The G5 - Is it really Slow????

[awulf]

One of my friends (anti-mac) has pointed out to a site displaying the 'real' benchmarks, and showing that the G5 is slower than the P4 in everyway.

Well here is the link:
http://www.overclockers.com/tips00408/

What if this is true?
What do you think?

[pEZ]

I'd still rather use a slower Mac than a faster PC anyway.

If those benchmarks are true, Apple has a bit of crap to wade through - because they're never going to hear the end of it. But, as most of the world consists of PC zealots, I tend to believe that those machines were tweaked beyond what is readily available to the consumer market. Also - their argument is slightly flawed. If Dell overclocked their machines for some tests, and Apple was merely demonstrating the power of the G5 sans overclocking, then those benchmarks really ARE fake. The average joe doesn't know how to overclock any machine.

If those benchmarks are false, however, it would only be typical of the petulent PC world that their beloved machines aren't the best anymore. Hooray for IBM and Apple.

[Billicus]

It's balogna...

[XnavxeMiyyep]

It's probably completely made up, or tweaked in some way that hurts the Mac and helps the PC. Regardless, I'd rather have my Dual 867 than the fastest Intels, AMDs, or HPs.

[janey]

there is no way to fairly compare a PC with a Mac.
In this matter, Apple tweaked the computers so the G5 will look as though it is faster than the PC's. On Overclockers.com, they tweaked it in favor of the PC's. And of course, there is no way to benchmark the hardware without having the software in it somewhere. Mac OS X and Linux and Windows are all different so of course if you run a benchmarking app on Linux you'll get different scores than you would in Windows.

Oh and don't trust anything on overclockers.com, that site is full of BS.

Don't trust me either, I'm biased :p but I seriously think that my Power Mac G4 is faster than my dual P4 Xeon :p when running photoshop and a few other apps i use often

[Cubeboy]

Overclockers pointed out that the Pentium 4 could score much better in SPEC if it used a Intel compiler, however, they failed to note that the G5 could also scale better had it used an AIX compiler as was clearly shown by the IBM's preliminary PPC970 SPEC scores. That's not to say Apple is completely without fault here, they probably did bench the G5 originally using the latest AIX compiler but the results were likely not as impressive as originally hoped. So in order to make the G5 look better, they compared a G5 running a specially optimized version of GCC 3.3, to a P4 running a standard version of GCC (which itself is poorly optimized for the P4).

So to answer your question, is the G5 fast? It certainly looks like it, is it the fastest desktop on the planet? It could well be but we'll probably have to wait for some benchmarks that aren't submitted by Apple to find out.

[tazo]

I agree with the article on many levels. on the guru and a flock comment. if apple came out with the equivalent of a 400 mhz celeron in their next big computer (like G5, etc), I believe a lot of mac users would buy it because of its pretty case. As a whole I think its fair to say that not everyone is like this, butr look what we pay for. true we have more reliability, some porgrams run faster, better stuff out of the box, good looks, but often times its a slower experience.

I am not trying to be a troll. I am not attempting to be represenative of all mac users, I just think its fair to accept that Macs are not going to be the fastest computers, right now.

I agree on both sides of the issue. I agree with mac users that macs are superior to pc's for the reliability and so forth that they offer. However I agree with the article in that it is true that Apple does hide intense details about their computers.

I guess I am just a moderate in this issue, a moderate on political topics, and a moderate person in reality.

[the future]

Originally posted by tazo
... I just think its fair to accept that Macs are not going to be the fastest computers, right now...

Why exactly is it "fair" to believe in the bollocks benchmarks of an obscure PC favoring website more than in the Apple benchmarks? And by the way, SCREW all those ********** benchmarks, they mean nothing at all as they can obviously be bent in any direction one likes to bend them. If the REAL WORLD PERFORMANCE of the G5 is anywhere near the demos at the WWDC keynote, the P4 is just yesterdays tech, period.

[Spock]

Originally posted by Billicus
It's balogna...

What is "Bologna"? Anyway people make these stupid benchmarks becouse they are insecure that their top of the line PC maybe slower than a Macintosh.

[Daveman Deluxe]

I read the article and I don't really trust either Apple's own benchmarks OR overclockers.com's benchmarks. Both obviously have too much invested in their side winning.

Benchmarks are almost completely meaningless anyway. http://arstechnica.com/cpu/2q99/benchmarking-1.html

[tazo]

Originally posted by the future
Why exactly is it "fair" to believe in the bollocks benchmarks of an obscure PC favoring website more than in the Apple benchmarks? And by the way, SCREW all those ********** benchmarks, they mean nothing at all as they can obviously be bent in any direction one likes to bend them. If the REAL WORLD PERFORMANCE of the G5 is anywhere near the demos at the WWDC keynote, the P4 is just yesterdays tech, period.

I can see where my debating of this article will lead to a warning, hurt feelings, loss of appetite, so I am not going to continue. You all read my comments.

