X-bit labs

Started by: 222 | Date 01/27/04

Comments: 37 | Last Comment:  01/30/06

[1-20 | 21-31]

1. Good article, a bit over my head in some areas.

The good thing about PowerPC setups...You don't need to look to Apple to get one!

Have a look at AmigaOne and Micro AmigaOne (PowerPC on Mini-ITX mobo)
(Yeah, it is the old Amiga brand, but they use PowerPCs now)

Also look into Pegasos II
=> http://www.pegasosppc.com/

The above come either with G3 or G4 PPC CPUs. No G5 atm...

2. umh page 8 and 9 are same...

Sorry, guys! It should be corrected now.

3. http://www.apple.com/lae/powermac/performance/
""World’s fastest” based on SPEC® CPU 2000 benchmark results and leading professional application performance tests against 3 GHz Pentium 4-based Dell Dimension 8300 and 3.06 GHz Dual Xeon-based Dell Precision 650. SPEC® CPU 2000 benchmarks run with GCC 3.3 and independently tested, full report available from Veritest. SPEC® and the benchmark name SPEC CPU 2000® are registered trademarks of the Standard Performance Evaluation Corporation."

The reason the Apple shows lower scores for the P4 is because it used gcc3.3. The Pentium4 came out in 2000. That gave the gcc team three full years with Intel's help to optimize performance for it. Forunately its schedueling isn't that different from other processors like we've seen since the original pentium so they even had a huge head start. The bottom line is gcc does a pretty good job on the P4, P3, and Athlon.

GCC for the 970 only had a couple months to get optimizations included. And the 970 with its packages of instructions really does not fit well into gcc's schedueling system. In fact I read Apple/IBM's engineers had to give gcc incorrect info about the 970 in order to improve its performance since it is such a different architecture!

GCC was choosen because it is an equal cross platform system. It has the exact same front end to compile whatever your language is into its intermediate format and then there is a chip specific back end. Intel's compiler is awesome at spec, but not much else. In face most game companies use MS Visual Studio! Even for the xbox! Why not use Intel's compiler? Because it really only shines on spec code.

IBM has a compiler for the 970/Power4 that performs better than gcc on the 970. I don't know what the spec results are with that, but I know they're better than gcc's!

I belive Apple's SPEC claims are the most fair way one could possibly imagine. Why not run some old Quake2 source code compiled with gcc, visual studio 7.1, intel's compiler, and IBM's compiler and run it on a 2Ghz G5, 3.2Ghz Pentium4, and 2.2Ghz Athlon64 and see the results. Or something like that. Take a good free CPU intensive application not part of spec and compile it on the different platforms with different compilers and see the results. I suggested Quake2 since then you could also test the video card.

4. Why are you misleading readers stating that G5 was first 64 bit desktop processor Victor? You want me to read remain part of the arcticle after total balony at the begining?

we all knew about alpha for desktop. But it's a long time ended story. They never became mass production aren't they?

I think bunny may have been referring to the Opteron.

bunny -- at the time the G5 came out the Opteron was not a desktop machine. I would say that it was just starting to make it to the desktop when the AMD64 CPUs came out.

5. Well, good BUT
There'a a remaining question: I was already told that g5 were bad at specfp,
bu I'm wondering if specfp is really representative of scientific calculation. As you know the 2200 g5 benchmarked sky high in linpack, much higher than clusters of better processoiors from the spec point of view. So I can't agree when you say in page 18 "Accordingly, you cannot expect this processor to compete successfully with x86 processors in serious scientific calculations".
It rely on the application but many simulations rely a lot on latencies, branch prediction... I've the experience of programme running slower on bi-proc, multithread P4 2.4GHz than on four 833MHz alpha which is far behind in spec.

So, right G5 aren't good at all with sepcfp but it can't tell anything on its hability to run scientific application.

6. Hasn't this all been discussed 100 times already? That's OK on a forum, but not worth an article?

7. Are you seriously doubting the performance of the G5 in scientific areas based purely on SPEC numbers?

Have you even tried to run any real optimized scientific software on the G5?

Isn't 'real world' what it's all about?

OS X is not as stable as it should be?

This article throws up more questions than providing answers.

