NASA recently benchmarked Apple’s dual 2GHz Power Mac G5 at its Langley Research Center in Virginia. The main purpose of the tests was to compare the G5 to the G4 for “computational fluid dynamics applications” however they also compare it to the Pentium 4.The test was well documented and concludes that “the G5 has about 22% better scalar floating point performance per clock cycle than the G4 systems tested and 32% better floating point performance per clock cycle than the P4 systems tested.”
IBM intends to use the PowerPC 970 (termed the “G5” by Apple) with its AIX operating system in workstation units soon, and Apple plans to begin shipping the Power Mac G5 with MacOS X 10.3 this fall.
Apple released today a 94-page PDF document detailing on the G5 architecture.
More about the G5 and an analysis can be found at DigitMagazine, while The Age has an article titled “Massive Mac“.
Insanely Great Mac informs us that the final release of Panther definitely won’t support beige G3 or earlier PowerMacs, nor will it support PowerBook G3s that lack built-in USB ports.
Some real validation of what Apple and IBM have been saying all along about the G5 Processor.
Interesting test but really, how well will OS X take advantage (if at all) with the new chip? I would guess to say that scalar floating point is not a main part of the OS.
Because I know a lot of people are going to read miss some of the details of the article, here are the main points not mentioned in the osnews.com summary:
– These tests are not threaded, so they use *one processor only*
– They estimate that the 2.0Ghz G5 is about 20% *slower* than the new 3.2Ghz P4 systems on scalar for this particular test (estimatee by extrapolating the P4 data from their current 2.66Ghz P4)
– However, they say to take the 20% with a grain of salt, because the G5 is a new chip and as compilers for it mature they could close or erase the gap
– The G5 performs about the same as a G4 clock-for-clock on vector operations, but of course the G5 has a higher clock speed.
– G4 was already a monster on vector, so G5 is even more so.
They don’t compare to P4 vector operations, but pretty much every benchmark I’ve seen shows Altivec beating every other vector unit out there handily. So to me, bottom line is that a 2GHz G5 is similar to a 3GHz P4 on scalar — maybe slightly slower — and probably significantly faster if the code uses Altivec.
It’ll be interesting to see more real-world tests as more shops get their hands on G5 systems. In particular, I’m curious about tests that can take advantage of both processors, and tests that need high bandwidth.
Some real validation of what Apple and IBM have been saying all along about the G5 Processor.
Not really. From the summary of the article:
The test was well documented and concludes that “the G5 has about 22% better scalar floating point performance per clock cycle than the G4 systems tested and 32% better floating point performance per clock cycle than the P4 systems tested.”
Consider the P4 currently enjoys a 60% lead in clock cycles. 32% better performance ber clock cycle is not sufficient to place the G5 ahead of the P4, as Apple contended in their claim that the G5 is “The world’s fastest personal computer” and also in their benchmarks.
Also notice they’re using the Portland Group and Absoft compilers, because they are dealing with Fortran 90 code. Once again, I’ll reiterate that GCC is *not* typically used for compiling scientific computing applications.
Some real validation of what Apple and IBM have been saying all along about the G5 Processor.
Not really, it says right there that the P4 does 255 MFlops and the G5 does 254 *LOL*
If I read the test correctly, the test is not bounded by memory. When you consider memory and the possible use of AltiVec, Apple’s claim may still be true (they do say fastest computer not computer with the fastest chip). I will also wait for the final verdict once compilers that are optimized for the G5 are released.
twocents
Not really, it says right there that the P4 does 255 MFlops and the G5 does 254 *LOL*
But 30% better MHz for MHz. In other words, remove Intel’s manufacturing edge and G5 comes out well ahead. Unfortunately, it’s not so obvious from NASA’s data that G5 is ahead of G4 on an equal-MHz basis.
“Not really, it says right there that the P4 does 255 MFlops and the G5 does 254 *LOL*”
That’s for one G5 processor, the machine has two. How many P4’s can you put in one box?
“Interesting test but really, how well will OS X take advantage (if at all) with the new chip? I would guess to say that scalar floating point is not a main part of the OS.”
Considering the fact that OS X offsets UI rendering to the GPU which of course depends heavily on floating point operations, you could bet that the UI will be significantly benefited by this, but then again, its just a UI it doesn’t require INSANE amounts of graphics processing. In the case of OS X, which is factor based, the UI is benefited by offloading to the GPU, but as we’ve seen in recent Macs Quartz extreme was ALREADY massive overkill for the UI… which ofcourse allowed it to do a lot more backflips without even a hickup.
So, will OS X’s UI be able to take advantage of the G5’s huge floating capabilities? Sure. Will you notice a difference? Almost certinly not. it was already incredible fast.
> That’s for one G5 processor, the machine has two. How many P4’s can you put in one box?
More than 2 of the Xeon variety actually!
Here’s another angle: how many G5s do you *need* to beat a single P4?
Spelling error:
factor based = vector based
“how many G5s do you *need* to beat a single P4?”
Depending on the Application… anywhere from as little as 1/7th of a G5 or as much as 1/2 of a G5.
“That’s for one G5 processor, the machine has two. How many P4’s can you put in one box?”
One. But you can fit a lot of Xeons in a box.
> Depending on the Application… anywhere from as little as 1/7th of a G5 or as much as 1/2 of a G5
Lets see…
G4=254, P4=255, hence it takes more than one G4 to beat a P4
Q.E.D.
“One. But you can fit a lot of Xeons in a box.”
Too bad the processor itself costs as much as a complete G5 tower. You can build a G5 cluster too you know.
Lets see…
G4=254, P4=255, hence it takes more than one G4 to beat a P4
It takes 1,0039370078740157480314960629921 G5’s according to my calculations 😀 [(1/254)*255]
> Insanely Great Mac informs us that the final release of
> Panther definitely won’t support beige G3 or earlier
> PowerMacs, nor will it support PowerBook G3s that lack
> built-in USB ports.
I don’t know about everyone else but this is news that I have been waiting for for quite some time. Thank goodness they haven’t removed support for all G3’s yet. This PowerMac 350Mhz still has some life left!