[Rower_CPU]

Why do all these people attacking Apple ignore the source of the benchmarks: Veritest?

It's Veritest's ass on the line if the scores were manipulated. It's easier for these PC weenies, running scared because of the G5, to pick on Apple than to fairly evaluate the scores.

As always, it comes down to real world benchmarks...when the G5s come out in Aug/Sep.

[nuckinfutz]

The more likely major culprit, though, is the compiler used to get the spec source code ready for processing and what was done with it. The x86 manufacturers (outside of the Opteron tests, which often used a beta version of Suse Linux 64-bit) used the general standards: either the Intel and MS C++ and Fortran compilers, while Apple used a compiler you've never heard of (at least in the x86 world).

This idiot has never heard of GCC??? :eek:
Explain to your friend that Apple had to standardize on the same compiler that would run all all platforms. GCC ,while being OpenSouce and perhaps not as efficient as Intels or IBMS compilers, runs on both platforms hence it's selection.

This guy is making the same mistake that the Haxial guy is. He's comparing Apples results and then crying foul because it doesn't match the numbers on Intel/AMD sites.


Finally, in the Apple testing, the code for the Apple machines were tweaked to the max with all kinds of libraries installed, while the code for x86 machines were largely default out-of-the-box experience.

Ummmm proof. There's a pdf from the Veritest that completely shows the configuration of the tested computers. FUD personified here.

[Billicus]

Originally posted by Spock
What is "Bologna"? Anyway people make these stupid benchmarks becouse they are insecure that their top of the line PC maybe slower than a Macintosh.

Being biased, I believe it's a bunch of Balogna because they're just trying to swing the numbers in their favor, obviously. One could also argue that Apple swings the numbers in their favor, so it's a toss up.

[nuckinfutz]

I am not trying to be a troll. I am not attempting to be represenative of all mac users, I just think its fair to accept that Macs are not going to be the fastest computers, right now.

Yes and that's your right. But some of us feel equally the opposite. We know have the pdf with the schematics of the G5 Mobo and I honestly can says it blows Intels Canterwood platform away.

I tend to think that Mac Site administrators have been taking a "wait and see" approach and it's the PC side that's generating all the heat.

[Cubeboy]

Originally posted by nuckinfutz
Ummmm proof. There's a pdf from the Veritest that completely shows the configuration of the tested computers. FUD personified here.

Actually the pdf clearly stated that the G5 system used a optimized high speed single threaded malloc library as well as some register tweaks (to reduce snoop times although this is somewhat acceptable since they used dual cpus). Regarding GCC, the GCC compiler used on the G5 (GCC 3.3 Build 1379) is specifically optimized for the G5 processing architecture whereas the GCC compiler used by the P4 (GCC 3.3) is the stock version. GCC as a whole, is quite poorly optimized for the P4.

[nuckinfutz]

Good info Cubeboy. That's the whole conundrum we have. Each compiler benefits a particular processor. To focus on the highest SPEC numbers solely would be to use the compiler on each platform that produces the highes SPEC numbers but not necessarily the highest performing software.

The best thing to do would have been to let Apple claim what they wanted and waited for Benchmarks. Yet the furor shown by some PC sites really borders on the rediculous. As a user of both platforms I can say that there will never be an absolute advantage for either Platform. PC users thought they had that but as we see ...the chasm between Hardware platforms can be erased very quickly.

It has always been my contention that designing good Software is a more difficult task than Hardware.

The G5 Motherboard is SWEET. Shiza..I thought Canterwood was decent but Apple really kicked azz on designing this Mobo. If it's reliable as it should be we have a nice platform for the next couple of years.

[reflex]

Originally posted by Spock
What is "Bologna"? Anyway people make these stupid benchmarks becouse they are insecure that their top of the line PC maybe slower than a Macintosh.

Or vice versa ;)

[Cubeboy]

Originally posted by nuckinfutz
Good info Cubeboy. That's the whole conundrum we have. Each compiler benefits a particular processor. To focus on the highest SPEC numbers solely would be to use the compiler on each platform that produces the highes SPEC numbers but not necessarily the highest performing software.

The best thing to do would have been to let Apple claim what they wanted and waited for Benchmarks. Yet the furor shown by some PC sites really borders on the rediculous. As a user of both platforms I can say that there will never be an absolute advantage for either Platform. PC users thought they had that but as we see ...the chasm between Hardware platforms can be erased very quickly.

It has always been my contention that designing good Software is a more difficult task than Hardware.

The G5 Motherboard is SWEET. Shiza..I thought Canterwood was decent but Apple really kicked azz on designing this Mobo. If it's reliable as it should be we have a nice platform for the next couple of years.

