Just thought I'd make a thread where people can discuss or argue about which is better and why, but no price talks here please. Just hardware to hardware comparisons along with Snow leopard benefits.
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.
Nehalem Mac Pro vs. Harpertown Mac Pro under Snow Leopard
- Thread starter jjahshik32
- Start date
- Sort by reaction score
I cannot discuss any Snow Leopard benefits with you at this time, as I don't have Snow Leopard, and neither do you.
I'm curious how grand central will efficiently schedule the usage of all 16 cores or rather the physical 8 cores with 2 threads from the Nehalem machines.
Its going to be a crazy.
Its going to be a crazy.
There won't be.
Those are hardware properties, and don't have anything to do with the OS.
its not 16 cores. its 16 virtual cores, and 8 physical cores.
it will schedule the usage as efficiently as it will for a computer with "only" 8 physical cores. because its the same thing. the only difference is you have 8 physical cores that dont "share" threads (have full power) and 16 cores with 2 sharing one physical core.
do the maths, if the computer spec is 8x2.26 and you make 16 cores out of this.
i would imagine it'd be better for Safari
but considering logic under certain circumstances already fills 7full cores (due to OS limits, which will be removed in SL and prolly next revision of logic), it will make no difference than it already is. it will fill 95% of hard raw power which is 15% different, which already shows now in all the benchmarks. (oh, beside logic benchmarks, which works craptastic with nehalem, actually, it works 60% worse than with previous gen).so release of snow leo/logic for logic studio means yeah, nehalem will be boosted severly, but that is only because it is at the moment doing worse than the previous gen and it has to get on par before beating it
the evan benchmark: 17tracks nehalem octo, 65tracks 08 gen octo.. its very simple. 17tracks because it only uses 7cores of of 16 and because these cores only have half the power of a physical core.
These forums lack a good selection of "smilies" so I found an appropriate one elsewhere...
I was totally expecting
It kind of summarizes what I think of jjahshik32 and his texas troll posts.
A word on hyper-threading as there appears to be much confusion on these forums about it:
In a nutshell hyper-threading allows for quicker context switching, between repeatedly used contexts, due to a complete second set of physical registers available per core. In now way whatsoever does hyper-threading provide additional processing power. You still have only 8 cores to do actual work.
So what is a context then? A context is the term used to describe everything an OS needs to know about a particular thread of execution. In simple terms this means a context must contain a complete copy of all CPU registers, registers being where all data must be placed inside the CPU before any calculations can be performed on it. There are also registers that hold information like the address of the next instruction to execute etc.
Each time an OS switches contexts the current context must be copied out to RAM and the new one loaded in. This takes a long time from the CPU's perspective which is why you hear arguments from some software developers that multi-threading their apps could possibly slow them down (generally referred to as "task switching overhead").
So the idea behind hyper-threading is to have a second set of physical registers in the CPU allowing the CPU to "remember" info about an additional context without having to read it all in from RAM each time. It's job is to help alleviate the task switching overhead.
That is all. There is no extra "processing power" at all provided by hyper-threading. It is simply an attempt to tackle a fundamental bottle neck in current CPU technology and therefore squeeze a little more juice out of a CPU.
The above is devoid of a lot of detail but I really don't want to put you all to sleep. The above is enough for anyone to get a useful understanding of what HT is.
Hope that helps.
In a nutshell hyper-threading allows for quicker context switching, between repeatedly used contexts, due to a complete second set of physical registers available per core. In now way whatsoever does hyper-threading provide additional processing power. You still have only 8 cores to do actual work.
So what is a context then? A context is the term used to describe everything an OS needs to know about a particular thread of execution. In simple terms this means a context must contain a complete copy of all CPU registers, registers being where all data must be placed inside the CPU before any calculations can be performed on it. There are also registers that hold information like the address of the next instruction to execute etc.
Each time an OS switches contexts the current context must be copied out to RAM and the new one loaded in. This takes a long time from the CPU's perspective which is why you hear arguments from some software developers that multi-threading their apps could possibly slow them down (generally referred to as "task switching overhead").
So the idea behind hyper-threading is to have a second set of physical registers in the CPU allowing the CPU to "remember" info about an additional context without having to read it all in from RAM each time. It's job is to help alleviate the task switching overhead.
That is all. There is no extra "processing power" at all provided by hyper-threading. It is simply an attempt to tackle a fundamental bottle neck in current CPU technology and therefore squeeze a little more juice out of a CPU.