[Some uninteresting numbers on how many G5s you need to beat a P4…]
Does it really matter? Some code will run faster on the G5, some on the P4… The G5 might be more efficient, but I heard that the P4 could be scaled up to 10GHz. Each architecture have its own advantages.
Oh, and before I start hearing the benefits of RISC architecture and why the G5 is so l33t… The G5 can’t run natively >90% [my guess] of the software on the market today. The best technology doesn’t always “win”.
“The G5 can’t run natively >90% [my guess] of the software on the market today.”
And yet it runs 100% of all the software I need it to.
The thing that everyone else seems to have overlooked (the PC freeks or so concerned about loosing the pissing contest) is the fact that Nasa is looking at using G5s. That says a lot in my book.
Good for you, but not everybody is in the same situation as you.
Anyway, How do you know that all your software will work flawlessly on it? Did you tried it?
“Good for you, but not everybody is in the same situation as you.”
Fortunately, OS X has all the necesseray software to keep 90% of the computing populace happy.
“Anyway, How do you know that all your software will work flawlessly on it? Did you tried it?”
I do (and have been) every day since OS X’s release.
First off, GO FORTRAN!! it will live forever in engineering. I hope as they mentioned breifly that there will be some G5 optimized FORTRAN compilers coming out.
Next, it’s nice to see some very real world results coming that are all about the cpu. x86 vs PPC comparisions should do all they can to take the os and such out of the picture, this means using the most optimized compilers for each cpu.
I wish they could have done something with the dual cpus. I find it odd that this program wasn’t threaded seeing it was a fortran code that seamed to be meant for clusters.
I think the last line was very important to.
“Finally, it is important to note that the current test does not factor machine cost or intended use into the picture, and that can have a large impact, especially in clustering applications.”
I think this will be a killer for the mac. For building a large cluster and you don’t have large funds, or want to get the most bang for the buck, you can be sure they will be building their nodes, not buying dells. Or maybe doing like Sandina (sp?) national labs did and use Shuttle XPCs. Either way when doing this the price performance of the x86s will still probably hold solid. with that being said, seeing a wall of G5’s in racks as a cluster would be something to look at for sure.
“I think the last line was very important to.
“Finally, it is important to note that the current test does not factor machine cost or intended use into the picture, and that can have a large impact, especially in clustering applications.”
I think this will be a killer for the mac”
Hold on there bud, you don’t even know how much Apple plans on pricing G5-equipped X-serves yet. Considering the fact that the the PPC970 is significantly less expensive than the XEON, Apple/IBM will have a huge advantage here.
Fortunately, OS X has all the necesseray software to keep 90% of the computing populace happy.
Right… Then explain me the low market share?
I’m not anti-Apple, in fact, I’m interested by the G5, but
I’m not blind at the facts. Most people I know don’t even consider an Apple computer (or any other OS than Windows) when they want to get a new computer just because of that.
I do (and have been) every day since OS X’s release.
But the G5 is a new CPU… It wasn’t even made by the same company. How can you be sure? *sigh*
< “Fortunately, OS X has all the necesseray software to keep 90% of the computing populace happy.” >
“Right… Then explain me the low market share?”
When most people use the word “market-share”, what they really mean is “installed-base.”
For example, while Apple’s Macintosh market-share may be 3 percent, its installed-base is approximately 10 to 12 percent of the computing industry, a figure that’s roughly similar to that of Linux based PCs.
When these figures are coupled with the remaining alternative operating systems on the market, Windows installed-base works out to be somewhere in the way of 80 percent — a far cry from the 95 figure that is often touted.
So how does market-share play into the picture you ask?
Market-share is determined by quarterly or annual sales figures. The problem with market-share statistics is that it implies that all computers retain the same level of usability over time. It assumes that once a computer is sold, it will retain its productivity status for as long as its parts continue to function.
Unfortunately, usability statistics and replacement purchasing habits of consumers vary significantly between platforms thus causing the market-share figure to look skewed.
Linux users (for example) are known to keep aging computer hardware useful long after it was left for dead by its former Windows using owner. The open source community consistently manages to squeeze every last ounce of processing power from even the most aged hardware available.
Similarly, Mac users are known to keep their computers as primary productivity tools until the gears fall off. This is really a testament to the quality that Apple incorporated into its hardware and software over the years.
Unfortunately, the incorporation of quality into these platform’s coding efforts will only fuel the notion that they are far less popular as what they are as long as market-share is the most commonly used gauge to determine platform popularity.
“Most people I know don’t even consider an Apple computer (or any other OS than Windows) when they want to get a new computer just because of that.”
So you’re saying that they don’t consider a Mac because of their misconceptions about market share.
“But the G5 is a new CPU… It wasn’t even made by the same company. How can you be sure? *sigh*”
because its of the same lineage. The G4 was PPC. The G5 is PPC. There wont be any lost compatibility for that reason.
Quote from conclusion…
Based on an extrapolation of current P4 results, the 2GHz G5 would lag newly announced 3.2GHz P4 systems in Jet3D scalar floating point performance by about 20%, but this kind of comparison is best deferred until G5-aware compiler tools become available (since a 20% performance gain is well within the potential of compiler optimization).
I found this to be the most important part of the whole article. Since I couldn’t be bothered reading the whole thing, I’ll save it for a rainy day.
You can only really compare to architecture according to how well the same apps haved been optimised for each architecture.
Until then on paper the G5 looks better than a P4, but obviously Intel are pursuing different angles on the P4, eg. Lets see how high our Gigahertz can go.
It is like how some car manufacture create superb engines, whereas other decide lets give it more power and waste more fuel!
For example, while Apple’s Macintosh market-share may be 3 percent, its installed-base is approximately 10 to 12 percent of the computing industry, a figure that’s roughly similar to that of Linux based PCs.
I sincerly doubt that their installed base is that high, but it’s still interesting, and I could be wrong. Do you have any source?
So you’re saying that they don’t consider a Mac because of their misconceptions about market share.