That sums it up quite well, the PC community has blown this thing COMPLETELY out of proportions, any company will introduce a few benchmarks that will make their product look favorable, it just so happens that the entire pc community only gets all riled up about this when Apple does it. Personally I think the G5 is a great cpu, and the Powermac G5 could well be the fastest personal computer in the world, but I'll be basing my conclusions off of personal experience and third party reviews rather than benchmarketing. And your right, the G5 motherboard is definitely one kick a** motherboard.

[tazo]

Somebody needed to say that;)

[zim]

overclockers.com... humm where have i heard that name before, oh ya it is this geek that lives here in massachusetts and i would not personally trust a word he says about macs, he is full of it. his only passion in life is to tweak out his pc with his water cooling system so that he can gain higher frame rates for his video games. seems to me that if a computer needs to be tweaked, then the computer is not all that it is suppose to be.

[Sun Baked]

It's marketing, you talk up the best features to sell to the brainless sheep.

If you fall for it based on marketing alone, you deserve to get fleeced.

If everyone trusted the commercials, we'd all think the Hyundai is the best car in the world -- or for that matter that Dell is king of the world, and the Gateway needs steak sauce.

[MorganX]

Originally posted by nuckinfutz
GCC ,while being OpenSouce and perhaps not as efficient as Intels or IBMS compilers, runs on both platforms hence it's selection. sted computers. FUD personified here.

hence the worthlessness of apple's benchmark. you're going to buy software that runs the best on your machine. if a benchmark is going to give you a realistic picture of performance, then both machines should use optimized software. The best fastest compiler for mac against same for intel.

the machines don't run the same OS, so why run the same compiler? Everyone knows Windows is faster than Linux, and OS X (in my personal experience.)

the issue will be over when G5s ship and people start using them.

[MorganX]

Originally posted by tazo
I agree on both sides of the issue. I agree with mac users that macs are superior to pc's for the reliability and so forth that they offer. However I agree with the article in that it is true that Apple does hide intense details about their computers.

I guess I am just a moderate in this issue, a moderate on political topics, and a moderate person in reality.

We're a secret society. Normal people who use Macs.

[iJon]

ive been quite intrigued by all these pc users. no more does cute cases, colors, computers named after a fruit, built in screens, and all of that matter anymore. strictly speed. it is normal for these people to bash macs, but whenever its about speeds its just, its slower. i have never seen so many pc peopel go out of their way and time to prove something wrong. i think most pc users are in denial that we pulled the rabbit out of the hat like this. but what i have noticed are the more educated pc users are giving apple credit, they had reasons for not liking apple and now they are taking it back. i hear many comments like i would consider this machine if it played my games, or whatever. then you have the typical pc users who arent so smart. for example, i was over at the americas army forum getting the latest info for aa for mac (comes out this month be prepared) and some guys already set up the mac bashing thread. so i went and checked it out. i finally found a post with reasons for bashing macs. they said they suck because the emacs and imacs are stuck with one monitor and we have only 1 button mine. damn it was funny, and i let them know. we may not shock them now (i think we already did) but i know that ibm roadmap is hella scary for the pc users unless amd and intel have got something up their sleves too.

iJon

[Stelliform]

Regardless of benchmarks, my TiBook is the fastest computer I have ever used. Why? Because it works 99% of the time, it turns on instantly, and the OS is stable. I beat PC laptops to the punch almost every time since my Mac comes on instantly and talks to a DHCP server faster. And I find my Mac browses PC networked file shares faster than a PC.

So regardless of benchmarks, for me the Macintosh is a faster computer because on average I can do my work faster.

[CaVoLo]

Originally posted by Stelliform
Regardless of benchmarks, my TiBook is the fastest computer I have ever used. Why? Because it works 99% of the time, it turns on instantly, and the OS is stable. I beat PC laptops to the punch almost every time since my Mac comes on instantly and talks to a DHCP server faster. And I find my Mac browses PC networked file shares faster than a PC.

So regardless of benchmarks, for me the Macintosh is a faster computer because on average I can do my work faster.

exactly i belive this has been said before....but we shouldnt base the mac vs pc debate on raw speed alone. I would base on productivty and how fast/well your work is done on which ever computer you choose to do it on. THis mumble jumble about G5's not being fast seems like an excuse for these guys to keep wasting there money on buying pc's. For me a Mac is WAY WAY WAY better than any pc!!

[Durandal7]

This is utter nonesense. For a site that deals with computers he should be ashamed of himself for his apparent lack of any knowledge about how software optimization works.

He wants to play the "the Dells weren't running special MS optimized code" game, does he?

If he had been running the G5 (which he does not have access to I might add) with G5 optimized code he would have found the G5 to put out far more processing power.

NASA ran independent benchmarks of the G5s running with only one processor and found that in all ACTUAL REAL WORLD applications involving mathematical computations that the G5 outperformed the P4 in every way. In fact with both processors activated the G5 outperformed the P4 by 95.3% yielding 498 MFlops compared to the paltry 253 MFlops of the P4.



The Bottom Line Is:
Do not trust benchmarks from someone who does not have what they are benchmarking.