The above is devoid of a lot of detail but I really don't want to put you all to sleep. The above is enough for anyone to get a useful understanding of what HT is.
Hope that helps.
so release of snow leo/logic for logic studio means yeah, nehalem will be boosted severly, but that is only because it is at the moment doing worse than the previous gen and it has to get on par before beating it
the evan benchmark: 17tracks nehalem octo, 65tracks 08 gen octo.. its very simple. 17tracks because it only uses 7cores of of 16 and because these cores only have half the power of a physical core.
I don't understand what you are saying. Have you tested Logic studio on a nehalem? And is it worse? I'm seriously doubting between 2.8 '08 and 2.26 '09. So if Logic is performing worse on the 2.26 '09 then I can't buy that.
Are there more benchmarks available yet? I only know the test from Loyd Chambers and Barefeats. But although they both did good in-depth testing, they focused on things that where not really very helpfull for me. Too bad for me, luck for others that needed that tests.
I have yet to see a head to head comparison between 2.26 '09 and 2.8 '08 with the same amount of RAM (and thus comparing DDR2 with DDR3) and same graphics card. What is te speed gain, what is the gain of hyperthreading, what apps work better on the '09, does Logic benefit from nehalem, Does FCP, and so forth ...
If you know where to find these tests, then please drop the link here.
cheers
the "logic" benchmark is a "selfmade" benchmark but its the one that actually reflects usage in real life, its been posted somewhere on this forum that nehalem 2.26 took only about 20 (I apologize for wrong figures in previous post, but they were not much off) tracks whilst 2.8 08s took from 50 to 70 (depending ram/os clutter)
nehalem mac pros will get soft updates to make logic work alright, but if you ask me the performance bump from 08 wont be as noticable because logic does not work efficiently even on the 08 8 cores..
so we with nehalems and penryns are all waiting for logic getting a good multithreading fix..
(does anybody remember the early 10.5.1+8.0.0 logic CPU spikes on even 3.2 machines?..)
https://forums.macrumors.com/posts/7283897/
here you go. ive ran this "benchmark" on both mbp and mp and its proves somewhat useful.
personal result for mp: 60tracks and one core idling (and i mean 0% idling)
now this is completely logical why nehalem got such low score, because logic can only utilize what, 7cores, thats 7 out of 16 splitcores of nehalem.
thats why i said it needs to get on par with 08 before it can beat it, but my guess is it will beat it for few tracks at most..
btw, on GS guesses are that 2.26 will match or beat 2.8 for few tracks at most and 2.93 will waste it
i know this benchmark is far from perfect, but its the only one audio guys currently have..
mind, logic has more issues with HDD speed and CPU than RAM speed... big things done in ram are usually loaded from HDD which is in most cases ive seen the biggest slowdowner for audio
Great it didn't even look like you read the original question if "just hardware to hardware comparisons" are wrong.
the "logic" benchmark is a "selfmade" benchmark but its the one that actually reflects usage in real life, its been posted somewhere on this forum that nehalem 2.26 took only about 20 (I apologize for wrong figures in previous post, but they were not much off) tracks whilst 2.8 08s took from 50 to 70 (depending ram/os clutter)
nehalem mac pros will get soft updates to make logic work alright, but if you ask me the performance bump from 08 wont be as noticable because logic does not work efficiently even on the 08 8 cores..
so we with nehalems and penryns are all waiting for logic getting a good multithreading fix..
(does anybody remember the early 10.5.1+8.0.0 logic CPU spikes on even 3.2 machines?..)
https://forums.macrumors.com/posts/7283897/
here you go. ive ran this "benchmark" on both mbp and mp and its proves somewhat useful.
personal result for mp: 60tracks and one core idling (and i mean 0% idling)
now this is completely logical why nehalem got such low score, because logic can only utilize what, 7cores, thats 7 out of 16 splitcores of nehalem.
thats why i said it needs to get on par with 08 before it can beat it, but my guess is it will beat it for few tracks at most..
btw, on GS guesses are that 2.26 will match or beat 2.8 for few tracks at most and 2.93 will waste it
i know this benchmark is far from perfect, but its the only one audio guys currently have..
mind, logic has more issues with HDD speed and CPU than RAM speed... big things done in ram are usually loaded from HDD which is in most cases ive seen the biggest slowdowner for audio
Allright, nice thread. Although some guy squeezed about 40ish tracks out of his 2.26, but that is still too low for such an expensive machine. Pay more, get less is not quite the salespitch I had in mind
But as the posters in that thread mentioned, wait until Logic gets a fix or an update, or try it without hyperthreading in the meantime.
