i've just been curious about the 4 vs. 8 MB cache thing, because i don't know about a possible new processor-line and all woodcrests (aka xeon core 2 duo) have 4 MB shared cache, whereas the clovertons (aka xeon core 2 quatro) have 2 x 4 MB cache (assumably each 2 cores sharing one cache)
generally i'm not very happy with the lack of precise specs on the apple website (chipsets, bus-speed, memory timing, ect), too often one had to rely on third party information or find it out the hard way by oneself ...
as an example it is not at all clear to me which chipset would improve performance if 4 banks of same memory is used opposed to *normal* paired sticks and i can only assume it is the intel 5000X ...
if this is the case than i don't understamd why we don't have the option to choose between PCIe 16 lanes and PCI-X 100/133 MHz as other servers offer - selecting a soundcard is currently somehow painful (either for the quality or for the pocket)
christian

... but to step in to the original topic: the intel macs are significantly more performant than the G5s, although i'm convinced one could get even more out of it using some hand-tooled compiler-options (and loosing the universal binary thingie)
there are not such detailed performance reports as for earlier macs, because i'm not willing to purchase some *newest mac* all three months and we don't have our hands long enough on certain models to check out everything in detail. also the machines are so d***d fast that one had to play around a while with various harddrives and soundcards for a while to locate the current bottlenecks.
christian

@Another User said:

as an example it is not at all clear to me which chipset would improve performance if 4 banks of same memory is used opposed to *normal* paired sticks and i can only assume it is the intel 5000X ...
if this is the case than i don't understamd why we don't have the option to choose between PCIe 16 lanes and PCI-X 100/133 MHz as other servers offer - selecting a soundcard is currently somehow painful (either for the quality or for the pocket)
christian

From what I can determine, the 5000X would be that chipset-- only because it was after the release of the 5000X that the recommendation for RAM sticks in 4s emerged. Even there, some people were under the impression that this was to the exclusion of installing sticks in pairs, but this now seems to be untrue.

Even so, there are no specs that I know of that show any benchmarks to illustrate the performance benefits. One would *think* that any performance benefits such as this would be a serious credential for the sake of selling a better Mac.

And yet, I found it most interesting that the Apple rep continued to refer to the difference between the 2.66 and the 3.0 as a 4Mhz bump in CPU-- where it's only 3.4Mhz. It may be a small exaggeration, but even with the law of averages, that number would be rounded down to 3Mhz and not up to 4Mhz. In the context of what this difference might benefit VI Cube, I've yet to get the sense that it would matter very much at all for the difference in price.

But, I hope we can get to the bottom of these issues with continued discussion by sharing out knowledge, gathering a diversity of information, and filtering out the facts from the fiction.

These are good machines-- and I am encouraged by Sonrise's expression of a lack of regret for having a 3.0. It would be nice to compare the 2.66 with the 3.0 to further determine the differences in performance.

I, too, have been happy with Apple overall. I'm just not pleased with some of the confusing and cryptic data they share these days

I was trying to get fancy and post an image of my "about this Mac", to show the specs re backside cache. Then I realized I didn't know how to include an image in the post - can anyone explain?

Regardless, it lists:

Machine Name: Mac Pro
Machine Model: MacPro1,1
Processor Name: Dual-Core Intel Xeon
Processor Speed: 3 GHz
Number Of Processors: 2
Total Number Of Cores: 4
L2 Cache (per processor): 4 MB
Memory: 9 GB
Bus Speed: 1.33 GHz
Boot ROM Version: MP11.005C.B04
SMC Version: 1.7f8

For what it's worth, there it is.

@plurye said:

For what it's worth, there it is.

...and it's worth a great deal, plurye!! Thanks so much.

I just got similar specs from another friend who has a 2.66:

Machine Name: Mac Pro
Machine Model: MacPro1,1
Processor Name: Dual-Core Intel Xeon
Processor Speed: 2.66 GHz
Number Of Processors: 2
Total Number Of Cores: 4
L2 Cache (per processor): 4 MB
Memory: 2 GB
Bus Speed: 1.33 GHz
Boot ROM Version: MP11.005C.B04
SMC Version: 1.7f8

Another engineer/friend of mine who JUST got his 3G two days ago is now returning it. He bought it on the premise that the 3G's L2 cache was the other major difference from the 2.66 based upon printed material from his rep. This is turning out to be untrue since it seems ALL of the MacPro's have the same L2 cache size... which is 2 x 4 MB = 8 MB.

