Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Dual 2.0GHz G5 PowerMac Pictures and Benchmarks
- Thread starter MacRumors
- Start date
- Sort by reaction score
Analog Kid
macrumors G3
AIFF from hard drive
Just making sure people noticed the iTunes rip is from an AIFF file on his hard drive.
Doesn't invalidate the benchmark, and may even be the better way to do it, but don't compare to ripping CD's...
Just making sure people noticed the iTunes rip is from an AIFF file on his hard drive.
Doesn't invalidate the benchmark, and may even be the better way to do it, but don't compare to ripping CD's...
Check out barefeats, too
Barefeats.com has a couple of G5 benchmarks as well. Check out his comparison of the G5s with Xeons and Athlons, and his Photoshop/Cinema 4D/Bryce comparison...
Barefeats.com has a couple of G5 benchmarks as well. Check out his comparison of the G5s with Xeons and Athlons, and his Photoshop/Cinema 4D/Bryce comparison...
Re: AIFF from hard drive
Good point - the CD only reads at 32x, so putting it on the disk eliminates the CD drive bottleneck. But that is probably a better test, since it is testing the G5 and not the drive.
Good point - the CD only reads at 32x, so putting it on the disk eliminates the CD drive bottleneck. But that is probably a better test, since it is testing the G5 and not the drive.
SeaFox
macrumors 68030
MacBandit
macrumors 604
Re: Here they are
Try using more jpeg compression and less scaling.
If you don't have a program to do this in I suggest Graphic Converter.
Try using more jpeg compression and less scaling.
If you don't have a program to do this in I suggest Graphic Converter.
Developer Support in a Meeting
This is kind of interesting. I placed a call in to ADC Support (800 793-9378 option 2) to see if anything has changed with my ADC order.
Instead of getting a person or an "everyone is busy" message, I got a recording telling me that all of the ADC Support staff was in a meeting that started at 11:30 central time and would last through the rest of the business day. I wonder what they're talking about?
This is kind of interesting. I placed a call in to ADC Support (800 793-9378 option 2) to see if anything has changed with my ADC order.
Instead of getting a person or an "everyone is busy" message, I got a recording telling me that all of the ADC Support staff was in a meeting that started at 11:30 central time and would last through the rest of the business day. I wonder what they're talking about?
The fact that Cinebench gets about 400 when a dual Xeon gets well over 500 is very disappointing.
So from this, I would expect Lightwave and Maya to still be a tad slower on the Mac.
And I thought the G5 was going to be a lot faster. Now why would I think that?
So from this, I would expect Lightwave and Maya to still be a tad slower on the Mac.
And I thought the G5 was going to be a lot faster. Now why would I think that?
Because once Cinebench and the other benchmark programs take the G5 into account, then the G5 just might be faster.
This is all similar to what happened when the PowerPC was originally released. Old 680x0 code ran really slow (until the emulator became more efficient), and eventually developers released "Fat" programs. The release of the G5 is similar, except there is not emulation mode for 32-bit PowerPC code.
Because once Cinebench and the other benchmark programs take the G5 into account, then the G5 just might be faster.
This is all similar to what happened when the PowerPC was originally released. Old 680x0 code ran really slow (until the emulator became more efficient), and eventually developers released "Fat" programs. The release of the G5 is similar, except there is not emulation mode for 32-bit PowerPC code.
No, it's nothing like the move to PPC. The PowerPC was a completely different chip architecture, and software ran in emulation that made it think it was running on 680XX. Just like Windows emulators are slow, this emulation was slow, but not as noticeable because the first PPC processor was so much faster than the 680XX processor. So when software became PPC native, it wasn't operating under the emulation, so it was much faster.
The IBM 970 is a native PPC processor. There is no emulation here.
Software can be optimized for the G5's differences, but we won't see a speed increase on the level of 35-40%, which is what we would need to see in Cinebench for it to overtake the dual Xeon.
This is all similar to what happened when the PowerPC was originally released. Old 680x0 code ran really slow (until the emulator became more efficient), and eventually developers released "Fat" programs. The release of the G5 is similar, except there is not emulation mode for 32-bit PowerPC code.
No, it's nothing like the move to PPC. The PowerPC was a completely different chip architecture, and software ran in emulation that made it think it was running on 680XX. Just like Windows emulators are slow, this emulation was slow, but not as noticeable because the first PPC processor was so much faster than the 680XX processor. So when software became PPC native, it wasn't operating under the emulation, so it was much faster.
The IBM 970 is a native PPC processor. There is no emulation here.