No no, I was answering to your previous point: the software. Sorry if I wasn’t clear enough.
because its of the same lineage. The G4 was PPC. The G5 is PPC. There wont be any lost compatibility for that reason.
Theorically, yes, but AFAIK, they’re not manufactured by the same company. That’s why I think that it *could* have some issue with some (probably poorly-written) software. You don’t have a G5 at home so you can’t certify that everything will be alright, like I can’t certify that every x86 program will work flawlessly on the Athlon64. I hope you understand what I mean.
Anonymous (IP: —.ph.ph.cox.net) wrote:
is the fact that Nasa is looking at using G5s. That says a lot in my book.
What I don’t quite understand is: with any type of scientific research conducted by use of computer power, the researchers claim that accuracy is far more important than computing speed. Yet the first generation of Altivec was not able to make use of double-precision floating point operands. Therefore, it may produce less precise numbers/figures/what have you.
I remember hearing that IBM will add their own extensions to the Altivec unit on the 970 but does anyone know if it’ll be able to handle double precision floats?
“Do you have any source?”
I did, but now i cant find it anymore. I’ll keep looking. If/When I find it I’ll post it in this thread.
“Theorically, yes, but AFAIK, they’re not manufactured by the same company.
manufactured no, but designed yes. The powerPC processor was jointly designed by Apple IBM and motorolla in what has since been regarded as the “AIM alliance”
“You don’t have a G5 at home so you can’t certify that everything will be alright, like I can’t certify that every x86 program will work flawlessly on the Athlon64. I hope you understand what I mean.
I do understand, but the Athlon64 wasn’t designed in cooperation with Intel as was the case with the G4 with Apple IBM and Motorola. The G5, while being designed solely by IBM (and to a smaller extent Apple) will use the same core technology that made up the G4 as well as the G3. They’re all PPC.
Really, your analogy would be closer if you were to ask if your PIII software will be compatible on your PIV.
“I remember hearing that IBM will add their own extensions to the Altivec unit on the 970 but does anyone know if it’ll be able to handle double precision floats?”
If they add anything to it, they will be adding to the genuine altivec instruction set (not a variation of altivec or even a “compatible” version as some have suggested) IBM has specificaly mentioned that it was using genuine altivec technology.
Why the hell did they use an old version of the Portland Group compiler when is very well known that Intel optimizing F90 compiler is almost unbeateble and further more freely available for Research institutions??
In numerical computing you cannot just evaluate the processor itself, you have to consider also the compiler availabilty, performance and pricing. See what AMD is also doing with the Opteron http://www.theinquirer.net/?article=10292
Side Note: Where I work (University) I bought the Absoft compiler for a MAC, do you want to know the price? Well 1000$. Intel compiler is free and has SSE2 hand optimized (by intel engineers!!) BLAS 1 2 3 and LAPAC libraries
Obviously, a lot of people seem to be selectively reading the NASA document. So let’s take a look at this comment:
“Consistent with earlier Jet3D tests, the vector version of Jet3D runs an order of magnitude faster than the scalar version (speedups of 10X-13X are typical).”
So while a P4 has a 60% advantage is clock speed to the G5’s 32% better performance per clock, when you run code that uses the Velocity Engine, you get 10-13 TIMES the performance for regular scalar floating point operations.
Let’s repeat: when you use vector calculations instead of standard floating point, the G5 deliever 10 to 13 times the performance on average.
Of course, not everything can be optimized to use the Velocity Engine, so you don’t get 10 times a performance boost across the board since in real-world situations, you’ll be mixing both scalar and vector calculations. This is essentially what we saw in the app bake-off, when we saw the G5s run highly computationally intensive apps 2-3 times faster than the dual Xeon.
Even with SSE, there’s no way a P4 is touching a G5 when it comes to Velocity Engine optimized, floating point intensive apps.
manufactured no, but designed yes. The powerPC processor was jointly designed by Apple IBM and motorolla in what has since been regarded as the “AIM alliance”
[…]
I do understand, but the Athlon64 wasn’t designed in cooperation with Intel as was the case with the G4 with Apple IBM and Motorola. The G5, while being designed solely by IBM (and to a smaller extent Apple) will use the same core technology that made up the G4 as well as the G3. They’re all PPC.
Really, your analogy would be closer if you were to ask if your PIII software will be compatible on your PIV.
Um, well, nevermind then. I forgot that IBM was in the alliance, hehe.
“Um, well, nevermind then. I forgot that IBM was in the alliance, hehe.
“
: )
Not really, it says right there that the P4 does 255 MFlops and the G5 does 254 *LOL*
Unless you include the Altivec scores (2755 MFlops / 5177 for the dual). There’s no doubt that the G5 is competitive with the P4 on most apps, and really fast on Altivec enhanced apps.
I don’t quite see why everybody is so involved in the numerical characteristics of the G5.
I mean, Apple *must* stress the fact that they produce the “fastest” PCs, they are sellers, not critical opinionists… and this must be corrected by the community with the truth.
But… the G5 seems to be (if I’m not wrong…) a very fast CPU which dissipates very little heat (around 30W).
So, who cares if that goes like a 1800+XP or a 2.6P4?
My 1800+ with crawling slow sdram 100Mhz is a powerful machine when it comes to rendering, and with the MEncoder makes almost 22 fps in DVD->DivX compression; a G5 would surely do faster, why bother it is not fast as a 3.06-3.2P4 which makes 82 or more watts making your machine noisy as hell with fans?
The old G4 was a good chip. 2 years ago.
Now they were really trying to survive selling “top of the specs” G4 13xxMhz… G5 brings a lot of computing power more, cool computing power, and it runs the incredible OsX or (my choice) Linux.
It’s your alternative, but do you really take care of a slight 10 or 20% more in computing power when buying?
I mean, 10% more means usually *no* difference at all, unless you look at the benchamrks.
Now that the performance gap has been, if not removed, at least levelled (this is what is important, IHMO) I’d buy a PowerBook G5 if I had the money (and if they did it 🙂 but I’ll have to stick with a P4m.
