6-Core Version of Apple's New Mac Pro Appears in Benchmarks

[friedmud]

Are these really only (single (slower than imac?!))-1000/ (multi)2700+ more than the new iMac's?
http://browser.primatelabs.com/geekbench3/159719

Or am I totally misreading/misinterpreting something?

EDIT: Ah guessing the 32bit / 64bit is the culprit?

Actually - the processors in that iMac are "Haswell" processors which are a newer architecture than the "Ivy Bridge" based Xeons the new Mac Pro will run.

Haswell brings several architecture improvements which can speed up floating point calculations. We see about a 30% boost over Ivy Bridge / Sandy Bridge in our (compute intensive) scientific applications.

So - it's not entirely unreasonable for a 4 core iMac to beat (or score similarly) to a 6 core Mac Pro.

The value of the Mac Pro is to go to _more_ than 6 cores. 12 Ivy Bridge cores (if you can utilize them all simultaneously like we can) will be WAY faster than 4 Haswell cores...

 

[deconstruct60]

....
Are they just going to ignore the Mini this year?

They do have a large product announcement hole that runs from January-June over last couple of years.

They have also ignored the classic MBP 13" so far too, but don't want to drop it.

 

[thejadedmonkey]

Yeap, my mom would think it looks like a modern blender after realizing is not a trash can.

I took that quote to be serious, as a high end coffee machine is about the price of a Mac Pro.

 

[deconstruct60]

We see about a 30% boost over Ivy Bridge / Sandy Bridge in our (compute intensive) scientific applications.

So - it's not entirely unreasonable for a 4 core iMac to beat (or score similarly) to a 6 core Mac Pro.

If get a 30% boost but there is a 50% increase in core count why is that a reasonable expectation? 4 * .30 is 1.2 versus 2 whole additional cores. Throw on top 20% bigger L3 cache for 6 core and have a hanidcap in that area too.

As long as can fully fill the 6 cores fp pipelines and keep them full, the microarch differences aren't got matter as much as adding full fledged cores (> 2). Both v2 (Ivy Bridge ) and v3 (Hawell ) have AVX instructions so there is a huge "lack of AVX" gap that the v2 had over 2011-2012 gen Mac Pro CPUs for SIMD float throughput.

The value of the Mac Pro is to go to _more_ than 6 cores.

Questionable. The 8 and 12 core versions come at E5 2600 prices; not E5 1600 prices. You'll be paying around $500 per core for those additional cores. That is actually more than a whole haswell 4 core model in the iMacs. (even the top end one).

So it better be that there is both very tight coupling in memory to the other work going on for that additional core.

12 Ivy Bridge cores (if you can utilize them all simultaneously like we can) will be WAY faster than 4 Haswell cores...

WAY more expensive too. Value is in $/performance not just biggest baddest geekbench score.

If have stuff that is just simply "embarrassingly parallel" in every way the GPUs may be the differentiator between Mac Pro and other Mac models.

 

[50548]

These machines are going to be blazing fast, and I have to admit that there is something in me that wants to get one just because it's built in America by Americans. USA, USA, USA!

America is a continent, not a country...as in "South America", "Central America" and "North America" - unless you still believe in the Monroe Doctrine, of course

 

[Bibbler]

That's wonderful. And I'm sure this thing has a market somewhere.

Now where is the simple Mac in a mini tower case (that can be easily opened) with an i7, user serviceable Card Slots (including GPU), RAM, and drives selling for about $899

Yep, the xMac. Apple users have only been asking for it for about 10 years.

 

[milo]

America is a continent, not a country...as in "South America", "Central America" and "North America" - unless you still believe in the Monroe Doctrine, of course

America |əˈmerikə|(also the Americas )
a landmass in the western hemisphere that consists of the continents of North and South America joined by the Isthmus of Panama. The continent was originally inhabited by American Indians and Inuits. The northeast coastline of North America was visited by Norse seamen in the 8th or 9th century, but for the modern world the continent was first reached by Christopher Columbus in 1492.
• used as a name for the United States.

American |əˈmerikən|
adjective
of, relating to, or characteristic of the United States or its inhabitants: the election of a new American president.
• relating to or denoting the continents of America: the American continent south of the tropic of Cancer.
noun
1 a native or citizen of the United States.

So "America" is a country as well as a continent. Not that any of that is remotely on topic.

 

[friedmud]

If get a 30% boost but there is a 50% increase in core count why is that a reasonable expectation? 4 * .30 is 1.2 versus 2 whole additional cores. Throw on top 20% bigger L3 cache for 6 core and have a hanidcap in that area too.

Well - our 30% is just a ballpark number... and we're certainly not perfectly optimized for Haswell (like this benchmark may be). I was just pointing out that there are reasons a Haswell core in an iMac might compare favorably...

Questionable. The 8 and 12 core versions come at E5 2600 prices; not E5 1600 prices. You'll be paying around $500 per core for those additional cores. That is actually more than a whole haswell 4 core model in the iMacs. (even the top end one).

So it better be that there is both very tight coupling in memory to the other work going on for that additional core.