[simX]

Originally posted by Billicus
Being biased, I believe it's a bunch of Balogna because they're just trying to swing the numbers in their favor, obviously. One could also argue that Apple swings the numbers in their favor, so it's a toss up.

Grrrrrrr.

It's "Bologna", which comes from the city in Italy for which it's named after. How it got named after that city, I have no idea.

[Anticipat3]

iJon nailed it dead on on this one, I think.

Slashdot has some threads like this:
http://apple.slashdot.org/apple/03/06/24/1232237.shtml?tid=126&tid=181

which I've found have the best arguments from both "sides." When the dust cleared, it seems that the general consensus is that the benchmarks were as fair as possible, and that the G5 really is faster.

Here's the qick responses to this idiot's article:

1. SPEC/Compilers. Apple's choice was either to use GCC (which all Linux and OSX users use daily) which runs on both platforms, or use ICC for the intel box, and an optimized Power4 compiler for the G5. It's a lot more fair to use the same compiler on both machines, so that's what Apple did. Had they used optimized compilers for both machines, they'd likely be under more fire yet, as this would be much LESS fair. When people say that "apple made the intel numbers lower than what dell gave them as," that's because Dell was using ICC and not GCC, not because Apple fudged numbers.

2. "They Used Red Hat... not Windows." Guess what kids, Apple VP Greg Joswiack answered a few questions about this, and it turns out that they used Linux because it made the PC look BETTER -- other sites show the same thing; the PC gets more spec points under linux than under windows.

3. It's also a big consensus that the real world app performance couldn't have been fudged, and that there's no doubt that the G5 knocked the P4 on its ass in real world performance tests.

4. Another number people have been debating is the quake 3 figures apple posted on their site -- showing a P4 getting 275 fps and a G5 getting 350 fps at 640x480. The gamers screamed -- P4s are capable of 400fps with such setups, they yelled, and pointed to benchmarks from Toms Hardware, etc. However, at the end of the day, it was pointed out that Apple was running default config (sound ON)for 1.32 and runing Demo4, while Tom's was using 1.17 and Demo1 and had some tweaks (sound OFF). Benchmarks were run again all over the web, and sure enough -- on 1.32 Quake 3, with sound on, and running Demo4, the P4/R9800 gets between 270 and 280 fps.

Anyways... believing all of Apple's propaganda would lead you to believe even your iBook is an "awesome game machine," but in the case of the G5, I think the box really is as fast as they're claiming it is.

[patrick0brien]

-awulf

You see, if the soured PC zealots (yes, the do exist) feel that if they shout, bitch, and scream loud enough, they can make the G5 slower in their minds, and thereby drowing us out.

But we know better don't we? :D

[hacurio1]

Originally posted by Cubeboy
Actually the pdf clearly stated that the G5 system used a optimized high speed single threaded malloc library as well as some register tweaks (to reduce snoop times although this is somewhat acceptable since they used dual cpus). Regarding GCC, the GCC compiler used on the G5 (GCC 3.3 Build 1379) is specifically optimized for the G5 processing architecture whereas the GCC compiler used by the P4 (GCC 3.3) is the stock version. GCC as a whole, is quite poorly optimized for the P4.

Ok Cubeboy, could you please explain what exactly does the malloc library do. As far as I know, it’s only for memory management purposes (no optimization there). Also, you forgot to mention that the Xeon had 2GB of ram and the G5 had 1.5GB. Could you please provide some links form ***reliable*** sources backing up your statement that GCC 3.3 is optimized for the ***PowerPC*** G5 and its poorly optimized for the ***x86** PIV. Thanks.

[ddtlm]

hacurio1:

Well I'm not gona dig up any links, but anyway there are certainly tests out there showing that GCC 3.x on a P4 is a lot slower than ICC 6/7 on a P4, whereas on an Athlon the compilers are "pretty close" with ICC still leading. The conclusion that GCC doesn't like P4's is pretty reasonable, I've even seen quotes from GCC programmers talking about it being at least partly due to a lack of information about how to do proper instruction scheduling.

As far as GCC 3.x being well optimized for a PPC970, well I don't believe that. Probably better optimized than for a P4, but it still lacks the sort of trickery ICC 6/7 can pull off with autovectorization.

[Cubeboy]

Originally posted by hacurio1
Ok Cubeboy, could you please explain what exactly does the malloc library do. As far as I know, it’s only for memory management purposes (no optimization there). Also, you forgot to mention that the Xeon had 2GB of ram and the G5 had 1.5GB. Could you please provide some links form ***reliable*** sources backing up your statement that GCC 3.3 is optimized for the ***PowerPC*** G5 and its poorly optimized for the ***x86** PIV. Thanks.

All the tests in the SPEC CPU2000 benchmark suite have a memory footprint of between 100 and 200 MB so having lots of memory isn't going to affect the scores.