Ciao,
Lorenzo
well G5 = 64 bit system, thought P4 = 32 bit. Lets see when Intel and AMD really but their own REAL 64 bit chips out. Then Apple can back his stuff to pocket and start to move at home
“Apple *must* stress the fact that they produce the “fastest” PCs, they are sellers, not critical opinionists… and this must be corrected by the community with the truth.”
Corrected?! They aren’t wrong.
“The old G4 was a good chip. 2 years ago.”
its still a good chip, just not for ultra high-end uses like the G5 will be ocupying.
“Now they were really trying to survive selling “top of the specs” G4 13xxMhz…”
Applications that took advantage of the G4’s Altivec instruction set, caused the G4 to remain very competative with the Pentium. Unfortunately, not enough developers integrated altivec code into their apps early on. While that is changing now, it was a bit too late about 4 or 5 months ago, when Apple DESPERATELY needed them.
“G5 brings a lot of computing power more, cool computing power, and it runs the incredible OsX or (my choice) Linux.”
While you are correct that it will run Linux, Linux needs to be updated to take advantage of the computerized fans. Without that support the fans will all go on in unison, causing the quiet G5 to be as loud as some of the more noise PCs. Thankfully, that wont keep Linux users from taking advantage of all that extra speed. I’m sure that eventually the OS will get support for the computerized fan system
“do you really take care of a slight 10 or 20% more in computing power when buying?”
For most functions, the G5 will be significantly more than 10 to 20% in speed over the Pentium.
>Theorically, yes, but AFAIK, they’re not manufactured by the >same company. That’s why I think that it *could* have some >issue with some (probably poorly-written) software. You don’t >have a G5 at home so you can’t certify that everything will >be alright, like I can’t certify that every x86 program .will >work flawlessly on the Athlon64. I hope you understand what I >mean.
You should be really read a book on instructions set architectures and compilers before posting responses. You might also want to understand the history of the powerpc before making claims of incompatibility. AFAIK, the PPC was a joint effort by Motorola, IBM and Apple. Motorola and IBM have manufatured PPC chips for Apple. Motorola stopped making ppc chips because of financial reasons. IBM decided to make a scaled down verison of thier Power4 + Altivec and call it PPC 970 which is what apple brands the G5.
Any code, poorly or elegantly written, in a higher level language should be compiled into correct object code by a compiler (obviously code that compiles). If the compiler has bugs it will probably generate bad instructions and those should execute the same on a G4 or a G5 (cause illegal instructions faults or the like) since they are built around the same instruction set.
I am sure IBM has a lot of experience in testing and validating thier processors. Compatibility would be the last of my concerns with two processors using the same instructions set. Because the the instruction decode logic would be the first to be tested and validated.
I would be more nervous moving from and x86 platform to the itanium.
“well G5 = 64 bit system, thought P4 = 32 bit. Lets see when Intel and AMD really but their own REAL 64 bit chips out. Then Apple can back his stuff to pocket and start to move at home
“
The G5 is really only the beginning. Being based on the Power 4 it definately has legs, but the power5 is already out. IBM has said that the Power 5 will be 400% faster than the Power 4. Assuming that speed also translates to the consumer dirivitive (980?) things will be very interesting.
Also of interest is the fact that it has been said that the Power 5 will be designed to reach significantly greater GHz numbers which should allow Apple to run in paralyl with Intel which is doing a nice job of increasing GHz be ever-extending the processors pipeline stages.
“You should be really read a book on instructions set architectures and compilers before posting responses.”
its probably appropriate to lay off him, he DID eventually say, “I forgot that IBM was in the alliance”. He was very good natured about recognizing the mistake..
To the PC fans — keep using your PC — I really don’t care.
On the software issue — who uses even 1% of the available Windows software. There’s a lot of junk out there. But, if you have to use some software that’s only available on Windows, then buy a PC.
I have yet to find any software functionality I wanted to use that was not available on Mac OS X. In every case, the Mac alternative has been as good as or better than the Windows alternative.
Finally, don’t forget that Mac OS X is Unix based. There are literally thousands of open source programs available for the Mac.
What’s wrong with you, examining things calmly and rationally and coming to a logical conclusion about what’s best for you? Where do you think you are? We want flames, man! FLAMES!
(-1, Sarcasm)
N.B.: I’m aware Apple did bring this on themselves with the “fastest computer EVAR!!!” claim, but realistically, even if by some miracle they’d been quiet and modest about introducing the G5s, critics would still be piling on ’em. I observed to a friend the other day that while Mac people can be arrogant bastards about PCs, they tend to attack the computers; PC users tend to attack Mac users. The idea that someone could buy a machine with so many supposed drawbacks (speed, price, software selection, mouse buttons, whatever) and not only be steadfastly happy but tell friends, “Yeah, these are great machines!” obviously makes some people absolutely livid.
Before the G5 ships, Intel will reduce prices on Xeons on July 17th making Dual Xeon machines much cheaper.
And at the same time, Intel will introduce a 3.06ghz Xeon with 1MB cache.
With the 1MB cache (vs. the G5 at 512K cache), the Dual Xeon will be much faster.
http://www.theinquirer.net/?article=10303
Before the G5 even ships, AMD will be shipping their 2Ghz Opteron, also with 1MB cache, that is also faster than the G5. And you can get Opteron systems today, from 1 processor to 8 processors.
So all this goes to say… if you want a G5 just go ahead and buy it. Performance is so fluid in the PC market, that many benchmarks expire before machines even ship. It’s just noise.
One. But you can fit a lot of Xeons in a box.”
‘Too bad the processor itself costs as much as a complete G5 tower. You can build a G5 cluster too you know.’
What are you on! A 2.8 GHz Xeon processor costs about US$600 and a 2.4 GHz about US$450. A quality dual Xeon MB is $400-1000.
The idea that someone could buy a machine with so many supposed drawbacks:
(speed
Apple’s hardware is faster. No drawback here…be it perceived or actual.
price
Apple hardware, when compared spec for spec in both hardware and software is either slightly more expensive, the same price, slightly less or significantly less. No drawback here…be it perceived or actual.