Was there a mention that the G5 was designed to be used in multi-processor systems?

8. Not a bad article. I would just like to point out a few things that were left out - perhaps because some of the information was not yet available when this article was being written.

1. IBM has just announced new compilers that are optimized for the G5. Some estimates show that recompiling programs with these compilers will result in significant speed increases with very little effort. The new compilers are fully compatible with Apple's xCode programming environment, which is very fast as well.

2. Every upgrade to OS X has resulted in substantial improvements in speed and responsiveness. As further optimization for the G5 occurs, this will only accelerate. This is directly the opposite of Windows updates, which are often much slower than their predecessor. Don't forget that the next update to WindowsXP is not anticipated until 2006.

3. It's my opinion that the hardware platform that Apple/IBM have developed has more scalability potential than current and proposed PC systems do. Keep in mind that symmetrical processing, including 4 and 8 processor systems are possible with the G5, but not with the Pentium4, extreme or otherwise.

4. Your comments on the cost of Virginia Tech's are invalid. Do you really think that the cost of the 3,000 s.f. facility housing the computers should be part of the installation cost? Do you really think that and the volunteer labor was worth over $200 million? Does anyone question whether these costs were included in the "legitimate" supercomputers that cost up to $350 million? Also, it was just announced that VT will be converting the PowerMac G5's to xServers that will provide MORE speed in one-third the space. Let's face it, Apple has scored a coup in terms of price/performance that has caused many to take notice. This includes the FBI, CIA, NASA and many more. Theoretically, a setup 7 times the size of VT's would perform twice as fast as the #1 supercomuter at 1/10th the price.

9. hmm, weird weird weird ;)
But you forget one thing, veritested came up with all those numbers, not Apple.

As mentioned by 'Steve'sBaby', Apple's #'s come from Veritest. So before you burst a blood vessel about *Apple's* numbers.... they're not Apple's, they're from a third party testing firm.

Get it straight.

10. What were the exact system specs used in the IBM test?

11. nice article, interesting perspective.
A couple of things though.

1) Comparing your information with that of "John Stokes" in his "Inside PPC970 part 1 & 2" (http://arstechnica.com/cpu/index.html) at Ars Technica a lot of your questions where addressed or at least a different analytical approach was taken; clearing up some of your statements.

2) Your obvious and very cavalier comments on OsX only show your lack of research in that aspect of the platform.
The transition between os9 to osX was very smooth considering the difference between these Operating systems (Os9 is to OsX what Windows is to Solaris).
Your repeated statement about major applications not being ported over was as false of a statement as Apples performance claims.
Of most major Applcations to be ported over in a timely fashion (within a year of OsX10.1) Quark was the last hold over, only appearing almost 2 1/2 years later.
(But this is the same company (Quark) which took 6 years to upgrade their product from version 4 to version 5).
Further more a slew of new applications appeared on the Mac which weren't there before notably 3d heavy weight from Discreet "Maya".
Again just gleaning over some of the material available on the web would have prevented this kind of misinformation.
You can get really really detailed and concise review of the operating system from "John Siracusa" in his article series on Mac Os X, again on ArsTechnica, (http://www.arstechnica.com/reviews/003/panther/macosx-10.3-1.html).

Lastly your attack on the performance claims made by Apple on the release of the G5 is very justified.
Apple did present itself in a very stupid and problematic way.
There is no justification for that.
Then again, Intel, AMD, Nvidia, ATI, et al; do the same thing on a regular basis, and are not lambasted for it in the way Apple has been.

Problematic in your attack of Apple's BenchMarks was the lack of commentary on the compilers used in all of the different benchmarks presented.
What exactly was the point of not comparing the fastest available processor from all of these companies?
I'm not saying the 3's where faked or anything, why not put the all of the processors on the table together?
This information is readily available.

These and a few other inconsistencies make this article less credible and also in an academic sense a failed paper.
Because of it's depth and of the other issues addressed above, there needs to be effective footnoting as well as a complete bibliography that is deeper than 9 links.