Malloc Libraries helps allocators manage large numbers of objects ("heaps") that would otherwise cause fragmentation which would degrade the allocator's performance as the heap size grows. They can also reduce swapping and maintain better locality, as well as other things which I will not delve into now. How much a malloc library can affect performance would depend on how much of the total execution time your application spends on allocating memory. Considering your typical application (and I'd expect SPEC to be similar) spends 40% of it's total execution time allocating memory, performance can be significantly improved.

Regarding compiler comparisons, look up the FLOPS comparison done by Aces hardware or go to any of the links below:

http://www.willus.com/ccomp_benchmark.shtml?p2

http://www.coyotegulch.com/reviews/almabench.html

http://www.polyhedron.com/compare/win32/f77bench_p4.html

http://www.polyhedron.com/compare/win32/f90bench_p4.html

http://www.polyhedron.com/compare/linux/f77bench_p4.html

http://www.coyotegulch.com/reviews/intel_comp/intel_gcc_bench2.html (Note the Scimark 2.0 scores Tychay)

http://www.intel.com/software/products/compilers/techtopics/compiler_gnu_perf.pdf

The time that Intel moved to ICC was about the same time that they introduced the Pentium 4, GCC has since had very few optimizations for the Pentium 4 from what I've read, it doesn't even have a scheduler.

GCC 3.3 Build 1379 used for the G5 was Apple's own version and has never been seen before until the benchmark and the introduction of the G5, GCC 3.3 used by the Dell was downloaded off the internet. For all we know, Apple could well have spent the better part of the year optimizing the compiler for the G5.

Also it seems that my original suspicions about hyperthreading improving SpecRate have been proved to be correct. Hyperthreading does indeed, increase the performance of SpecRate on a single processor pentium 4.

Note these two official SPEC submissions, the Dell submission had hyperthreading turned off, the Intel submission had hyperthreading on.

Dell SPECint Rate Base: 12.6
Dell SPECfp Rate Base: 12.7

Intel SPECint Rate Base: 13.8
Intel SPECfp Rate Base: 13.6

Overall a Pentium 4 with hyperthreading turned on improved SPECint Rate by nearly 10% and SPECfp Rate by 7%. Looks like somebody <cough> Joswiak </cough> was lying.

[nuckinfutz]

You mean "Joswiak" Wosniak has nothing to do with this.

;)

[Cubeboy]

Originally posted by nuckinfutz
You mean "Joswiak" Wosniak has nothing to do with this.

;)

Whoops, better edit that, my sincerest apologies to Mr Wosniak. :)

[FlamDrag]

What I really wish Apple would have done is lots and lots of benchmarks in addition to the ones that they did.

The problem with these numbers (irrelevant as they may be) is that were dealing with a small data set. It would be better if they had run these tests + Intel optimized compilers, + Apple opt. comps. + run Unix, Linux, Windows (many flavors), OS/2, virtual PC, turned the G5 upside down, whatever! Just do a whole bunch of things and show us all the data and let us make up our own minds.

In my opinion, it helps Apple's case if they lose a couple of tests.

At the end of the day - I'm buying a Mac anyway and I'm just happy to have faster machines available. Some WIN users will never ever buy a Mac - no matter how fast.

[Cubeboy]

Originally posted by Anticipat3
Another number people have been debating is the quake 3 figures apple posted on their site -- showing a P4 getting 275 fps and a G5 getting 350 fps at 640x480. The gamers screamed -- P4s are capable of 400fps with such setups, they yelled, and pointed to benchmarks from Toms Hardware, etc. However, at the end of the day, it was pointed out that Apple was running default config (sound ON)for 1.32 and runing Demo4, while Tom's was using 1.17 and Demo1 and had some tweaks (sound OFF). Benchmarks were run again all over the web, and sure enough -- on 1.32 Quake 3, with sound on, and running Demo4, the P4/R9800 gets between 270 and 280 fps.


Barefeats conducted their own Quake 3 benchmark out of suspicion on this, with the same configurations as was used by Apple, a Pentium 4 with a Radeon 9800, scored between 36 FPS to 43 FPS faster than Apple's Pentium 4.

[Edot]

Apple made no claims that the G5 was the fastest processor in the world, but that the PowerMac G5 was the fastest computer. The huge pipes make this computer faster in real world applications because these applications need to manipulate memory. I think people are twisting "World's Fastest Personal Computer" into "World's Fastest Processor", which Apple has never claimed. I think the real world tests Apple showcased proves that the G5 is the fastest computer.

The processor is not the computer!!:D

Not that the G5 isn't fast!

[Rower_CPU]

Originally posted by Cubeboy
Overall a Pentium 4 with hyperthreading turned on improved SPECint Rate by nearly 10% and SPECfp Rate by 7%. Looks like somebody <cough> Joswiak </cough> was lying.

Which compiler?

[Cubeboy]

Originally posted by Rower_CPU
Which compiler?