“software selection”
While there is indeed more software for Windows than for the Mac, many of those applications for Windows are lesser versions of a superior version. The superior version (available on Windows) is typically also available on the Mac. if it isn’t there is typically an equivilent or superior alternative. No drawback here…be it perceived or actual.
“mouse buttons”
Just like Windows, Macs support mice with multiple buttons. No drawback here…be it perceived or actual.
“Before the G5 ships, Intel will reduce prices on Xeons on July 17th making Dual Xeon machines much cheaper.”
Considering the fact that the Dual XEON machine( while also being slower) cost more than $1000 more than a similarly equipped Mac. I sure hope they reduce those prices GEEZ!
“And at the same time, Intel will introduce a 3.06ghz Xeon with 1MB cache.”
Wow, that will halp with the speed deficiency very nicely actually.
Some real validation of what Apple and IBM have been saying all along about the G5 Processor.
Like always (doh), I beg to differ. Look at this, ” 32% better floating point performance per clock cycle than the P4 systems tested.” P4 of course having more clock cycles per second…
And NASA doesn’t go further into other variables (read the documentation) like optimization, operating system, compiler, etc. Apple, using SPEC, went into there and use a set of configurations only a minority of PC users use (with RedHat and GCC).
Too bad the processor itself costs as much as a complete G5 tower. You can build a G5 cluster too you know.
You also can build a P4 cluster, you know.
Obviously, a lot of people seem to be selectively reading the NASA document. So let’s take a look at this comment:
“Consistent with earlier Jet3D tests, the vector version of Jet3D runs an order of magnitude faster than the scalar version (speedups of 10X-13X are typical).”
Holy cow. You’re right; I don’t know how I could have missed this. So basically the G5 seems to be somewhere around a high-end P4 on scalar, but if you can take advantage of Altivec… wow.
Also notice that while they don’t graph it, they do mention how the G5 scored with both procs enabled. In scalar, it almost exactly doubled the single-proc performance, which shows that the SMP overhead is low. For some reason on the Altivec version it didn’t come as close to linear speedup. But once you’ve gotten a 10x speedup, I doubt you’ll care too much if the other proc only gives you another 1.75x instead of 2x!
I also think you’re right about this explaining Apple’s WWDC application demos. The 2x performance of the G5 to the Xeon in the demos was probably an average of the fact that it’s about even on scalar, but way way faster for small portions of code that can be vectorized. Bandwidth might also have come into play in the demos as well.
You should be really read a book on instructions set architectures and compilers before posting responses. You might also want to understand the history of the powerpc before making claims of incompatibility. AFAIK, the PPC was a joint effort by Motorola, IBM and Apple. Motorola and IBM have manufatured PPC chips for Apple.
Maybe, but you should really read a book on “how to read a forum”, especially posts made 30 minutes before yours… And yes, I did really forgot that IBM was in that alliance. I’m not a Mac user, after all.
Btw, there’s always a difference between theory and the real world. Yes, I know that CPUs with a identical instruction set should behave exactly the same way, but bugs happens.
While there is indeed more software for Windows than for the Mac, many of those applications for Windows are lesser versions of a superior version. The superior version (available on Windows) is typically also available on the Mac. if it isn’t there is typically an equivilent or superior alternative. No drawback here…be it perceived or actual.
Macs don’t have a whole lot of applications Windows have. Sure, it has the top few major applications (Office, Photoshop, Quark, etc.), assuming people are
a) Willing to fork out a whole lot of money when they are happy with cheaper PC software doing exactly as they wish (e.g. Paint Shop Pro over Photoshop). But on the worser extend…
b) Willing to leave a niche application for a “all-round” application that doesn’t do as some people need. In other words, the top few applications does not, and I repeat, does not cover every single niche. For example, a whole lot of Malaysia lawyers (I’m willing to venture a bet that this is the same elsewhere) would drop dead if WordPerfect is snatched away from them. WordPerfect, the last I check, isn’t available on the PC.
I personally don’t mind moving to the Mac (I just question the investment I need to put in with leaving the PC – new software licenses, new hardware, etc.) because all my applications are available on the Mac. BTW, neither Paint Shop Pro nor WordPerfect Office, or any other applications like AutoCAD 2004, aren’t available IIRC on the Mac.
If the Intel processors are soooo good, how can that be, 32 pieces of Power4+ processors kicks 64(!) pieces of the high end Intanium craps in TPC-C?
http://news.com.com/2100-1010-1022355.html
Come back to the floor poor x86 belivers!
If I remembered correctly, when G4 was released, it was not only tonnes faster than its x86 counterpart, but somewhat cheaper. A month or two ago, G4 was not only loads slower and tonnes more expensive than it’s x86 counterpart. And if I remembered correctly, it didn’t took long for x86 to close the gap.
Now look at G5s. It may be faster than x86, but largely it is around the same speed. It is loads (somewhat) cheaper than Xeon machines from all major OEMs. How long is that gonna last, especially since the gap is smaller than when G4 was first released? That depends on how fast Apple and IBM churn out upgrades and reduce prices. How often would Apple reduce prices and release new machines? Unless they are trying out a new bright business strategy – not as fast as Intel (or AMD for all that matters).
What I’m trying to say is, enjoy flaming Intel and the PC world while you can, cause in a short year, or maybe even less, they would be back with vengeance. Unless by some miracle on Apple’s (and IBMs) part.
>>The G5 can’t run natively >90% [my guess] of the software on the market today. The best technology doesn’t always “win”.
The G5 can run 32 bit and 64 bit OSX applicaitons natively.
Oh wait, you were talking about how it won’t run windows software natively. It doesn’t even occur to me anymore that someone would want to run windows software on an OSX box now that I’ve used OSX for a couple weeks. It would be like trying to run an Apple IIe application on OSX.
There are still more good games for Apple ][ than Mac OS X.
And Diversi-DOS kicks ass on whatever bloatware crap they have in OS X.
