PowerMac G5 Benchmarks (again)

[MacRumors]

For those still following the G5 Benchmark discussions... Luxology posted some details surrounding their demo and their belief that this represented an accurate methodology in comparing systems:

The demo set up itself was designed to require a large number of computes and to push a large amount of data in and out of the chip to show both processor speed and bandwidth. I believe the demo accomplished this in an effective and very fair manner.

Meanwhile, xlr8yourmac.com posts a user submitted reference of a Java Linpack Benchmark. This user notes that the Dual G4 1.42 achieves 90Mflop/s, while the new Dual 2GHz G5 scores 236Mflop/s. While other systems references are given, the benchmark also tests the specific java implementation along with processor speeds... so comparisons across platforms is not necessarily representative.

Users are reminded that all user-provided benchmarks of PowerMac G5's have been done in uncontrolled situations with pre-production machines with non-ideal code.

Macuarium.com (Spanish) posts photos and a first hand account with a PowerMac G5.

 

[a9mike]

More benchmarks...

More benchmarks on these demo machines don't mean squat. We won't really even know what it's capable of until people (like me hopefully) get production models in their hands. Then we shall see...

The Power Of The Macintosh!

Hopefully <crosses fingers>

 

[Pete_Hoover]

I agree. I think we should just wait until the full production models are shipping before we come to any conclusions. That is, if we can stand to wait. It is hard, but I will try.

 

[e-coli]

I don't see any numbers posted????

 

[Rx7 Fan]

Stop all the argument already... Just know that its fast!!! Who really cares if its the fastest in the world or not (I don't)...PCs people just can't accept the fact that APPLE has caught up in CPU performace/benchmarks!!!

 

[P-Worm]

Behold! The Power of the G5! And the people shall come in twos to recieve the blessings of the new processor!

P-Worm

 

[Abstract]

*gasp* Oh no, Steve was wrong!!! It's only the 2nd fastest computer in the world.

*starts strangling puppies and slitting his wrists*

 

[allpar]

Still haven't answered...what about the single-processor G5s? All the specs seem to be at the dual 2.0...which should blow away the single 1.6 by a factor of at least 2:1...

 

[Freg3000]

It really doesn't matter. And we won't know until they really start shipping in volume in August/September anyway. So we have to wait 2 months to truly know. Big Whoop.

 

[k2k koos]

a few words

So the G5 is an amazingly powerful computer, eh sorry, work of art!

So what if there are other less well designed mix and match products that go by the name of PC out there that do perform as good, not as good, or better than a G5, they still do not run OS X!

XP might be the best OS from MS to date ( opinions vary on this matter too...)it is a long shot from OS X jaguar, and the gap is widening with the introduction of Panther later this year.

Bottom line is, the Mac is an overal beter computing experience!

It has its "short comings" (hence the ongoing development and improvements), but they are easely forgiven. Someone once metioned that a PC is used, a Mac is loved and cared for. I couldn't agree more!

Happy computing!

 

[Megaquad]

Anyone knows someone who could run a test on G5. Flash benchmark in my sig.

 

[fpnc]

From the bits and pieces I'm seeing it appears that the new dual 2.0GHz G5 is about twice as "fast" as the previous top-of-the-line G4 Power Mac. This may be a best case number since I've seen some lower results (i.e. the G5 won't be twice as fast in every cpu-bound task), but I've also seen some over 2X results so maybe twice is a reasonable average.

This would also fit with Apple's comparisons at WWDC since Apple marketing is still claiming that the G4 Power Macs are just as fast or faster than top-end PCs (a claim which I believe is very difficult to support in the "real world").

So, if Apple marketing claims that the G4 is still competitive with top-end PCs and if the G5 is up to twice as fast as the G4 then that would certainly allow room for Apple to construct benchmarks showing that the dual G5 is twice as fast as a dual 3.06GHz Intel Xeon.

In any case, in most situations PCs are not twice as fast as the G4 Power Macs. Thus if you extend this line of reasoning I think it's not too difficult to see that the new G5s will at least be competitive with PCs, and in cases where applications make good use of the dual G5 it could indeed be faster than any PC (but __not__ twice as fast).

One area where there may be some issues is AltiVec (Velocity Engine). We know that AltiVec enhanced applications can perform very well on a G4. However, I'm a bit concerned that the G5 may not show much of a improvement with existing AltiVec code. This may be why Apple hasn't directly compared the G4 to the G5.