I have to admit, it is all very confusing. I waited for 6 months now to be ready for the new faster MP's who claimed to be lower in price, nut instead of being faster and cheaper, the buzz on the boards is that is is only marginally faster but a lot more expensive.
And the worst thing is that I don't see any piece of software to become the next jezus and give this 2009 MP's a sudden jolt of holiness. Not Snow leopard, not Logic 2 not FCP 2 ... it will improve a bit, but it wont double the performance all of a sudden.
But then again, I could be wrong. Then I'm stuck with an old 2.8 for the next 5 years, which 'will just do', while all the new stuff is doing the leap.
aaaaaargh, this unclarity frustrates me. I was Apple minded, but it is waning.
Steve, get your ass over here and save us
Snow Leopard is hype. Every computer will see a boost, if there is a boost. There is no "trick" that Nehalem has that hasn't been taken advantage yet. Anyone who says otherwise doesn't know what they're talking about.
So the supposed 40% boost in latency is a myth?
Latency? The nehalem processors are faster clock for clock compared to the harpertown models of yesteryear. Any advancements Snow Leopard has will be equally taken advantage of throughout the line-up.
I don't know where people get this garbage. :/
I don't know where people get this garbage. :/
Okay, I'm merely recalling a pretty heated discussion that some member brought up (ahem) claiming that there would be that latency increase. Not naming names, but it started with "jj", and ended with "32".
I'm not lending any credibility to those numbers, but they sounded interesting.
Offf that topic, you're saying that a Harpertown (2008) chip is clock for clock slower than the 2009 Nehalems? That very much interests me. Describe that a bit for my education if you can. . .
I'm not lending any credibility to those numbers, but they sounded interesting.
Offf that topic, you're saying that a Harpertown (2008) chip is clock for clock slower than the 2009 Nehalems? That very much interests me. Describe that a bit for my education if you can. . .
Benchmarks showed that Nehalem chips were up to 40% faster than their Penryn Core 2 counterparts in some situations. That's correct. That's the extent of their gain, there is nothing else that Snow Leopard can "unlock." Boosts in speed will be through a tightly coded OS. More substantial boosts will be available for multi-core machines and machines with powerful GPUs (The Mac Pro will probably see a better jump than a Mac Mini).
As far as the "faster clock for clock" thing, there isn't much to it. Hyperthreading uses unused cycles, and other differences in the architecture also help. There isn't much to explain. A more extreme example would compare a 3Ghz P4 to a 3Ghz i7 processor. If you only enable one of the four cores in the BIOS (On a PC), the i7 core will smoke the P4.
Got it. Also did a Google search on that same topic , using "clock speed Nehalem Harpertown".
Even last year's releases and early evaluations say what you'd just posted. Interesting information here from a Nov 13, 2008 review.
Apple has traditionally used the Xeon server-class processors for their Mac Pro line. The current Mac Pro uses the Xeon ("Harpertown") 5400 series processors and is due for an update. Despite comparable clock speeds, the Nehalem-based processors have been shown to offer clock-for-clock improvements performance improvements up to 29%.
Meaning that the 2.26 Nehalem (theoretically) is comparable with a 2.93 Harpertown in some cases.
Even last year's releases and early evaluations say what you'd just posted. Interesting information here from a Nov 13, 2008 review.
Apple has traditionally used the Xeon server-class processors for their Mac Pro line. The current Mac Pro uses the Xeon ("Harpertown") 5400 series processors and is due for an update. Despite comparable clock speeds, the Nehalem-based processors have been shown to offer clock-for-clock improvements performance improvements up to 29%.
Meaning that the 2.26 Nehalem (theoretically) is comparable with a 2.93 Harpertown in some cases.
Snow Leopard is hype. Every computer will see a boost, if there is a boost. There is no "trick" that Nehalem has that hasn't been taken advantage yet. Anyone who says otherwise doesn't know what they're talking about.
i admit that i don't know anything technical about processors, but my understanding was that the QPI of the Nehalem architecture would actually allow for a greater utilization of Grand Central's scheduling advances compared to the harpertowns.
is that incorrect?
I'm a former CPU designer for AMD. Among other things, QPI reduces the penalty for moving processes from processor to another, or for accessing memory attached to processor A from processor B. (To be fair - I am assuming QPI behaves like AMD's HyperTransport).
Grand Central presumably is aware of the various penalties and could theoretically take advantage of the reduced penalties in certain situations. I wouldn't think this is a very important difference.