I mean, Apple had to get rid of the floppy drive so that people couldn’t run their Apple excellent ][ software, but had to move to Mac.
🙂
“Not really, it says right there that the P4 does 255 MFlops and the G5 does 254 *LOL*”
But hey, the P4 does 255 32-bits MFlops and the G5 does 254 64-bits MFlops.
I just noticed that the G5 does not support ECC RAM. That means all those extra bits may not work right! There is no guarantee that the system is working properly!
The Space Shuttle is going to blow up if they use the G5!
Better stay on those ECC Xeons and Itaniums, NASA. Now is not the time to blow up another Shuttle…
On the software issue — who uses even 1% of the available Windows software. There’s a lot of junk out there.
a) Everyone uses a different .1% of Windows software and b) Windows users have a lot of choices within a single category.
I’m just curious though why Apple zealots think PC users are threatened. x86 architecture is a common butt of jokes. It’s not suprising to hear it can be improved. I hope IBM gives AMD/Intel a run for their money, without harming AMD.
For clustering apps, cost is a vital factor. But OTOH, for workstations you often want to maximize the per-processor strength.
Anyway, we should keep in mind we’re talking about floating-point, which is unimportant for most people.
“But hey, the P4 does 255 32-bits MFlops and the G5 does 254 64-bits MFlops.”
Nope… the P4 has almost full IEEE 754 FP support, and the FPU performs double precission. I.e. those were indeed 64bit MFlops on the pentium
.
I like the G5 myself, but give Caesar unto Caesar that which belongs to Caesar.
AM I the only one that thinks this test is bullshit?
Is it fare to compare A PowerMac G5 2x2GHz with a P4 1×2.66GHz?????
Why don’t they compare with P4 2x2GHz???
And yes I know there is more than Hz, but should the total be “equal”
????
Its not even fare to compare a 2x1GHZ with a 1x2GHz… so why this!!!??
from the article
“Note that even though the G4 and G5 systems have dual processors, detailed benchmarks in the present study pertain to a single processor only.”
read the article first
just another bunch of guys trying to do their job – may be to make the foam stick
> read the article first
Yeah I read it. .but still I am asking the questions of fareness!!!
Build a box with a double-P4 configuration…if you can… I hope now you’re understand why.
HAHA.. please I already know that…
I am not asking about that..
I am simply asking if its fare to test like this..
the answer is NO its not fare…
Your answer is no, but my answer is yes.
Or is it fair to compare 64 Itanium 2 with 32 Power4 ?
Yes, beacuse they are used to the same target and in the same costal range.
Well, we have to look at the big picture here.
A few years ago, The Intel/Microsoft was almost the only combination you could get in PC, then came AMD with the Athlon, then Linux, then Apple OSX then AMD (again) with Opetron/X64 then IBM with the G5, then… etc etc
Nowadays you have the freedom to choose the combination of processor/OS you want on your box (with some limitations obviously) instead of being stuck with just Intel/Windows.
This fact only is priceless and that’s what people should be looking at not argueing about a few percent difference in perf.
Now, one of Intel’s strenght’s is its long pipelines (and as a result) it’s high frequenzy. Now, it would be rather unfair to use a 2GHz Pentium 4 which is rather old, without all the new Intel features that can be used in this test (higher cache for example), just to have a “fair” competition. The G5’s high floating point calculations per clock cycle is its strenght, the P4’s sheer amount of clock cycles is its strenght. Just like it would be unfair to compare both processors on just MHz, it would be unfair to compare both processors on your suggestion.
Thanx rajan r.. thats the answer I was looking for
And also thought about..
They also used a 2.66GHZ P4. I do think that they should have performed the test with a 3 or 3.2 GHZ P4. Maybe even a dual Xeon.
from the Jet3D Test…
“but a look at the normalized levels reveals that the G5 is nearly identical to the G4 in terms of vector performance per clock cycle. Thus, the increased raw vector performance of the G5 is largely due to its higher clock speed.”
There is little to no optimization of Altivec in the G5s. The gains come from moslty MHZ so in this aspect there is still more power to be extracted from the G5 not even factoring in issues with non optimized compilers and Panther which will probably be more optimized for G5.
The only thing that really matter is that Apple was stuck in a bad place – little incremental speed boosts in the G4, slow bus and slow-down in memory speed and al the ramifications that go with that.
Now they have 64 bit computing, very fast processors, very fast bus and RAM. Now we can run OS X and it will scream. This is great news for all of computing. Competition is back, Apple is healthy and can compete again. This helps move computing forward.
on a G5!!!
I thought people might be confused by the test where the P4 had 255MFLOPS and the G5 scored 254. If you read the article a little more closely however, just above the test is this:
“Though dual processor benchmarks are not presented in detail here, it is worth noting that the G5 system benchmarked at 498 MFLOPS and 0.125 MFLOPS/MHz for scalar Jet3D performance when two processors were used.”
That’s impressive for more than one reason. One is how well this chip scales up in performance in multiple processor systems. By adding a second processor, it only dropped from 0.127 MFLOPS/MHz to 0.125 MFLOPS/Mhz. The P4 scored 0.096 MFLOPS/MHz with a single processor.
The 498 MFLOPS score for the G5 compared to a 255 MFLOPS score for the P4 puts this test on par with Apple’s benchmark.
This reminds me a bit of the republican democrat arguments here in the US.
25% of either side will never admit that the other side can be right…ever. Its all black and white. Polarization. I am right you are wrong.
30-40% people are roughly moderates and will listen to some logic. The rest just don’t care.
Likewise, Apple could have a 20 GHz 128 bit G28 processor with 30 Gig of memory tomorrow sold for $1400 and you’d still have these arguments. Discussion is good but don’t dilude your selves logic is not as powerful as protection of one’s own views.
For whatever reason there is a need to see apple down. I don’t know why but it is impossible for some to accept that apple can do right. Just ignore it and focus on those 30-40% moderates and the group that just don’t care.
Quote from IBM, that Apple was involved in developing the G5.