The Testing Configuration for the Dell can be found here:
http://www.spec.org/cpu2000/results/res2002q4/cpu2000-20021113-01829.asc

The Testing Configuration for the Intel pc can be found here:
http://www.spec.org/cpu2000/results/res2002q4/cpu2000-20021202-01888.asc

Both systems used ICC and IFC 6.0.

[the future]

Originally posted by Edot
Apple made no claims that the G5 was the fastest processor in the world, but that the PowerMac G5 was the fastest computer. The huge pipes make this computer faster in real world applications because these applications need to manipulate memory. I think people are twisting "World's Fastest Personal Computer" into "World's Fastest Processor", which Apple has never claimed. I think the real world tests Apple showcased proves that the G5 is the fastest computer.

Smart thinking! You're spot on with this observation which also neatly explains the much bigger advantages of the G5 SYSTEM in the real world app demos than those of the G5 PROC in the benchmarks.

[mikeyredk]

try this

http://www.osnews.com/story.php?news_id=3997&page=1

[Rower_CPU]

Originally posted by Cubeboy
The Testing Configuration for the Dell can be found here:
http://www.spec.org/cpu2000/results/res2002q4/cpu2000-20021113-01829.asc

The Testing Configuration for the Intel pc can be found here:
http://www.spec.org/cpu2000/results/res2002q4/cpu2000-20021202-01888.asc

Both systems used ICC and IFC 6.0.

Ok, so maybe under GCC, HT incurs poorer performance just like Joswiak said?

[Sedulous]

It's all about real world use. When September rolls around what will it matter what someone else says when you can personally burn a P4 with your new G5.

[Cubeboy]

Originally posted by Rower_CPU
Ok, so maybe under GCC, HT incurs poorer performance just like Joswiak said?

Not so sure about this, the Dell did beat the Intel in SPEC int and fp, so it's pretty reasonable to assume that the reason the Intel beat the Dell in SPEC rate had to do with how SPEC rate works, not how the compiler works.

By the way, hows the weather in San Diego, still perfect, I assume?

[solvs]

It's always gonna be us vs. them. If the Mac is slower it's, "who cares if it's a little slower, it's still better". If the PC is slower, number are scrutinized over machines that haven't even been released yet. And "OMG, a business is using deceptive marketing to make us buy their products" (like they all don't do it). And "well, I still can't use this program or game on it", or whatever.

Who cares?

The G5 may be faster at some things, the P4 or Opteron for others. Apples are better for some things, WinTels better at others. This debate will rage on, and people are still going to buy whatever they feel suits their needs.

The G5 is fast. I, for one, am happy to have some hardware parity. Call me a zealot, but I consider myself merely a drowning fish.

Think about that one for a second. :D

[cuneglasus]

Spec is nothing but slight of hand.It has never produced results that matched real world performance.I was reading just yesterday on aceshardware I believe about how the p4 slapped the athelon silly on spec but was murdered in real world tests.This was written a few years ago.The only reason Apple used it was so they could make the claim of "Worlds fastest computer" useing an "industry standard" benchmark.Apple used its own tricks to level the playing field,which of course meant getting rid of intels compiler that even the head of the microprocessor report admitted was "tuned" for tests like spec.Check out the submitted scores on spec.org.Even AMD uses intels compiler.It is foolish to base your perceptions of performance on a handful of simple app fragments in a benchmark you have to pay for in an industry where intel has called the shots for decades.The results look so skewed its clear the spec code is biased towards x86 and intel in particular.Of course the thought will outrage some I'm sure.

[dongmin]

Originally posted by tazo
if apple came out with the equivalent of a 400 mhz celeron in their next big computer (like G5, etc), I believe a lot of mac users would buy it because of its pretty case.

ummm, NO.

No professional, who makes their living off of their computer, is gonna plunk down $3000 for a 400 mhz celeron system, not even for a super-duper-Ives designed case. There is a simple reason that PowerMacs have not sold well until now: they were slow. And there is a reason now why they're flying off the shelves (figuratively speaking): they are fast.

please, even if you're gonna exaggerate, don't cite such silly hypotheticals.

[solvs]

I'd have to agree with dongmin here.

The original 500 MHz G4 was pretty cool when it finally came out, but you remember the nightmare that was Motorola (even then IBM had to bail them out).

Then OS X came out, and when it matured into .1 and .2, Apple was even cooler. But their processors were slow. And I don't mean slow, as in slower than Intel or AMD. I mean slower than they should have been. Remember how unimpressed we were with the 1.42 G4s even?

Now, finally, we will have the G5 (fast) and 10.3 (even faster). Even if a $4,000 Dual Xeon or Opteron slightly beats it in benchmarks, it's still fast. And still faster than the G4s and 10.0-.2 (maybe even OS 9).

