diglloyd's 2.93GHz 'Nehalem' Mac Pro Review!

[jjahshik32]

I'll bet you $500 you'll see pretty much the same memory throughput speeds under Linux.
OS X is not some magical piece of software.
It simply goes as fast as it can on the underlying hardware.



Funny how none of those are specific to OS X, eh?

Uh then wth is grand central all about? Does linux have something similar to grand central, nope.

SL is the first of its kind, period.

 

[jjahshik32]

Since Snow Leopard isn't supposed to support PPC machines anymore, aren't the applications etc. installed with SL stripped of PPC code thus making them leaner and the total install smaller?

No, actually SL still supports PPC.

 

[rylin]

How is Leopard bloated?? My clean install is only 7.1GB.

It's bloated in terms of resource usage.
Feel free to read this or watch the SL introduction clip.

 

[jjahshik32]

It's bloated in terms of resource usage.
Feel free to read this or watch the SL introduction clip.

Those clips are missing and those are just vague little descriptions. Also is that guy running snow leopard on a nehalem? I dont think so. He's still running the beta on the older 08 models, in which to him doesnt feel that much different because the Nehalem will benefit the most.

 

[rylin]

Uh then wth is grand central all about? Does linux have something similar to grand central, nope.

SL is the first of its kind, period.

Grand Central is not unique to Nehalem.
It's (as far as anyone can tell) a new scheduler, alongside an API that let's developers give the scheduler hints on priority and "weight" of the workload, to better be able to schedule unrelated things on different cores.

Grand Central has nothing to do with memory throughput.

 

[jjahshik32]

Grand Central is not unique to Nehalem.
It's (as far as anyone can tell) a new scheduler, alongside an API that let's developers give the scheduler hints on priority and "weight" of the workload, to better be able to schedule unrelated things on different cores.

Grand Central has nothing to do with memory throughput.

I didnt say Grand Central was unique to Nehalem. You guys need to stop twisting my words around for your own benefit. Grand central will be there regardless of which machines you are using.

The real question is how well certain hardwares will benefit from Grand Central. And the Nehalems will benefit the most.

Good lord just take account to the 2x faster memory speeds, QPI, no more slow FSB to dip into, 40% less latency, faster RAM 1033mhz, 2x hyperthreading in each core as a whole and you will have a much faster result with SL and grand central.

 

[rylin]

Good lord just take account to the 2x faster memory speeds, QPI, no more slow FSB to dip into, 40% less latency, faster RAM 1033mhz, 2x hyperthreading in each core as a whole and you will have a much faster result with SL and grand central.

This has:
a) Nothing to do with Snow Leopard.
b) Nothing to do with OS X.
c) all the hallmarks of someone talking about stuff he doesn't understand.

Allow me to reiterate:
From everything we know so far, Grand Central has nothing to do with memory, and everything to do with the scheduler.

Nehalem does not have "2x hyper threading".
What it does have is hyperthreading, effectively running two threads at a time, per core, with a performance impact per thread.

 

[Horst]

Ok then it'll be released in june and get ready to sell your previous gen 8 core when you see the monster results of the nehalem.

And dont forget, I told you so.

Funny you don't see any I told you so remarks re. the Nehalem , which was supposed to shake our world.

Bottom line for me, the new MPs are great computers, but if you are using an earlier Mac Pro right now , there is no reason to switch - they are excellent machines as well.

As said above, no OS update will magically turn any hardware into something complitely different.
Unless you rememver how OSX rendered the G3s useless.

Here I hope OSX is loosing that belly it's been schlepping around since day one, and gets mean and lean again - no widgets, a lag-free Finder, Spotlight rewritten, a usable search function, Dashboard, iSync and TM strictly optional, etc etc .

Multi-core support made easier for developers is great, just don't expect results very soon.

Monster results for iLife apps might be in a bit sooner - but who the hell cares ?

 

[jjahshik32]

This has:
a) Nothing to do with Snow Leopard.
b) Nothing to do with OS X.
c) all the hallmarks of someone talking about stuff he doesn't understand.

Allow me to reiterate:
From everything we know so far, Grand Central has nothing to do with memory, and everything to do with the scheduler.

Nehalem does not have "2x hyper threading".
What it does have is hyperthreading, effectively running two threads at a time, per core, with a performance impact per thread.

2x meaning 2 threads in each core or am I wrong?

Of course grand central will take use of allocating the cores or (scheduler of cores) but if you have can schedule the cores and the hardware runs more efficiently and faster due to QPI and its hyperthread technology and 2 threads in each core, wouldnt you agree that it will run faster?

Especially when memory will be a separate issue running in the background and it being 2x faster in all memory tests even with the current leopard and benchmarks thats been out?