Chakib Akrout, (IBM): Yes.
DMN(http://www.digitalmedianet.com): Can you give us an idea of what the difference between this new G5 is and the Power4 chip that existed a year ago or so? Just give us a roadmap of where you’ve gone from there to here.
Akrout: The G5 is actually a derivative of Power4. It’s basically the same processor which we took from the Power4 and we brought it to G5. We’ve added in the G5 beside the processor, the Velocity Engine, which optimized with the same kind of pipeline, with more enhancements we were able to put there. It connected to the fastest front-side bus, which we developed with Apple, which is able to get up to 1GHz. For this we used the most leading-edge process technology we have today. 130 millimeters, with the silicon-on-insulator, copper technology, eight levels. This is our leading edge technology from architecture, micro-architecture optimization and process technology.
I realized after you two didn’t know what each other was fighting about… 🙂
ah ha pc losers, even in “real world” stats you still act like losers. Risc is good!
This is the fifth or sixth post in this thread that preach the same message – if ya don’t like the debates – stay out of it. I don’t butt in (well, not lately, I stopped) into debates I don’t like and ask them to stop. No one will listen to you.
rajan,
I have nothing against debate as my presence will indicate. By the same measure i can’t ignore the apparent need of some to bash apple regardless of the company’s actions, regardless of facts. And its SOOOOOOOOOO OBVIOUS that well i have to mention it.
I agree, I’m a great fan of AMD and Intel’s processors. It just it seems when it comes to Apple people bash them when they get something right. So who is living in reality destortion? Good question, when even Nasa says positive things about the G5 performance against the P4.
Apple debates make OS NEWS possible, look at the traffic that comes in just for Mac news. LOOK at the number of post by each Apple article,,,,,,funny!
Apple’s benchmarks which were highly criticized by some gave the Dual 2GHz Power Mac G5 a 194.5% performance advantage over a 3GHz in SpecFP base tests. The G5 getting a score of 15.7, and the P4 getting an 8.07.
NASA’s study found the Dual 2GHz Power Mac G5 to score 498 MFLOPS for their Jet3D performance. A P4 running at 2.66GHz scored 255 MFLOPS. This is a 195.3% performance advantage for the G5 in this test. Making adjustments to scale the P4 up to 3GHz and using compilers aware of the G5, these test results look remarkably similat to those Apple and Veritest published last month at the WWDC.
Dear Stupid,
Please enlighten us which store NASA goes to, CompUSA or BestBuy, to purchase the chips that run the Space Shuttle.
“The Space Shuttle is going to blow up if they use the G5!
Better stay on those ECC Xeons and Itaniums, NASA. Now is not the time to blow up another Shuttle…”
(Your’e kidding, right? Forgive me, this is typical of someone shooting their mouth off and being insensitive)
Please continue.
“Does too”
“Can not”
“Oh yeah”
“Yeah” …
P4 (2.66)=255
P4 (3.2)=255+20%=300 (estimated)
Though dual processor benchmarks are not presented in detail here, it is worth noting that the G5 system benchmarked at 498 MFLOPS and 0.125 MFLOPS/MHz for scalar Jet3D performance when two processors were used.
so
dual G5=498 (actual test, and add 20% estimated for FORTRAN optimisations).
=o)
Dear to you also,
Please be careful about the shuttle incident. There are some of us on here that new a person who died on that day.
Ken
This benchmark means ziltch!
We know that both processors don’t do the same amount per clock so what is the point comparing them on MFLOPS/MHZ — none
If they are really comparing processors why not test an opteron or an athlon as well. I think because it will make the G5 look bad in MFLOPS/MHZ comparison. So again is there any point in comparing processors based on MHz — none.
If their aim is price/performance as it should be, they should care about the MHz. They should care about what compilers. The way to do it is this: Set a conservative price range for the whole system (hardware,software,compilers). Test all comparable systems that fall within the range and give a price/performance ratio for the specific type of task being tested. The smallest value is the winner and the one consumers should care about.
Alternatively (more difficult), find competing systems that perform withing a performance window and determine price/performance ratio and conclude as above.
i find it quit bad that they didn’t put some work into the vector unit, as it is very comparable to the G4 and not very well integrated into that ripped off Power4 core. hope IBM will work on that, should be nice to see this together with the FP throughput of the Power5.(sure to check ars technica)
they should not care about the MHz
> Finally, don’t forget that Mac OS X is Unix based.
I think Apple has just been sued by the Open Group for this statement. Apple has been a member of the Open Group for years, but always refused to get a UNIX98 certification for Mac OS X, while constantly advertising their OS as UNIX based.
The point is to some extend that I also find this kind of “marketing arguments” quite strange. Pure UNIX manufacturers like Sun or IBM get a UNIX98 certification and never used “UNIX based” as an argument for their OS. Honestly, it is the extra features that make those OS attractive. Somehow I can’t help me to ignore the impression that Apple actually wants to benefit from the technical achievements of these UNIX manufacturers.
“Macs don’t have a whole lot of applications Windows have. Sure, it has the top few major applications (Office, Photoshop, Quark, etc.), assuming people are
a) Willing to fork out a whole lot of money when they are happy with cheaper PC software doing exactly as they wish (e.g. Paint Shop Pro over Photoshop). But on the worser extend…”
But buying a Mac does NOT cost a whole lot of money. Rather, it costs either a little more, is at the same price, is slightly less, or significantly less when you match both computer’s hardware and software inclusions as close as possibe.
“b) Willing to leave a niche application for a “all-round” application that doesn’t do as some people need.”
That comment assumes that the Mac equivilent is not a better solution as is often the case.
“BTW, neither Paint Shop Pro”
And what does Paint Shop Pro offer me that I couldn’t get elsewhere on the Mac?
“nor WordPerfect Office”
And what does WordPerfect Office offer me that I couldn;t get elsewhere on the Mac?
“or any other applications like AutoCAD 2004”
You do have a point there. While there ARE alternative CAD software solutions on the Mac, AutoCAD has them all beat.