I still say yippee. :p

[MarkCollette]

I posted an explaination here of all these benchmarking discrepancies:

http://slashdot.org/comments.pl?sid=69975&cid=6373223

Furthermore, to address this specific instance, one should note that these overclockers never did any benchmarking themselves, they merely copy and pasted official scores as given from the manufacturers. The only thing he said that remotely gives credence to the comparisons are that the amount of RAM used was the same ( 2GB ). We don't know which types of hard-drives are used, what OSes and versions were used, etc. So, right off, those numbers are irrelevant.

Secondly, this whole issue of using the ICC compiler for the P4s. What Intel has done, is spent a lot of time creating a compiler that is very good at compiling simple programs, like benchmarks. At my work we use Visual C++ 6 on WinNT, Solaris CC on Solaris, and whatever True64 on Alpha uses for a compiler. Each one of these has defficiencies, forcing us to use a common subset of functionality. We do not even consider using ICC since it could never handle the C/C++ language features that we use. Thus ICC is a nice toy that is very good at pumping out benchmarks, but is almost never used in large scale software projects, and thus cannot compare to GCC, which is used by many large software projects. That is why it was valid for Apple to use GCC for the P4. But to be more legitimate, they should have instead used the newest MS VC++ compiler, as that is the #1 compiler for x86, which would have forced them to do this on MS Windows, not RedHat. Or they could have done both.

None-the-less, the question of speed, when both chips are now in the same class, is merely a masterbatorial affair that only affects 2% of the market. Expect PC users to cling to their speed crown, as that is the only thing they can cling to, against the ever growing superiority of the computing experience on the Mac.

[Cubeboy]

Originally posted by MarkCollette
I posted an explaination here of all these benchmarking discrepancies:

http://slashdot.org/comments.pl?sid=69975&cid=6373223

Furthermore, to address this specific instance, one should note that these overclockers never did any benchmarking themselves, they merely copy and pasted official scores as given from the manufacturers. The only thing he said that remotely gives credence to the comparisons are that the amount of RAM used was the same ( 2GB ). We don't know which types of hard-drives are used, what OSes and versions were used, etc. So, right off, those numbers are irrelevant.

Secondly, this whole issue of using the ICC compiler for the P4s. What Intel has done, is spent a lot of time creating a compiler that is very good at compiling simple programs, like benchmarks. At my work we use Visual C++ 6 on WinNT, Solaris CC on Solaris, and whatever True64 on Alpha uses for a compiler. Each one of these has defficiencies, forcing us to use a common subset of functionality. We do not even consider using ICC since it could never handle the C/C++ language features that we use. Thus ICC is a nice toy that is very good at pumping out benchmarks, but is almost never used in large scale software projects, and thus cannot compare to GCC, which is used by many large software projects. That is why it was valid for Apple to use GCC for the P4. But to be more legitimate, they should have instead used the newest MS VC++ compiler, as that is the #1 compiler for x86, which would have forced them to do this on MS Windows, not RedHat. Or they could have done both.

None-the-less, the question of speed, when both chips are now in the same class, is merely a masterbatorial affair that only affects 2% of the market. Expect PC users to cling to their speed crown, as that is the only thing they can cling to, against the ever growing superiority of the computing experience on the Mac.

In my opinion, ICC 4 was a benchmark compiler, ICC 5, 6, and 7 were all pretty darn good production compilers and theirs quite a few companies that use them to make production software (Oracle, SGI, IBM, Newtek, CERN, MusicMatch, ALIAS/Wavefront, Bell Laboratories to name a few). True theirs a lot fewer applications compiled with ICC than say GCC or VC++ but thats not to say they can't effectively be used for large scale software projects.

"The Intel compilers have performed excellently on our ROOT code. On average, the Intel C++ Compiler for Linux produces executables that run 30% faster than ones produced by gcc 3.2. Due to the excellent compatibility with the GNU compilers, the porting effort was reduced to a minimum. Furthermore, the strong technical support that is provided with the Intel Compilers enables very quick turnaround on fixes for customer issues."
Dr. Fons Rademakers
Senior Scientist
CERN
Geneva

" We are impressed with the results we have seen using the Intel Version 6.0 Compiler with IBM DB2 database software. The performance, reliability and compatibility of the compiler and the quality technical support Intel provides were important factors in selecting the Intel compiler for the future development of DB2 for Windows and Linux, for both the IA-32 and the Itanium architecture."
Berni Schiefer
Distinguished Engineer and Manager of DB2 Universal Database Performance and Advanced Technology
IBM

"We recently switched to the Intel C++ Compiler from a leading compiler for use on Maya 5.0's dynamics code. The output ran so much faster - up to 90% faster - than before, that our engineer didn't actually believe it and rechecked his data. To say the least, we are impressed with the Intel product."
Kevin Tureski
General Manager of Maya Engineering
Alias|Wavefront

"MUSICMATCH, Inc.: "After optimizing our code, (with the Intel complier) MP3 encoding speeds dramatically increased, and MUSICMATCH Jukebox was able to rip MP3 files from audio CDs at a blistering 35x speed for the first time"
Jim McLaughlin
Director Hardware Alliances
MUSICMATCH, Inc."