This would and could have been an excellent paper but in the end (as well as in the beginning --"...put aside the glorious PC to take a look at the Mac..." from the opening ) the article becomes a tainted piece of propaganda, made to make the distinguishing pc buyer comfortable with his/her purchase.
For some reason it feels similar to the works in a pseudo-science called eugenics by "scientists" in the early/mid part of the 20th century. ( it explained societal disparities in Race/Gender/Class in such a way as to excuse Hitler's and others bigoted practices)

12. Nice ideology exercize.
I have gone throough part of it.
It lacks synthesis.
But most of all it is driven by an upfront idea of the "glorious PC" and looks like a very good propaganda. Maybe Paul Thurott's site (http://www.winsupersite.com) will link to this article very soon.

I am very happy for you that you are happy with your PC.
But that is no reason to try to loose common sens.
You are right talking about mac "zealots". There are some. Please do not be worse than they are.

The world is made of diversity.
Not everything is black or white, nor beige.

NNext time, try to do it in 1 page. Maybe it will make sense

13. Interesting article. Mainly a read for Mac people. Really why would a PC person care, processor alone is not reason to switch, and it has no bearing on whether to buy and Intel or AMD box next.

So the conclusions don't make much sense. You talk in the body about the good branch prediction unit, the fast front side bus, and a few other design choices. The only conclusion you make is that Apple cheated on their benchmarks. Ok, but what about summing up the value of all the other things about the processor. What and where are these things valuable to the design on computers and programs. Aside from Megahertz which neither IBM or AMD wins over Intel, what else do these processors bring that make them better or worse. You did a little bit a thinking and research up until the conclusion. Which was all old news and little summing of the performance outside of two benchmarks.

So you probably didn't bother testing an actual machine but ran off info on web searches alone. I suggest you get a little more hands on if you really want to know the machine.

Yes really! This was exactly what I was expecting from an article at Xbitlabs: More hands on and at least som new benchmaking!

But I do thought that processor power/price ratio is a good reason to switch. At least for people having every app. has multi-platform code

14. The new PowerMac G5 is a masterpiece, the design is absolutely gorgeous. You can’t find a PC that looks like that. Apple is the greatest company in the world that builds computers. They care about us; they build things for us and adjust them to our needs. Try to make someone like Bill Gates to do something like that and he will say “Are you completely insane?”. I just wanted to say this!

About “The Fastest there is” part, I don’t know. It is very difficult to compare so different systems like that. It’s almost impossible to get accurate results. I used to use pc before, and I just regret that I didn’t convert earlier. Everything is much easier now and much better looking. Windows XP was a big mistake and Longhorn, when it will be available (probably 2020), will unfortunately be the same but with the difference of some few adjustments. But that is just what I think…

I’m completely satisfied with my new PowerMac G5 with dual 2Ghz processors =)

15. This was a very interesting article. The authors are clearly knowledgeable about CPUs and had some interesting insights. I was rather disappointed, though, that they didn't do more legwork on it. For example, Ars Technica already has a PPC970 review up at http://arstechnica.com/cpu/02q2/ppc970/ppc970-1.html and http://arstechnica.com/cpu/03q1/ppc970/ppc970-0.html. I would have been very interested to see what comments the present authors had on this previous work. I expect that if the present article had read the Ars Technica report they might have been able to benefit from it as well and perhaps it would have been an even better article. The same goes for the benchmark issues. There's been a very long and vociferous debate on the net about this issue and the authors clearly were not aware of the ins and outs of it or they would have addressed both sides of it. Finally, some of the comments made did not appear to be supported (I'm a professor and we learn to be very careful what we say) and they made me concerned about credibility issues. For example, they state that a current problem for the Mac OS X platform is that many popular programs have not been ported to it yet and cite this as a significant problem. I can't think of any such programs at this time. Quark XPress was a laggard by about a year and this prompted much comment at the time but by this point pretty much everything has been ported that is still actively being supported. Can the authors actually cite any programs to support this comment? This sort of thing makes me worried about the credibility of the other parts of the article that I do not have the expertise to judge. The article had a fairly partisan tone to it that makes me uncertain how much I can believe its conclusions, I'm sad to say. Nonetheless, I found it very interesting and thought-provoking. I'm hoping the authors might post a follow-up where they could address some of these questions and concerns.