But I agree with others, it will likely be several more months before we know how the G5 performs. In fact, it might be mid to late September before we see any detailed comparisons ("real world") to the PC and/or G4.

 

[Catfish_Man]

236MFLOPS/s? Nice! We've got the world's first accelerating computer (the S in MFLOPS stands for seconds, so MFLOPS per second is millions of floating point operations per second per second, which is a measure of accelerating floating point power).

Sarcastic comment aside, 236 MFLOPS sucks badly. I realize VMs are slow, but Sun and Apple really need to work on Java, and the Java VM for OSX. For comparison, my dual 867 G4 can get 6,000+ MFLOPS running Altivec Fractal Carbon demo which is a vectorized C app (so at the other end of the efficiency spectrum).

 

[SiliconAddict]

It’s the fastest computer until someone comes out with something faster. Then someone will come out with something even faster. And then someone. BAH! People seem to miss the point. The G5 finally brings Apple in line with AMD and Intel. A few points faster or slower? (Or more then a few depending on your benchmarks.) Who cares!?!? From any report you pick biased or unbiased the numbers speak for themselves the G5 IS fast. I find it somewhat hilarious how many people on zdnet.com are getting their pants in a massive, twisted bunch over this. Everyone needs to give credit where credit is due. Apple got some horses under the hood once again.

Now give me a new Powerbook dang it!

PS- Please tell me the irony isn’t lost on the fact that Apple and IBM now have such an interesting relationship. Wasn’t Jobs insanely anti-IBM at one point?

 

[JoeRadar]

Originally posted by allpar
All the specs seem to be at the dual 2.0...which should blow away the single 1.6 by a factor of at least 2:1

Real-world performance will vary greatly depending on how well a particular application is threaded.

If your application uses only a single thread (probably most applications these days), a move from 1.6 to a dual 2.0 may only see a 25% performance gain.

 

[JoeRadar]

Originally posted by SiliconAddict
PS- Please tell me the irony isn’t lost on the fact that Apple and IBM now have such an interesting relationship. Wasn’t Jobs insanely anti-IBM at one point?

The original Macintosh ad was an implicit stab at IBM and its "you will conform" mentality. I think Jobs is/was anti-establishment, and IBM was the establishment.

Now Microsoft is the establishment, and IBM, with its strong embrace of Linux, is seen more as a rebel.

Of course, Apple has been working closely with IBM since the original PowerPC, which I think was derived from the IBM RT technology.

 

[3.1416]

Originally posted by fpnc
I'm a bit concerned that the G5 may not show much of a improvement with existing AltiVec code. This may be why Apple hasn't directly compared the G4 to the G5.

It will probably depend on how much memory bandwidth the code requires. If it doesn't require much (like RC5 which fits entirely in cache) then the G5 won't be better than the G4 on a per-cycle basis. But for stuff like video processing that streams loads of data, the G5's vastly superior memory architecture should provide much better performance.

As for Apple not comparing the G4 to the G5 directly, that wouldn't be a good idea no matter what the results. Either way it makes one of them look bad.

 

[Lyle]

Originally posted by SiliconAddict
Please tell me the irony isn’t lost on the fact that Apple and IBM now have such an interesting relationship. Wasn’t Jobs insanely anti-IBM at one point?

According to legend, the Apple Computer slogan "Think Different" was a play on an old IBM corporate slogan, "Think".

 

[Jeff Harrell]

Originally posted by Catfish_Man
(the S in MFLOPS stands for seconds, so MFLOPS per second is millions of floating point operations per second per second, which is a measure of accelerating floating point power).

FLOPS can just as easily be the plural of FLOP, which is "floating-point operation." So MFLOPS can mean "millions of floating-point operations per second" or simply "millions of floating-point operations" depending on the context.

 

[inkswamp]

Why the fascination with numbers

Before the G5 was announced and now after, I felt (and still feel) that numbers concerning machine speed is only so much circle-jerking amongst geeks. Does anyone really care if the numbers show the G5 to be the biggest and best? I'm still sailing along happily with my iMac G4 700Mhz and cannot possibly imagine what the big deal is in having the slightest edge in processor speed at or near 3Ghz. I understand that there are high-end usage concerns where buyers want every last bit of speed they can get (I deal with that myself at work), but to concentrate solely on processor speed is ludicrous, whether Apple is in the lead or not.

But now the various Internet forums are buzzing with contentious and pointless arguments about the validity of Apple's numbers or whether it was all rigged or how things were fudged or blah blah blah blah blah... Who cares? It probably was fudged. It's all marketing and Apple's not the only company guilty of that, if in fact they did nudge things a little in their favor.