WAY more expensive too. Value is in $/performance not just biggest baddest geekbench score.

If have stuff that is just simply "embarrassingly parallel" in every way the GPUs may be the differentiator between Mac Pro and other Mac models.

Definitely more difficult to use two iMacs for compiling, running and debugging than using _one_ Mac Pro with 12 cores. For real jobs I'm going to go use my 12,000 core cluster... but that's not convenient for day to day development...

The price of this machine is a drop in the bucket compared to the amount I cost my company. If it saves me even 2% of my time over _one_ year it will already pay for itself. Each year after that is gravy.

These boxes are tools for professional use. When you are making money using that tool the only question is whether it is worth it in a cost/benefit analysis.

Our stuff is not written to take advantage of GPUs (although we do fairly well with Intel Phi)...

 

[td2243]

It seems premature to be equating benchmarks to a machine when the tests are simply tests similar parts. Correct?

I'll probably wait until the Mac Pro's are on the market with not-so-speculative benchmarks on the table.

 

[LordVic]

Curious as i don't have these stats myself.

how does this compare to similarly built hardware thats not the Mac Pro?

say, build your own desktop with similar internals. are we looking at similar performance? Worse? better? is there something to the Mac Pro that makes this somehow "better" than building your own?

I know there are subjectives, such as OSx, and design of the casing itself.

But when it comes to raw performance and numbers, What is the cost to performance?

----------

America is a continent, not a country...as in "South America", "Central America" and "North America" - unless you still believe in the Monroe Doctrine, of course

as someone in Canada, While technically in the American Continent. Canadians by no way consider themselves American.

We try and distance ourselves and our national pride from those of Americans.

its like Calling New Zealanders Australians.

----------

That's wonderful. And I'm sure this thing has a market somewhere.

Now where is the simple Mac in a mini tower case (that can be easily opened) with an i7, user serviceable Card Slots (including GPU), RAM, and drives selling for about $899

Yep, the xMac. Apple users have only been asking for it for about 10 years.

I think this is a very good question.

Apple has never really been able to penetrate the PC market. Even now, despite having made some headway, are sitting in Sub 10% of the overall PC market.

I believe one of the leading causes is a lack of product. Apple has some very nice niche products, But they do not have the "one box" to cater to many.

The iMac is a good all in one PC for people looking for your typical "work" style computer, that has a little bit of "oomph".

The Mini is good for those who are looking for the bare minimum in computing in an appliance that can just be plugged in to go.

The Mac Pro has been up to now for those who need serious computational power for workstation related activities. Very high end, and is expensive.

But the average gamer. The typical Computer user who wants the occasional upgrade. Desktop class components with some horsepower to play modern games at high settings has absolutely no product in the Apple lineup to suit them.

And Apple has made no effort to try and win these people over. it's a very large market segment that isn't going away. Even in this "post pc" (I hate that term) market, the typical PC Gamer is going absolutely nowhere. They still need the robust consumer grade computer to game on.

They just have no choice but to go elsewhere instead of Apple. I Think Apple mis-judged the new Mac Pro. I think the design and internals should have been aimed at this market instead. Leaving the Professional market with similar design to the old Pro model, with lots of upgrade room internally and less proprietary devices

 

[twietee]

It's a just a tool to use. I imagine the 12 core maxed spec out with 1tb pcie-ssd onboard 64gb ram and best GPU is going to cost me near $8-10K - but it'll make it's money back for me in a few weeks.

Oh and they are tax deductible so essentially free over time anyway.

If 64GB serve your 12/24cores well enough, good for you. I mean it. Otherwise, if people can't actually use dualGPUs the whole pricepoint of this machine is moods. And as a professional you just don't buy a machine and wait afterwards for your programs to get updated so you can use it properly.

And just because it's a tool doesn't mean it's making a lot of sense that way. Don't see many plumbers rocking golden plungers, even if they could write it off.

But I don't want to badmouth this machine, it's a fine tool. But sadly not serving my needs - at all.

 

[mrhick01]

little disappointing score. when macbook Pro 15" scored over 13.000. so not that big of a difference.

It is over a 40 percent improvement. And that is using the 32-bit version of geekbench.

40 percent is a lot if you make money from your computer.

I'm looking forward to getting the 6 core. It's over twice as fast as my 8 core 2008 Mac Pro, so it's time to make the jump. I have dabbled with audio and video editing over the years, it will be great to work with a dedicated workhorse computer.

 

[kappaknight]

The MP quad core is also 3.7 as opposed to that chip at 3.5 so hopefully it will be faster even on the 32 bit. The 12 core is a 2.7 so that will bring down single core performance, we'll have to see to what degree once there are tests on GB 3 and of course 64 bit tests for all models.

Leak out lower results as opposed to higher 64 bit ones? Doubtful, if Apple was going to do it intentionally they'd be idiots to leak 32 bit results. These are people leaking on their own, who haven't bought the paid version of GB.

Serious question: if someone's willing to pay for it, could they not make a 12 core 3.7 if they could achieve that speed with the 4 core?

How ironic. Isn't that calling the kettle?