So far, there is no other difference between the 2.66 and the 3.0 except for the reported 3.4MB bump in CPU clock.

Thanks again! This feels like a very healthy discussion already!

@plurye said:

I was trying to get fancy and post an image of my "about this Mac", to show the specs re backside cache. Then I realized I didn't know how to include an image in the post - can anyone explain?

The image would already have to be posted elsewhere on the internet.

Given that, open the image in a separate browser window and copy the address.

In the forum message composition window, use the IMG tags-- starting with ''

For example, quote this very message and study the code used to post this image...

http://www.sweetwater.com/images/items/VISymphCube.jpg">

@JWL said:

So far, there is no other difference between the 2.66 and the 3.0 except for the reported 3.4MB bump in CPU clock.

I think the difference is actually 340MHz, or about 13%.

Steven

@JWL said:

For example, quote this very message and study the code used to post this image...

Thanks - I've copied the message to my computer for future reference. And while we're at it I'm attempting to quote some text from your message, which is something else I've been unable to do before now!

PL

@JWL said:

So far, there is no other difference between the 2.66 and the 3.0 except for the reported 3.4MB bump in CPU clock.

I think the difference is actually 340MHz, or about 13%.

Steven

Ooops-- what was it I said about my being a moron? [:P]

Mhz, not MB to start-- sorry.

Darn decimals: .34Ghz w/ factor of 1k = 340Mhz

Thanks for the correction!

Okee-dokee...

Is the difference of 13% on the 3G enough to matter compared to the 2.66 where loading VI Cube instances is concerned (given 4GB of RAM or more are installed on both machines)?

JWL, thanks for confirming the 5000X - then it's clear ... this is from the documentation of a board with a 5000P
http://vsl.co.at/upload/users/449/dualchanneldualbranchmemory.png">
i don't think the difference between 2.66 and 3.0 will be significant, even noticeable - but there are always configurations needing each single quantum of CPU power(eg. altiverb)

the quadcores are at 2.66 currently (producing a lot of heat) - maybe apple waits for the quadcore low voltage processors reaching 3 GHz or more ... to make an argument why purchasing a dual quad looking better
christian

CM --

OK, now I need clarification! [:)]

If you glance up higher in this thread, you'll see the specs for my machine.

1. My understanding it that the RAM limit is 16 GB for the dual 3.0 xeon Mac pro. Does the 32 GB possibility mentioned in your last post exist for my machine, or is that a later generation of Mac Pro?

2. After doing some research, I and the person who helped me acquire the Mac concluded that I was OK leaving my pair of 512k chips in my computer, as long as we got everything in the recommended slot order. Is that info incorrect? Should I take 'em out, leaving just the two 4 GB chips?

I'm not an expert in this area, as you can tell!

Thanks,
PL

ad 1) at least the 5000P _supports_ 16 slots (grouped in 2 branches each 2 channels each 2 banks), this doesn't say the board _has_ 16 slots (actually it has 8, as the macPro has). i've not heared about 4 GB mamory sticks so far, so the limit will be 16 GB ...
ad 2) that's hard to say and would need some in-depth reading of the chipset specs ... does throughput decrease if a second branch or channel is equipped with sticks of different size? i really don't know and only some nifty RAM benchmark test would give us numbers ...

but 8 GB seem to be enough - if you're really interested i'd try to take them out for testing and *feel* the difference - make sure you don't have any static electricity (get your wrist grounded) when touching the sticks and keep them in an antistatic bag in case you do so.
christian

Thanks - I figured 32 GB referred to a theoretical upper limit of the board, not the current Mac pro.

But what was the source of the info you pasted into your message, with the underlined warning not to mix sizes and speeds of RAM?

the source was the manual of a server motherboard ... as you can see even in the last line 8 slots stay empty. but wait - re-reading it by my own i see the underlined note: do not mix ..., so there might be a reason for it ...

Hey, CM-- thanks for the additional info. It is as revealing as it is confusing!! At least we're getting somewhere.

Of course this site from Macworld has some interesting info:

http://www.macworld.com/2006/08/firstlooks/macprofaq/index.php

Excerpt regarding L2 and system busses:

But no matter how you configure a Mac Pro, there are still a few constants: every model includes 4MB of shared L2 cache per processor and two 1.33GHz system buses, one per processor.