Basically all of you guys are saying with SL the differences between a Nehalem vs. the previous gen 8 core machines will have equal performances, but that will NOT be the case. Of course the Nehalem will be faster on all accounts.

 

[ceres]

I didnt say Grand Central was unique to Nehalem. You guys need to stop twisting my words around for your own benefit. Grand central will be there regardless of which machines you are using.

The real question is how well certain hardwares will benefit from Grand Central. And the Nehalems will benefit the most.

Good lord just take account to the 2x faster memory speeds, QPI, no more slow FSB to dip into, 40% less latency, faster RAM 1033mhz, 2x hyperthreading in each core as a whole and you will have a much faster result with SL and grand central.

....which a) a guess at this time
and b)
I´d reckon that Apps running on the same Nehalem MP in SL will not outperform their Win7 counterparts on the same MP at all.

 

[VirtualRain]

This has:
a) Nothing to do with Snow Leopard.
b) Nothing to do with OS X.
c) all the hallmarks of someone talking about stuff he doesn't understand.

Allow me to reiterate:
From everything we know so far, Grand Central has nothing to do with memory, and everything to do with the scheduler.

Nehalem does not have "2x hyper threading".
What it does have is hyperthreading, effectively running two threads at a time, per core, with a performance impact per thread.

QFT!!!

@jjahshik32 you really are mis-informed and need to read what others are saying here, digest it and try to understand the relationship and roles played by the CPU cores (executing code), the OS schedular (scheduling tasks on cores to execute), and the application (generating the tasks).

All three components need to be optimized for parallel multi-threaded operation, but the failure of any one (e.g. the application) will mean that the whole thign runs inefficiently. Similarly, massive improvements in any one or two will not yield significant improvements... all three must be optimized for any improvements to be realized.

Intel has created a CPU that is optimized for multi-threaded execution (although the last generation is no slouch). Apple is creating an OS that will be optimized for multi-threaded execution (although the current OS X is not bad as it is). And some, but not all, applications are already optimized for multi-threaded execution.

There is no opportunity here for significant gains over-night. Subtle improvements... yes. Drastic improvements... no! (unless you are talking about OpenCL where there is opportunity for orders of magnitude improvements on some tasks, but then again, all three elements, GPU, OS, and application need to be optimized for it).

 

[jjahshik32]

QFT!!!

@jjahshik32 you really are mis-informed and need to read what others are saying here, digest it and try to understand the relationship and roles played by the CPU cores (executing code), the OS schedular (scheduling tasks on cores to execute), and the application (generating the tasks).

All three components need to be optimized for parallel multi-threaded operation, but the failure of any one (e.g. the application) will mean that the whole thign runs inefficiently. Similarly, massive improvements in any one or two will not yield significant improvements... all three must be optimized for any improvements to be realized.

Intel has created a CPU that is optimized for multi-threaded execution (although the last generation is no slouch). Apple is creating an OS that will be optimized for multi-threaded execution (although the current OS X is not bad as it is). And some, but not all, applications are already optimized for multi-threaded execution.

There is no opportunity here for significant gains over-night. Subtle improvements... yes. Drastic improvements... no! (unless you are talking about OpenCL where there is opportunity for orders of magnitude improvements on some tasks, but then again, all three elements, GPU, OS, and application need to be optimized for it).

You know what, I dont want to argue anymore because whatever I say or whatever you say wont matter to each of us.

Well just have to see when snow leopard is released and we'll have to come back here or on a new thread that will decide which hardware benefitted the most from SL.

Oh well I'm tired and its already 2pm, its about time for me to go to the Apple Store to pick up the precious.

 

[rylin]

2x meaning 2 threads in each core or am I wrong?

Of course grand central will take use of allocating the cores or (scheduler of cores) but if you have can schedule the cores and the hardware runs more efficiently and faster due to QPI and its hyperthread technology and 2 threads in each core, wouldnt you agree that it will run faster?

Especially when memory will be a separate issue running in the background and it being 2x faster in all memory tests even with the current leopard and benchmarks thats been out?

Basically all of you guys are saying with SL the differences between a Nehalem vs. the previous gen 8 core machines will have equal performances, but that will NOT be the case. Of course the Nehalem will be faster on all accounts.

Hyperthreading was originally a messy hack to not completely waste Netburst's insanely long pipeline.
As for Nehalem's implementation, while better than the Netburst variation, it's still not traditional SMT (which is to natively run two or more threads per core, with no performance penalty).

What it really boils down to is that under heavily threaded scenarios, you might see a slight performance benefit.
That, however, is also OS independent (as long as the OS has support for "HTT" -- Windows does, OS X does, Linux does), and so doesn't have anything to do with neither OS X nor Grand Central.