Sarcasm is wasted on this one.

 

[tuartboy]

What's really interesting to me is that my 2013 rMBP actually outperforms the 6-core Mac Pro on single-threaded performance.

Here's my Geekbench run.

If these numbers are accurate, a rMBP is going to be faster than a Mac Pro for many tasks. It's amazing how far laptops have come.

 

[WestonHarvey1]

We try and distance ourselves and our national pride from those of Americans.

its like Calling New Zealanders Australians.

Which reminds me, while living in Australia, many Australians asked me where I was from. I had that little voice inside that said "Don't say America! It's wrong and racist to say America!", so I'd say United States. They'd say "...where?" Then I'd say "The States?" then I'd say... "The United States of .... America?" and then they'd say "Oh! America!"

Sometimes I'd get "Are you American, or Canadian?"

So it's not just here. America is a pretty common alias for USA in at least one other country.

 

[Nevaborn]

We can't decide on the new 12 core Mac Pro or a new coffee machine for the new kitchen.

The Mac probably makes the best cup of coffee in the world while running your life and your yearly taxes, all in something smaller than that £50 piece of junk youd buy that needs capsules.

 

[Ralf The Dog]

This will be an awesome machine to run Mathematica on. I wonder how many cores they'll allow to run on the personal license of their software.

licenses needed = sqrt(Cores on first CPU^2 + cores on second CPU ^2);

 

[milo]

It seems premature to be equating benchmarks to a machine when the tests are simply tests similar parts.

The assumption is that these tests are legit although it's possible someone is pulling a fast one. There is a quad core result that looks like an obvious hoax.

Don't see many plumbers rocking golden plungers, even if they could write it off.

If it helped them do their work considerably faster they would. Not the best analogy.

Serious question: if someone's willing to pay for it, could they not make a 12 core 3.7 if they could achieve that speed with the 4 core?

Apple can only ship what Intel has available. If you mean could Intel do it, it's probably a heat issue more than anything else. They might be able to run a 12 core at 3.7 with crazy cooling and low yields, but probably not practical to release to the public with their existing technology.

 

[TheRealTVGuy]

my 5.1 12 core will take this garbage can's school money

And what OS do you run? Winblows? Hackintosh?

Get real...

 

[deconstruct60]

Well - our 30% is just a ballpark number... and we're certainly not perfectly optimized for Haswell (like this benchmark may be). I was just pointing out that there are reasons a Haswell core in an iMac might compare favorably...

It is going to compare favorably with common core counts. It is not going generally going to claw back core count deficits. It is just one iteration ahead microarchecture wise and no major leap in computational instructions. Versus the 2-3 iterations with Mac Pro 2009-2012 versus 2013 comparisons.

There are some transaction memory improvements with Haswell but those probably won't have large spread, high value impact until they make it to the Xeon E5 (and above) class CPUs. Tight loop numeric kernels aren't going to see much benefits.

When you are making money using that tool the only question is whether it is worth it in a cost/benefit analysis.

Which means it is questionable (or at least worthy of getting/measuring answer to the question). The cost/benefit analysis may or may not work for some other folks. 4 and 6 core Mac Pros are useful and far more cost effective for a large group of folks. That is exactly why they are announced standard configs and the 8 and 12 core models are only optional BTO configs.

Our stuff is not written to take advantage of GPUs (although we do fairly well with Intel Phi)...

Similar issues of cost/benefit trade-offs per context. If have already been aiming at 10-16 core Intel x86 contexts then Intel Phi is an easier move. The chance of a Phi though running tightly coupled with OS X though is approximately zero though. So it isn't necessarily the most viable choice if looking for a "embarassingly parallel" solution that is present across the whole Mac line up going forward from 2013 on. OpenCL is.

 

[kustardking]

"later this"?

I think it's actually "next"

 

[vampyr]

Am I the only one that thinks these abitrary number "scores" from some program is meaningless??

It's about as useful as hooking up a crate motor to a dyno and saying that the engine can go up to 9000rpm. Hey... that's awesome... but until you put it in a car, you have no idea if that's a good number or not.

 

[MacPrometheus]

Processors just aren't getting much faster and that's the path we are on for transistor based CPUs.

The future is parallel, and I think that's one of the reasons Apple went for dual GPUs instead of dual Xeons to fill the precious little space in the Mac Pro enclosure. They're betting on a GPGPU future for high performance computing.

Maybe somewhere in the next few generations they'll find a way to cram another Xeon in anyway.

I agree with you half way. The future is stacking 3D. Why travel all the way across the CPU when you can move 1/100 the distance up. I can't wait until this tech hits the streets in force.

 

[milo]

Am I the only one that thinks these abitrary number "scores" from some program is meaningless??

Of course there's a limit to their value, but hardly meaningless. If I have a machine and know the score, and I know I need to get a faster machine, these scores are very useful to me in regards to a general sense of what machine to get and what kind of performance boost to expect.

Benchmarks aren't the same as real world work, but they give a general idea of how the relative speeds of different machines compare.

 

[montom]

Oh man! Textedit on this bad boy's gonna be shaweeeeeet!!!