Excerpt comparing PCIx to PCIe:

In the Power Mac, each PCI Express slot had a set bandwidth, expressed in terms of lanes—the graphics slots was the fastest at 16x, with one 8x slot and two 4x slots as well. With the Mac Pro, Apple says that when you boot the computer after installing a new PCI Express card, the OS will let you choose the amount of bandwidth to dedicate to that slot. Apple told us that the total amount of bandwidth available to the PCI Express bus is less on the Mac Pro than on the Power Mac, but said the ability to direct that bandwidth as needed should make up for such a shortcoming. They also told us that there's more power (total wattage) available to the PCI Express bus, letting you power two Nvidia Quadro FX 4500 graphics cards.

As for the underlined "do not mix...", that was the first example of "do not" I've ever seen. It is clear that this is akin to the PPCs where the 8-slot riser card was introduced. Installing in matched pairs was made quite clear at that time, and it has certainly carried over.

Installing in 4s remains a quandary-- I've yet to find a "do not mix.." in terms of sticks in 4s... new (?) with 5000X. I am clear on the rule of using matched pairs of RAM sticks still applies with all these models.

While the 13% increase is not especially earth shattering, I still wonder about it, given that we spend SO much time squeezing out every micro-ounce of each computer's resources to accommodate VI Cube. That this 13% is focused only on the CPU, it remains to be determined how this increase benefits all the other fixed parameters involved with shuttling a demanding VI.

Right now, I'm looking at a 2.66 and a MacMini for about the same price as a 3G...

And yes--- the 32GB statement was a brow-raiser!!! [[[:|]]] [[[:|]]]

Thanks again!
JWL

One other observation-- correction on my part...

The developer note says "you SHOULD not" which is rather different in meaning than "DO not".

I don't know if Apple is trying to be polite by using the passive voice so often, but it is becoming an increasing source of confusion...

hehe ... if somewhere it reads *you should not* then i'm always taking it as a *do not* if i want to squeeze as much as possible out of *it* ...

PCI(X vs. e) total bandwidth ... total bandwidth of _what_ ... the bandwidth delivered to all slots simultaneously or the plain addition of bandwidth per slot times the number of slots ... i doubt the chipset of the PPC delivers more bandwidth than the chipset of the macPro ... adding simply up slots (eg. 64 bit x 133 MHz x 4 slots) is nonsense ... we all know that formerly gigibit network cards couldn't deliver full bandwidth (either because of sitting in a 32-bit slot or sharing the bus with other devices) because the PCI-bus itself has been the bottleneck ... if someone gives us figures he should also say on what such figures are based on or related to ...

in any case this keeps a theoretical discussion, because we don't have the choice to select between options.
christian

@cm said:

hehe ... if somewhere it reads *you should not* then i'm always taking it as a *do not* if i want to squeeze as much as possible out of *it* ...

PCI(X vs. e) total bandwidth ... total bandwidth of _what_ ... the bandwidth delivered to all slots simultaneously or the plain addition of bandwidth per slot times the number of slots ... i doubt the chipset of the PPC delivers more bandwidth than the chipset of the macPro ... adding simply up slots (eg. 64 bit x 133 MHz x 4 slots) is nonsense ... we all know that formerly gigibit network cards couldn't deliver full bandwidth (either because of sitting in a 32-bit slot or sharing the bus with other devices) because the PCI-bus itself has been the bottleneck ... if someone gives us figures he should also say on what such figures are based on or related to ...

in any case this keeps a theoretical discussion, because we don't have the choice to select between options.
christian

As always, cm, you nailed it. But this is typical of Apple. When reading fixes made with 10.4.9, their so-called "details" would always say fixed a bug with _x_ parameter, or better functionality with _y_--- but better than what? Fixed a bug is fine-- but what was the bug?

If we keep peeling back the layers of this onion, we'll get to the bottom of things sooner or later.

i'd say it is not as much apple, although they tend to make their descriptions as simple as possibel, but hyper-enthusiastic "writers" (please note the quotes) of certain magacines or online-publications who pick up single statements not understanding about what they are talking about. unfortunately this also applies to sales people too often ...

the machines itself are great, so i do not understand why such attampts for glrification are even needed ...

but to return to a real issue ... what soundcards are you considering to use? i have not been too lucky on searching for PCIe interfaces. roumors say the quadcore will offer a PCI-X slot again, because it seems there are not too much PCIe options on the market ...
christian

Maybe this was the slot change that broke the camel's back and the world is rebelling.

I personally will think *very* hard before I ever buy another expensive PCI card. It's just too much to keep up with. FireWire interfaces for me from now on.