[MarkCollette]

Yes, we've established that the ICC compiler is pretty good at compiling number crunching code, which explains the media companies and scientists using it, but how about it's support of templates, namespaces, and its compliance to the ARM and to the ISO standards? Furthermore, how many third party development tools support ICC?

The point being, the important thing about computer preformance is how well real world applications run on it, which is determined to some degree by which compiler those applications are compiled under, which for the most part is simply not ICC. Thus, how relevent can benchmarks be which are compiled under ICC?

Since Apple chose to go against RedHat with GCC 3.3, it's clear that they are claiming that their platform is superior when using opensource software. Of course PC users are complaining that their other software, on other operating systems, compiled by other compilers are faster. Apple simply should have ommitted their claims of the fastest workstation, when they had only proven a very narrow area of being the fastest in.

[Cubeboy]

Mark Collete:

The point of my post was to demonstrate that ICC isn't solely used as a benchmark compiler but is also used by many big name developers to produce code for real world applications and software like Lightwave, Maya, Jukebox, DB2, and Oracle 9i to name a few. I don't think I need to prove that ICC produces very fast code in some instances.

[rjwill246]

Wintel stalwarts are going to keep saying that Apple tweaked GCC to their benefit when it seems clear that they didn't blatantly do so... they could have used Altivec and other CPU tweaks but Veritest have been pretty forthcoming on what they did and did not do. Was it a completely level playing field? No because that goal seems impossible but given that there were no OUTRAGEOUS advantages given to the Mac, the results show that the G5 is a fantasticaly capable chip. It's in it's early days too unlike the P4 and so time is all on IBM and Apple's side here. Further, the first tirade against these benchmarks specifically included the observation that the Intel chip would have performed much better had it been using an optimized compiler. What was left out of this "discourse" was the obvious... so too, would have the G5. In any case, there are now multiple articles popping up supporting the impression that the G5 is one of the best performers out there (start with MacBytes.com as one source) and no amount of handwringing from Wintellers is going to change the fact hat Apple and IBM have one hell of a CPU and that it will get faster and by the end of the year there will be real world observations to finally prove the point. Wait til the benchmarks come out that include every tweak conceivable, permitting the G5 to show its true form. There will be howls, of not fair but no passionate Winteller could stand to think that a Mac could trump the fatuous fatigued platform they love to struggle with.

[patrick0brien]

Originally posted by Edot
"World's Fastest Personal Computer" into "World's Fastest Processor", which Apple has never claimed.

-Edot

I think that's the most eloquent, elegant, and concise statement addressing this whole argument in the press I've yet heard. Damn good point.

Thank you for keeping your eye on the ball.

[Cubeboy]

The only real world benchmark I've seen for the G5 is Jet3d (simulation code, heavy in floating point math) and on that particular test, a single 2 GHz G5 achieves rough parity with a 2.66 GHz P4, despite that neither machines were using their best compilers. Quite a difference from Apple's SPECfp test where a single 2 GHz G5 would probably outperform a estimated 3.6 GHz Pentium 4. Of course, SPECfp is alot more bandwidth intensive than Jet3d but that alone can't account for such a big difference.

[yzedf]

Trollfest 2003 - SPEC scores

OK... the scores for the G5 are provided by Apple. The low scores for the x86's are by Apple. The high scores for the x86's are provided by SPEC on their website.

The best part to me is that everyone acknowledges that the G5 is what, 2x faster than the fastest G4's? The high end Intel chips are out now, and have been getting incrementally faster very 4-6 months. Yet people would argue that the G4 was as fast as the x86's. NOT.

Whatever... this is all worthless until true comparisons can be run. SJ up on stage is just that, staged.

[Cubeboy]

Regarding Jet3d, I've come across a 2 GHz Pentium 4 submission using a IFC compiler. Not sure if it's scalar or packed although SSE2 seems to lower performance on this particular benchmark. The test configurations seem different as the scores are dramatically higher. This would give a comparison of how a Pentium 4 with a optimized compiler would fare. Now if only we can get some results for a G5 with a optimized compiler.

Pentium 4 2 GHz/RAMBUS
RH Linux 7.3 / IFC
1 CPU-SCALAR(?)
842 MFLOPS

Pentium 4 3.2 GHz (assuming only linear increase in clockspeed)
RH Linux 7.3 / IFC
1 CPU-SCALAR(?)
1347 MFLOPS

PowerMac G4 1GHz
OS X 10.1.5 / Absoft f77
1 CPU-SCALAR
236 MFLOPS

PowerMac G5 2 GHz (Extrapolating from Craig's Numbers)
OS X 10.2.7 / Absoft f77
1 CPU-SCALAR
571 MFLOPS

http://vivaldi.bio.bnl.gov/asda/bb/archive/ccp4bb/ccp4bb.200208/7049.html