16. If you multiply $3000x1100. you get $3,300,000.

What part of that doesn't add up to you?

How is it not possible for Virginia Tech to build the thing for $5Mil?

Your closing comments leave a bad feeling about your entire article.

The G5 will reach 3Gig by this Summer and most likely surpasse the P4 by years end which is now stuck and climming very slowly. And as you mentioned, none of those test uses the Altivec engine. Which as you mentioned, is the Trump card.

As far as I'm concerned, Apple might have streched the truth a Bit, but you streched it much further.

Oh my god, so the G5 will (after 3 years of updates) beat the Pentium 4??

Wow, and there i was thinking that new technology wasn't supposed to surpass old stuff.

Do Apple owners even realise AMD exist?

Put simply Apple computers are for that wonderful demographic that like to get annoyed about things. It's a penis extention and as a result is bandied about exactly like one. Apple owners on the whole are those who believe owning a mac is a status symbol and thusly entitles them to be snooty about everything - after all when the local Ford thrashes your Ferrari, you're gonna cry into your Mochacino right?

17. "But why Decode? We have been taught that RISC means a Computer with a Reduced Instruction Set: simple instructions are ready to be fed directly to the processor."

If you were taught that, you were taught wrong. The classic 5-stage RISC pipeline design has a decode stage. The point is to make instructions simple to decode, not to remove decoding altogether.

As an example of how it might be easier to design a decoder for PPC, all PPC instructions are 32 bits wide and must reside on 4-byte boundaries. This makes it easy to build parallel decoders and decode many instructions at once. It is exceedingly difficult to do this for x86 because x86 instructions are variable length and have a convoluted prefix byte system.

"PowerPC970 uses an internal instruction set, which is completely different from the external set."

Wrong. The 970 is unusual for a PPC implementation in that its decode stage does replace some instructions with sequences of other instructions (most PPCs do not do this), but it does so only for a handful of instructions. Most are simply used as is.

18. You clearly have a good understanding of the technical operations of a CPU, but your understanding of business and marketing is sorely lacking. This article would be much, much better if it were devoid of any commentary on marketing. The errors in your article are numerous, but the most glaring:

a) The rebound of Apple being attributed solely to adopting "standard" components... if simply choosing "standard" parts was the reason for Apple's rebound, then by that measure alone, they would be king of the world if they had only moved to Intel from 68K... A ludicrous proposition.

b) The failure of the Alpha chipset and the subsequent sale of DEC to Compaq... If only we lived in such a simplistic world that the failure of chipset could be the core cause of DEC's sale to Compaq... Again, a ludicrous proposition.

The irony is that you condemn marketing fallacies yet resort to the same kind of broad overgeneralizations and sweeping statements to describe the business operations of the likes of Apple, IBM, and DEC. You wouldn't dream of posting unsubstantiated technical details without some sort of information to back it up, so provide your in-depth business case study or leave your unwarranted guesses out of the article if you want to preserve your credibility.

19.
After reading the several comments to this article I can only say, GOD HELP US. Im not going to even bother going into what was right or wrong with that article. It’s fair enough to point out any technical issues associated with this article. On the other hand the author has done a fair bit of research to write that article. As for the comment that a lot of the information was already available on the internet than why are you reading this article than!

I have a G4 powerbook laptop and it runs sweet with 10.3 OS X. Also I have an AMD 2.8 GHz and what you know, it runs sweet as well. They both use different applications to get the same job done. As to how fast they perform a particular task, hell their are to many independents.

One last thing.

Computers are logical, humans illogical. How the hell can an illogical human compare a logical computer?

Moral of the story. Everyone has their own opinions.


I just had an idea (A really stupid idea that is). Why doesn't Intel (or any CPU manafacture) just buy out the other CPU manafactures. Than we could just have a monolopy and no more wining about who makes the best cpu. Sounds kinda like microsoft dont ya think?

20. In this type of article, opinions should be clearly 'marked' and separate from factual description. The two should not be intertwined in the same sentence.

Why did the author feel the need to include a dual processor comment in the same breath as Apple Marketing? Was he trying to imply something?

As has been stated the article also contains factual errors.

Some of the opinion comments just come over as a person being 'smug'

[1-20 | 21-31]