Software can be optimized for the G5's differences, but we won't see a speed increase on the level of 35-40%, which is what we would need to see in Cinebench for it to overtake the dual Xeon.
MacBandit
macrumors 604
You're correct about the emulation but I don't believe so about the speed increase. The PPC970 has several advances over the current G4 or G3 line that are not being used by current apps and compilers. Once those are taken into account we could very well se a 25-50% increase in performance depending on the task.
Compiler, yes, optimizations, no. The developer of Cinema even states that with the optimizations it might reach 500. That's a quote. Dual Xeons are over 500.
With compilers, the question is which one do they use, and can they use the new one from IBM.
One thing, though, is I'm seeing results from Panther that are very impressive. So a dual G5 with applications that are heavily optimized running on Panther might be impressively fast. The problem is a lot of developers don't spend the time necessary to make a Mac version very well optimized. After Effects is a famous example from a developer that is supposedly one of the Mac's strongest.
I was under the impression that all software would be very, very fast on a G5 without all this "just wait for optimizations" talk. Apple's hype certainly implied that. So we are like we were with the G4 - a slower processor, but potentially faster with the right software optimizations.
With compilers, the question is which one do they use, and can they use the new one from IBM.
One thing, though, is I'm seeing results from Panther that are very impressive. So a dual G5 with applications that are heavily optimized running on Panther might be impressively fast. The problem is a lot of developers don't spend the time necessary to make a Mac version very well optimized. After Effects is a famous example from a developer that is supposedly one of the Mac's strongest.
I was under the impression that all software would be very, very fast on a G5 without all this "just wait for optimizations" talk. Apple's hype certainly implied that. So we are like we were with the G4 - a slower processor, but potentially faster with the right software optimizations.
The 1.8 has reached the Apple Dealer in Coff's Harbour I have seen it, got static off the alu case and looked inside which is a work of art. The Australian Apple site is still telling my dealer next week for the duals. I'm not holding my breath.
I've got my 23" cinema display frustratingly sitting in its box as even with a Mac Radeon card and the loan of a DVI to ADC, it won't work on my G4'd Beige Tower.
Yes it probably will, but Panther won't help benchmarks like this Cinebench stuff. Which once again illustrates the general uselessness of benchmarks in determining real-world performance and useability. Running multiple Apps and making heavy concurrent use of OS/GUI functionality will be where Panther/G5 will shine, I think.
That's true, and I am also somewhat disappointed with that. Although it is true that any new chip needs different instruction scheduling to make the most of it (even with the same native instruction set, and aside from the whole 32-64 question), I agree with you that the G5's raw CPU performance is probably still going to be generally a bit short of Intel--even with G5-optimized code. Still, it's nice that we are finally back in the same ballpark.
Not really. The G4 has other serious problems, like an absurdly slow memory bus. The G5 gets us away from that, and I think that is really one of the most important advantages of this chip for the Mac. The other major advantage is the seamlessness of the 64-bit transition, which needed to be done sooner or later as we approach the 4GB wall. So much the better that it is sooner, when most people haven't quite run into that wall yet. Code may be need to be recompiled for optimum scheduling, but we are not in a "32-bit emulation" situation.
itunes conversion
itunes conversion on my 1.6 G5 from the hard drive is between 20 and 21. This is for an AAC file at 160. Pretty good speed, beats my dual 867 by about 5X to 6X.
itunes conversion on my 1.6 G5 from the hard drive is between 20 and 21. This is for an AAC file at 160. Pretty good speed, beats my dual 867 by about 5X to 6X.
Actually Panther will help a great deal with apps like this. Remember that Panther has a new optimized Math Lib and completely rewritten memory management which is significantly faster.
Your wrong. Steve's WWDC keynote was very clear that with a simple and quick recompile Apps will do much better. Without a re-compile they will still do very well.
As far as performance comparisons, you're being ridiculous. Th 970's architecture and design outside of Altivec is radically different than the G4. The PPC architecture allows for binary compatibility though the implementations can be completely different, that is the beauty of the PPC architecture. By properly taking advantage the 970's internal parallelism and increased functional units, as well as its better prediction, Apps can be optimized significantly more than 30-40% as the Cinebench programmers stated. And that was before IBM's new xlC compiler which in some cases is showing over a 200% increase without any major changes.
To prejudge the 970's continued performance increases due to better software is premature. Why, OS X on the G4 is going to see another 30-40% speed increase with Panther. In fact, Apple has sped up some text functions by as much as 10x (1000%) with Panther on the G4.