 

[jjahshik32]

Hyperthreading was originally a messy hack to not completely waste Netburst's insanely long pipeline.
As for Nehalem's implementation, while better than the Netburst variation, it's still not traditional SMT (which is to natively run two or more threads per core, with no performance penalty).

What it really boils down to is that under heavily threaded scenarios, you might see a slight performance benefit.
That, however, is also OS independent (as long as the OS has support for "HTT" -- Windows does, OS X does, Linux does), and so doesn't have anything to do with neither OS X nor Grand Central.

My whole point is that if Grand Central has nothing to do with performance boost, then whats the whole point in it to begin with??

Isnt the whole outcome of Grand Central is for performance boost due to the efficiency.

 

[rylin]

My whole point is that if Grand Central has nothing to do with performance boost, then whats the whole point in it to begin with??

Isnt the whole outcome of Grand Central is for performance boost due to the efficiency.

Allow me to clarify, as I may have phrased it badly in one of the above posts.

Grand Central is meant to help performance while allowing for more efficient scheduling of resources by virtue of allowing developers to give the scheduler hints on what the current workload is.

A very crude version of this already exists on a per-process basis with "nice" and "setpriority" (open up Terminal.app and type "man nice" for more info).

It's a technology that's hardware independent, and so you'll see pretty much the same relative speedup on a '08 2.8 as a '09 2.26 or 2.93.

In other words (since the speedup is a relative figure), it can't be expected to drastically change the field.
Not only is it hardware independent, but it's largely microcode independent as well, so the extra instructions Nehalem is capable of don't factor into things.

If a developer doesn't know how to hint his/her code properly, the performance boost will be minimal.
Properly hinted, it'll run effectively spread out on several cores, which can give a reasonable performance boost -- but again, it'll be the same speedup for a 08 MP as a 09 MP.

 

[-js-]

Ok then it'll be released in june and get ready to sell your previous gen 8 core when you see the monster results of the nehalem.

And dont forget, I told you so.

I remember when Nehalem architecture was all the up-hub and surprisingly snow leopard was announced. It automatically hit me, oh I see why they are releasing SL due to the future of the architecture that takes better advantages.

This is just borderline insulting. I will NOT "get ready to sell" my MP08 in June when Snow Leopard gets released. I am not someone who needs to have the latest and greatest every year. Right now, I am still using a 733MHz G4 quicksilver, for example. I plan on keeping my new 2.8 octo MP08 for many years to come. I don't care how much faster the 09's become when SL hits. I know it will also speed up my 08 and I know that it's plenty fast enough right now.

Finally, a faster memory architecture doesn't mean DIDLY if you run out of memory. On octad 09 models, this isn't an issue. But on quad 09's, it certainly could be, or could become so. Since my budget won't allow for a 2.66 or 2.93 09 octad, that leaves the 08 models.

Period.

 

[jjahshik32]

I only wish jjahshik would stop parroting the Apple information, and FILLING these forums with the same, redundant information.

40% latency, bla bla bla. Snow Leopard will save us from global warming, bla, bla, bla. Good lord, take a day off already, or turn a corner and start relaying how this data is relevant to workflow, daily usage and computer choice.

many of us are still listening for quality information.

Ah seems like another MP 08 owner. Good to know.

 

[jjahshik32]

I already had the 2.8GHz 8 core mac pro last year sold it off. I'm looking into buying the nehalem very very soon.

 

[rylin]

I already had the 2.8GHz 8 core mac pro last year sold it off. I'm looking into buying the nehalem very very soon.

So?

What relevance does it have that MCHR has a MP 08?
Absolutely none, I tell you.

"Judge not lest ye be judged."

 

[jjahshik32]

So?

What relevance does it have that MCHR has a MP 08?
Absolutely none, I tell you.

"Judge not lest ye be judged."

It has the relevance of all the people that are arguing back and attacking and defending their old hardware when I made the claim that Nehalem will be faster than the older generation especially under SL which is infact true.

 

[nanofrog]

That doesnt make any sense. Since the current Leopard runs about the same compared to the 08 vs. the 09 models that it'll be equivalent on the SL?

Especially when the current Leopard isnt really utilizing much on the current Nehalems features nor architectures? I guess we would have to find out once 10.5.7 is released to see if any differences occur.

In terms of hardware, YES.

The gains SL may be able to produce over Leopard, would be replicated whether it's run on the '08 or '09 models. Both architectures would be able to run it, and the software, assuming it can remotely do what's promised, will be able to take advantage of the cores on both machines.

Any performance gains between the '08 and '09 models, both running SL, will strictly be a function of the hardware. Keep in mind, the biggest difference isn't the cores, but the Integrated Memory Controller + DDR3. (Thinking in terms of the CPU/memory systems, not peripheral devices such as PCIe devices, HDD chips,...).

