DDR with current chip = no more preformance !

[quote]Originally posted by Appleworm:
<strong><a href="http://www.barefeats.com/xserve2.html" target="_blank">xserve vs dual pm g4</a>

Why the Xserve is not at least 5-10% faster than the PM ?</strong><hr></blockquote>

like a lot of people said the extra DDR bandwidth is used by other server hardware (network card, storage,..), not the CPU.

I think apps that take advantage of QE and Jaguar will fair a bit better on the X-serve. As will anything that requires better disk performance (once you start adding second and third drives)

DDR the Xserve-way is not the same way as DDR needs to be implemented for the processor(s) to take advantage of the extra bandwidth.
DDR the 'True' way, which means a DDR FSB giving the CPU('s) 2,1 GBps bandwidth instead of 1 GBps (Xserve, powermacs) will be '(much) more performance!'.

[quote]Originally posted by Appleworm:
Why the Xserve is not at least 5-10% faster than the PM ?
Aw
<hr></blockquote>

Because the G4 is a 3 year old chip. A P3 wouldnt either benefit from DDR. They were both designed before DDR was invented and no one would have thought that the G4 would be in use after 3 years.

The G4 was designed for PC100 and not PC400 DDR, the performance benefits for an old chip is negligible.

The Athlon XP doesnt benefit from 333 MHz or 400 MHz DDR, why would an even older chip benefit from it?

Heck the Athlon XP dont even benefit from a 333 MHz FSB!

The Athlon was designed for 266 MHz DDR and a 266 MHz FSB - it dont get much faster when coupled with faster Ram or a faster FSB.

[quote]Originally posted by -@-:
<strong>

Because the G4 is a 3 year old chip. A P3 wouldn’t either benefit from DDR. They were both designed before DDR was invented and no one would have thought that the G4 would be in use after 3 years.

The G4 was designed for PC100 and not PC400 DDR, the performance benefits for an old chip is negligible.

</strong><hr></blockquote>

FALSE.
Amlost every altivec-enhanced algorithm is bandwith starved, even on a G4/500MHz.

[ 08-12-2002: Message edited by: Franck ]</p>

[quote]Originally posted by Franck:
<strong>

FALSE.
Amlost every altivec-enhanced algorithm is bandwith starved, even on a G4/500MHz.

[ 08-12-2002: Message edited by: Franck ]</strong><hr></blockquote>

What he's really saying is that the XServe's memory-to-CPU bandwidth isn't higher because the G4 itself isn't designed to be hooked up to buses quite that fast. Which is true.

What you are saying is also true though: if we could only get instructions/data past the bottleneck at the front door of the chip, the AltiVec unit could easily consume 10x as much data.

barefeats = clueless. barefeces would be more appropriate.

Besides, this topic in general (how much performance will DDR get us) has been beaten to death. There's no way to answer the question unless you know all the other specifics of the system architecture, and unless you know what kind of application is being run.

Okay, let's get on the same page here. Are we saying that Barefeat's benchmarks are flawed in this case? If so, then we can stop talking and move on. However, if we are to believe what Barefeats has given us, then this truly is bad news.

How can anyone say the tests aren't applicable because one is a server and one is a desktop? If they're using the same components, except one is using DDR-RAM, then why not compare their performance? Early Xserve benchmarks said it was much faster than the Dual 1GHz in certain cases. But if it truly isn't faster, or if it's only faster in cases that wouldn't benefit desktop users, then we need to take that seriously.

Apple most likely will be delivering Power Macs with the Xserve DDR hack. (I don't believe there is a new G4 was a new FSB, but I would like to be surprised.) If all we get is the G4 7450 with a higher clock speed and the DDR hack, then we should take these benchmarks seriously. If you don't believe the tests were valid, then please state why; if not please offer up a reason why extrapolating future Power Mac performance from these tests is invalid.

Finally, the G4 chip line may itself be three years old, but the chip we're using isn't. There was that terrible speed dump that Motorola addressed by lengthening the pipeline to seven stages, remember? If Motorola were competent or a good partner to Apple, then it would have designed the new G4 to be DDR compliant. It didn't and now we're praying that a new, unannounced version of the G4 may support DDR. Don't give Motorola an out by stating the G4 is three years old; the one being used right now is much newer.

[quote]Originally posted by Big Mac:
<strong>Okay, let's get on the same page here. Are we saying that Barefeat's benchmarks are flawed in this case? If so, then we can stop talking and move on. However, if we are to believe what Barefeats has given us, then this truly is bad news.

How can anyone say the tests aren't applicable because one is a server and one is a desktop? If they're using the same components, except one is using DDR-RAM, then why not compare their performance? Early Xserve benchmarks said it was much faster than the Dual 1GHz in certain cases. But if it truly isn't faster, or if it's only faster in cases that wouldn't benefit desktop users, then we need to take that seriously.

Apple most likely will be delivering Power Macs with the Xserve DDR hack. (I don't believe there is a new G4 was a new FSB, but I would like to be surprised.) If all we get is the G4 7450 with a higher clock speed and the DDR hack, then we should take these benchmarks seriously. If you don't believe the tests were valid, then please state why; if not please offer up a reason why extrapolating future Power Mac performance from these tests is invalid.

Finally, the G4 chip line may itself be three years old, but the chip we're using isn't. There was that terrible speed dump that Motorola addressed by lengthening the pipeline to seven stages, remember? If Motorola were competent or a good partner to Apple, then it would have designed the new G4 to be DDR compliant. It didn't and now we're praying that a new, unannounced version of the G4 may support DDR. Don't give Motorola an out by stating the G4 is three years old; the one being used right now is much newer.</strong><hr></blockquote>

I don't know the testing methodology, but those are the processor performance results that I would expect (more or less). MPX is maxed out with Apple's latest SDRAM implementation, so DDR isn't going to help anything that is completely dependent on the processor's memory bandwidth. I believe the reason most people are belittling the site's comments is that this has been known for a couple of months and yet the site seemed to expect some kind of a processor performance increase.

"barefeats = clueless. barefeces would be more appropriate."

I dont really care about the xserve,DDR is no great shaks anyway, but I would have to agree with this assesment of barefeats.I know they have some obviously wrong benchmark data on their site like in their photoshop test the 800 pm was slower than a 733 and even slower than the 667 powerbook (the one with no L3 cache).The 933 was slower than the 867 too.This was on their site for months and is probably still there,it was last week!Home brew benchmarks are dangerous as not one in 1000 seems to know what they are doing or how to judge the results.