost in relation to other computers on the list.

 

[AidenShaw]

true - but "Linux" is not "UNIX", and it's not "BSD"

OS X is a UNIX and they could even do a more pure BSD version with no HFS and all that and call it "OS X enterprice" even if it is NetBSD in a X wrapper.

The point is that most of the work in HPC is being done in Linux - and people don't want to have to port their Linux programs to UNIX and BSD variants. (X Window System support is not really an issue - superclusters don't use graphics cards.)

It's a path of least resistance issue - having to port to OS X is harder than porting to PPC Linux (and even harder than to Itanium Linux).

As far as all the "awards" go, let's see what's on the Top500 list next week - particularly at the high end.

"Traction" means more top supercomputers made from PowerMacs, not a bunch of beauty pageant awards. (An "Oscar" is nice, but bucks at the box office are what drives movies.)

 

[AidenShaw]

cost per production hour - infinite!

I think the main thing that stands out in everything the I have read about the Big Mac project is not how fast it turned out to be, but it's cost in relation to other computers on the list.

Unfortunately, there's no "general accepted accounting principles" for describing the cost of a supercluster.

The price commonly used for the Earth Simulator includes the cost of the entire building and everything inside. The Thor's Hammer system at Sandia is in a $9M building (http://www.hoise.com/primeur/04/articles/monthly/AE-PR-06-04-79.html).

The most quoted VATech price doesn't even include the cost of the air conditioning and power equipment!

Yes, VATech's computer had a low price tag, but when an article like the one you quote says that it cost "one-tenth less than traditional supercomputers" it's complete hogwash. The two price tags are not comparable because they don't include the same items.

The other big question is what the ROI of VATech is. It's been 11 months since they spent the $7M to $10M for System X, and it's not yet in production. That's quite a lot of money for nothing (yet)!

 

[Frobozz]

Kills? The #4 cluster in last year's Top500 was only slightly slower than VATech's. It used 2900 3GHz Xeons, or about 30% more CPUs for about the same power. "30%" hardly is "kills".

You mean 30% more CPU's that run at a 50% higher clock rate were slightly slower than the G5.

Yes, that "kills."

While it's obvious the MHz scale to performance ratio is not linear, a 3GHz G5 class IBM chip will apparently stomp on a lot (all?) of Intel / AMD fare.