The advantages SL is supposed to provide would be applicable to both machines. BTW, a great deal of overall performance gains will have to come from applications developed specifically to take advantage of what's provided under SL. Older, existing software likely hasn't yet been written to use the new API's. What SL can provide on its own (no applications involved), is it's supposed to be leaner. Leaner code has a nice little habit of performing faster. So I tend to think of it as needing fewer clock cycles to do the same task (usually how it actually goes).

Think of it in terms of testing. If you want to know the difference between A and B, in terms of hardware, then you have to use the same exact software packages. (OS, benchmarks, applications, drivers,...).

In terms of testing software, it has to be on identical machines to truly gauge it. In the case of the '08 and '09 MP's, having Leopard on the '08 and Snow Leopard on the '09 can only tell you the total difference in performance. Not how it came to be though, as it's likely a mixture.

I've the strong impression you're expecting miracles of SL, and it can't provide them solely on it's own. It can make it easier for applications to run faster, but they have to be developed to do so. On single threaded applications, the only possibility SL contains to speed them up, is GPGPU processing.

Just don't expect too much, as you'll be really disappointed. As I'm sure you've realized, software development is always behind hardware. Simple, but sad truth of the matter.

 

[jjahshik32]

In terms of hardware, YES.

The gains SL may be able to produce over Leopard, would be replicated whether it's run on the '08 or '09 models. Both architectures would be able to run it, and the software, assuming it can remotely do what's promised, will be able to take advantage of the cores on both machines.

Any performance gains between the '08 and '09 models, both running SL, will strictly be a function of the hardware. Keep in mind, the biggest difference isn't the cores, but the Integrated Memory Controller + DDR3. (Thinking in terms of the CPU/memory systems, not peripheral devices such as PCIe devices, HDD chips,...).

The advantages SL is supposed to provide would be applicable to both machines. BTW, a great deal of overall performance gains will have to come from applications developed specifically to take advantage of what's provided under SL. Older, existing software likely hasn't yet been written to use the new API's. What SL can provide on its own (no applications involved), is it's supposed to be leaner. Leaner code has a nice little habit of performing faster. So I tend to think of it as needing fewer clock cycles to do the same task (usually how it actually goes).

Think of it in terms of testing. If you want to know the difference between A and B, in terms of hardware, then you have to use the same exact software packages. (OS, benchmarks, applications, drivers,...).

In terms of testing software, it has to be on identical machines to truly gauge it. In the case of the '08 and '09 MP's, having Leopard on the '08 and Snow Leopard on the '09 can only tell you the total difference in performance. Not how it came to be though, as it's likely a mixture.

I've the strong impression you're expecting miracles of SL, and it can't provide them solely on it's own. It can make it easier for applications to run faster, but they have to be developed to do so. On single threaded applications, the only possibility SL contains to speed them up, is GPGPU processing.

Just don't expect too much, as you'll be really disappointed. As I'm sure you've realized, software development is always behind hardware. Simple, but sad truth of the matter.

Oh boy.. nevermind!! Yea the 08 model will be exactly the same speed as the current models!! There are you guys happy now, end of discussion.

 

[nanofrog]

Oh boy.. nevermind!! Yea the 08 model will be exactly the same speed as the current models!! There are you guys happy now, end of discussion.

You're missing the point, or not understanding it.

The '09 MP (Nehalem) architecture could run SL faster than an '08 model MP, but it will depend on specifics (Quad or Octad). Any performance gain between them, would be due to the HARDWARE differences, not Snow Leopard.

Even with Snow Leopard, it will depend on the application. If it was written to take advantage of multiple cores (multi-threaded), it will be able to run faster under Snow Leopard than Leopard, on THE SAME MACHINE. For single threaded applications, not really, unless SL can apply the GPGPU capability to it.

 

[rylin]

For single threaded applications, not really, unless SL can apply the GPGPU capability to it.

Of course, a Mp 08 with a graphics card capable of OpenCL or CUDA will also be accelerated.
Just wanted to get that in there before he starts spouting "HAHA I TOLD U ZO!!11 SEE?!?! THE 09 IS SUPERIOR BECAUSE OF THE GPGPU!!11five"

 

[jjahshik32]

Of course, a Mp 08 with a graphics card capable of OpenCL or CUDA will also be accelerated.
Just wanted to get that in there before he starts spouting "HAHA I TOLD U ZO!!11 SEE?!?! THE 09 IS SUPERIOR BECAUSE OF THE GPGPU!!11five"

All in all the last gen machines are capable of that BUT in the end the Nehalem will be still faster. Get my point??

I can see you guys just itching to type back responses to defend your machines to death that its going to be the same as the Nehalems. I cant wait for SL to arrive so it'll shut you guys up for good. I already know off to eBay it goes.