“Look at G5s. It may be faster than x86, but largely it is around the same speed.”
The SPEC benchmarks made it look like it was only slightly faster bet every real world test showed that it was WAY faster. It only confirms what I’ve been saying all along, “SPEC BENCHMARKS ARE NOT A TRUE GAUGE FOR SPEED.
“It is loads (somewhat) cheaper than Xeon machines from all major OEMs. How long is that gonna last, especially since the gap is smaller than when G4 was first released?”
Comparing the speed at which x86 was able to close the gap is not a fair comparison. Apple announced the G4 but could ship for (if I remember correctly) 4-6 months! That had NOTHING to do with the scalability of the G4 but had EVERYTHING with the INABILITY of Motorola to ship and ramp up.
“That depends on how fast Apple and IBM churn out upgrades and reduce prices.”
make no mistake, this situation will not be another Motorola-like induced catastrophe. The ramp up room for these chips is HUGE.
“How often would Apple reduce prices and release new machines? Unless they are trying out a new bright business strategy – not as fast as Intel (or AMD for all that matters).”
That argument assumes that price it the only way to compete.
“What I’m trying to say is, enjoy flaming Intel and the PC world while you can, cause in a short year, or maybe even less, they would be back with vengeance. Unless by some miracle on Apple’s (and IBMs) part.”
No need for a miracle. The roadmap for these chips is so strong it should have no problem keeping a significant leap ahead of x86.
>> “Finally, don’t forget that Mac OS X is Unix based.”
> “I think Apple has just been sued by the Open Group for this statement.”
And the Open Group was sued for trying to lay claim to a term which has long since been deemed generic.
“was sued”
Oops. Make that, “is being sued”
NASA’s study found the Dual 2GHz Power Mac G5 to score 498 MFLOPS for their Jet3D performance. A P4 running at 2.66GHz scored 255 MFLOPS: a 195.3% performance advantage for the G5 in this test. If we assume a direct correlation between MHz and MFLOPS for the P4 (which would actually overstate the performance of the P4) and increase the P4’s score by 12.782% this would give the 3GHz P4 a score of 287.594 MFLOPS. This is still a 173.16% performance advantage for the G5, and NASA states that a 20% increase in performance for the G5 would be reasonable “when G5-aware compiler tools become available.”
So it would seem NASA’s benchmarks go a long way in validating the benchmarks for the G5 that Apple released last month at the WWDC. In fact, NASA appears to be giving the G5 even better scores than Apple and Veritest did.
The vector tests that NASA performed to test the G5’s AltiVec instruction set produce some even more impressive results, and would be a good indication for why the G5 outpaced the Xeon and P4 by such dramatic amounts on real world tests (at times more than 700% faster than a 3GHz P4). “The vector version of Jet3D runs an order of magnitude faster than the scalar version (speedups of 10X-13X are typical).” The dual 2GHz G5 was benchmarked at 5177 MFLOPS (a 1040% increase over the scalar test) and 1.29 MFLOPS/MHz. This also seems accurate considering Ars Technica’s claim that the AltiVec engine wasn’t as well integrated into the G5 as it was in the G4. The 2GHz G5 (single cpu) scored 2755 MFLOPS, or 1.378 MFLOPS/MHz, which shows a slightly larger performance hit for vector operations than floating point operations when moving to a dual G5.
This is a translation from a French website macbidouille.com
It refers to the 970,980 and 990 (g5, G6, G7)
– Even further in the future of IBM – Lionel – 16:49:27
Here infos not confirmed on the future of the PPC 9xx
IBM is working on the engraving of processors in 65nm with East Fishkill. The production with this smoothness of engraving will start in 2005. They will take lead on INTEL which wastes time with its engraving in oil immersion, although the problems of birefringence were solved. IBM gave up this technology, because there are problems with bubbles of air in the fluid.
Roadmap envisages PPC 980 with 90nm for the third quarter of the next year. The maximum frequency that these chips will reach is not fixed yet with certainty, but they count on 4,5 GHz at the beginning of 2005, when the engraved successor with 65nm takes the changing.
At the end of 2005- beginning 2006, the PPC 990 engraved in 65nm will arrive. It should begin its career with 4,5 GHz and finish it to 9:) It will be derived from Power6.
Veiled what to take again confidence out of APPLE for years.
Yep, this seems to be the message PC fanboys have been missing when reading the NASA study.
2.66 GHz P4 = 255 MFLOPS
3.20 GHz P4 = 288 MFLOPS (extrapolated)
Dual 2 GHz G5 = 498 MFLOPS
Dual 2 GHz G5 (using AltiVec/Veleocity Engine) = 5177 MFLOPS
These are the RAW performance numbers, not normalized per clock cycle.
I don’t know about you, but 5177 MFLOPS compared to 288 MFLOPS is a extraordinary huge performance difference to me.
No wonder the G5s smoked Dual Xeons in Photoshop, Luxology, Mathematica, and Logic/CuBase by more than 2 to 1 in the WWDC app bake-off.
All these apps would have taken advantage of the 10-13 times performance speed-up offered by the Velocity Engine’s vector unit. Since in the real world, programs can’t utilize the Velocity Engine 100% of the time, that probably translates to a G5 having a 2-3 times performance advantage over the dual Xeons for computationally, math-intensive apps (which pretty much constitutes virtually every multimedia app today).
5177 MFLOPS compared to 288 MFLOPS – no wonder the PC fanboys are in a tizzy! It’s becoming obvious that all the hoopla about Apple fixing benchmarks is the first stage of denial. 😛
Apple has been making better software for years, everyone agrees; They just never had the hardware to back them up. Then every time they do crop up with better hardware, everyone criticizes them and says that it’s just not possible, PC hardware is always better they say. But now they’ve proven you wrong… TWICE, and some trolls STILL don’t believe them. It’s a sad world.
“They just never had the hardware to back them up.”
Lemme take that back, they havn’t had the hardware for the past 1.5 years. Their hardware has been faster for all years previous to that (exept for the tail end of each upgrade cycle for a particular Mac)