For my money, a far better comparison of speed and capability was the shoot-out done on stage at WWDC. Particularly impressive was the Mathematica demo. The G5 killed, but geeks don't care about that because it's not quantitative enough to satisfy their need for numbers. But it's that kind of demo that really matters and the kind of thing that non-circle-jerking-geeks can relate to.

 

[JoeRadar]

Originally posted by 3.1416
As for Apple not comparing the G4 to the G5 directly, that wouldn't be a good idea no matter what the results. Either way it makes one of them look bad.

I concur. To show the G5 whipping the G4 is to acknowledge that the G4 had lagged behind, especially considering the G5 was pretty much on par with the Xeon in combined int and fp spec marks (a little behind here, a little ahead there). Furthermore, Apple still sells G4s (iMacs, PowerBooks) and G3s (iBooks).

I just bought a 12-inch PowerBook yesterday. I don’t need to be reminded that the 867 MHz single G4 is woefully inadequate and virtually obsolete.

 

[SubGothius]

Originally posted by SiliconAddict
PS- Please tell me the irony isn’t lost on the fact that Apple and IBM now have such an interesting relationship. Wasn’t Jobs insanely anti-IBM at one point?

I think that was only back in the early days of Apple and the Mac, before Jobs got the unceremonious boot and went to start his NeXT big thing. Apple's been significantly cozier with IBM since the PowerPC era -- the PPC 601s ("G1s"? ) used in the earliest PowerMacs were made by IBM, which I found amusing to no end whenever I "lifted the hood" on my old 7200/75 and saw Big Blue's logo glaring back at me.

FWIW, I recall hearing that after NeXT stopped producing hardware and the NeXTStep OS became the Intel-compatible OpenStep, Steve turned to an IBM ThinkPad running OpenStep as his regular rig -- a combo he continued to use (to the chagrin of many) for quite some time after he returned to Apple, until PowerBooks with a mature build of Mac OS X were available for him to use instead.

BTW, if by some bizarre twist of fate I ever found myself in possession of a WinTel laptop, be assured that I'd nuke'n'pave it pronto with something NeXTStep-ish, like GNUStep on Darwin, or even LinuxSTEP...

 

[inkswamp]

Originally posted by 3.1416
As for Apple not comparing the G4 to the G5 directly, that wouldn't be a good idea no matter what the results.

Right, and besides, MacWorld and MacAddict routinely do these kinds of new Mac vs. old Mac speed tests anyway.

 

[MisterMe]

Originally posted by SubGothius
I think that was only back in the early days of Apple and the Mac, before Jobs got the unceremonious boot and went to start his NeXT big thing. Apple's been significantly cozier with IBM since the PowerPC era -- the PPC 601s ("G1s"? ) used in the earliest PowerMacs were made by IBM, which I found amusing to no end whenever I "lifted the hood" on my old 7200/75 and saw Big Blue's logo glaring back at me.

FWIW, I recall hearing that after NeXT stopped producing hardware and the NeXTStep OS became the Intel-compatible OpenStep, Steve turned to an IBM ThinkPad running OpenStep as his regular rig -- a combo he continued to use (to the chagrin of many) for quite some time after he returned to Apple, until PowerBooks with a mature build of Mac OS X were available for him to use instead.

....

Actually, Steve Jobs's relationship with IBM began several years earlier than you recall. Leading up to the System 7 era, Apple and M$ entered into a partnership based on TrueType. By then, Jobs had been ousted from Apple and had founded NeXT. Following the Apple/M$ partnership, IBM licensed NeXTSTEP.

 

[fpnc]

Originally posted by 3.1416
It will probably depend on how much memory bandwidth the code requires. If it doesn't require much (like RC5 which fits entirely in cache) then the G5 won't be better than the G4 on a per-cycle basis. But for stuff like video processing that streams loads of data, the G5's vastly superior memory architecture should provide much better performance.

That's true, the faster bus and memory on the G5 will help, but I'm wondering how good the G5's Altivec performance will be compared to the G4. From what I've read the Altivec units on the G5 may offer somewhat less performance per cycle than the G4. So it could be that a 1.4GHz G4 will perform some Altivec-enhanced tasks just as fast or faster than a higher speed G5. The G5 should be much faster in integer and floating point operations (and thus will look good in comparison to PCs), but for Altivec there may not be a big jump in performance.