MacRumors: Apple's New 8-Core Mac Pro Shows Up in Benchmarks

[M Most]

mike, I don't get you..oh hang on, you're the guy who thought the renouned and well respected BBlist was a scam and posted about it without checking..its where I bought all my lenses..

You have me confused with someone else. I honestly have no idea what you're talking about. I don't know what "BBlist" is and have never posted anything about it.

 

[Antony Newman]

With Intel's shrinking PC market, and their fabs not running at full capacity - I think Intel would jump at the opportunity to keep fabs 'more full' until their mobile entry is cemented.
They may charge $3k for 12-core chips ... but a knock down price for Apple would probably give them more of the volume they need. I GUESS that Apple will get these for circa $1k.

AMD have a lot to gain from a move that will level that CUDA advantage that Nvidia have, and put mounting pressure on companies to use OpenCL. Apples minimum configuration is going to be two AMD GPUs, and have probably guaranteed that they will only ship new Mac Pros with AMD for the first couple of years. I GUESS AMD will sell these to Apple for $1.5k for a pair.

Apples says they are planning this form factor for the next decade and one that will comfortably deals with 4k as standard. To make this a standard they'll need to shift boxes and I GUESS they will stop selling the old form factor.

I guess that
+) An entry level 8 core Mac Pro will appear at $3k (32GB of RAM + 512GB Flash).
+) An entry level 12 core Mac Pro will appear at $3.5k (32GB of RAM + 512GB Flash).
+) A maxxed out 12 core Mac Pro will appear at $8k (128GB RAM + 4TB Flash)

And that Apple will shift lots of entry level 8 core machines .. and make enough $$ to keep them happy on the maxxed out machines.

Who's holding the bets?

AJ

 

[luigivaltulini]

[/URL][/IMG]




a lot of us have already moved on... this was my first system, 16 cores, 32 threaded.

built a second one 6 core 12 threaded, 22k geekbench score, cost less then 4k to build. machine is def a 11 in my book as well.

hehe Tom, and when he gets' the new Mac Pro would have the new smbios.plist, so we can activate the turbo boost and go to almost 40000 points Geekbench

still a few months


Luigi

 

[Tom.Wong]

I know Lougi, i'm stoked for mavericks

true gtx 7xx and titan support as well. Right now ML doesn't really know how to take advantage of it, and open cl doesn't really work with gtx 6xx and above cards. Mavericks should fix all of that. Gonna be a whole new gen of performance. I'm also seeing very little performance increase with haswell processors over sandy bridge. Isn't even worth updating the processors. Haswell is all about a little better performance than Sandy, but better cooling and power management. Which really doesn't matter when you have liquid cooling and a 1200w power supply... great for Surface Pro 2's though

 

[luigivaltulini]

I know Lougi, i'm stoked for mavericks

true gtx 7xx and titan support as well. Right now ML doesn't really know how to take advantage of it, and open cl doesn't really work with gtx 6xx and above cards. Mavericks should fix all of that. Gonna be a whole new gen of performance. I'm also seeing very little performance increase with haswell processors over sandy bridge. Isn't even worth updating the processors. Haswell is all about a little better performance than Sandy, but better cooling and power management. Which really doesn't matter when you have liquid cooling and a 1200w power supply... great for Surface Pro 2's though

you're right ...
we just have to wait for this "fucking" SMBIOS
our Sandy Bridge and then take the flight )

 

[Mike Halper]

You know the reality - if you really wanna an 11+ machine that runs OSX, TB2, PCIe, etc. - you gotta hack it > http://www.asus.com/News/22a1IgaOOVSdTMoM

Personally, I find that motherboard to be rather useless with all those x1 slots. I really don't understand what the deal is with these motherboard manufacturers half-assing all these motherboards that are supposed to be pro/high end. No high end or pro user is going to have 4 x1 cards for those slots. On my system I have an x16 GPU (and want to add a second), 3 x8 cards, and an x4 card. 1 x16 and 2 x8 slots doesn't cut it. I understand you have Thunderbolt 2 ports on that board, but they aren't hot-pluggable in Windows or OS X (Hackintosh), you can quickly saturate that bandwidth, and sometimes they don't work. I'd rather see them add more PCIE chipsets to add more lanes and charge more money for them. IMO $500 isn't that much for a piece of hardware as important as the motherboard. I'd totally be willing to pay $700-800 for that board if it has the 2 TB2 ports in addition to 2 x16 double-width slots, and the rest x8... or even more x16.

 

[Mike Halper]

With Intel's shrinking PC market, and their fabs not running at full capacity - I think Intel would jump at the opportunity to keep fabs 'more full' until their mobile entry is cemented.
They may charge $3k for 12-core chips ... but a knock down price for Apple would probably give them more of the volume they need. I GUESS that Apple will get these for circa $1k.

AMD have a lot to gain from a move that will level that CUDA advantage that Nvidia have, and put mounting pressure on companies to use OpenCL. Apples minimum configuration is going to be two AMD GPUs, and have probably guaranteed that they will only ship new Mac Pros with AMD for the first couple of years. I GUESS AMD will sell these to Apple for $1.5k for a pair.

Apples says they are planning this form factor for the next decade and one that will comfortably deals with 4k as standard. To make this a standard they'll need to shift boxes and I GUESS they will stop selling the old form factor.

I guess that
+) An entry level 8 core Mac Pro will appear at $3k (32GB of RAM + 512GB Flash).
+) An entry level 12 core Mac Pro will appear at $3.5k (32GB of RAM + 512GB Flash).
+) A maxxed out 12 core Mac Pro will appear at $8k (128GB RAM + 4TB Flash)

And that Apple will shift lots of entry level 8 core machines .. and make enough $$ to keep them happy on the maxxed out machines.

Who's holding the bets?

AJ

I would really love to know where you're getting your $1K figure from. That's just insane. It's not in Intel's best interest to sell the CPUs that cheaply. They have no competition to worry about. It's not like Apple can suddenly switch to AMD chips, which there is no comparable product for anyway. There's no reason for Intel to sell a $2700 chip to Apple at a 63% discount. I'm sure Apple is getting a bulk discount, but it wouldn't be that much. Intel does have investors to answer to who would question those kinds of discounts that aren't necessary. I find it funny that so many people here don't think about things in a realistic manner.

 

[Mark L. Pederson]

Personally, I find that motherboard to be rather useless with all those x1 slots. I really don't understand what the deal is with these motherboard manufacturers half-assing all these motherboards that are supposed to be pro/high end. No high end or pro user is going to have 4 x1 cards for those slots. On my system I have an x16 GPU (and want to add a second), 3 x8 cards, and an x4 card. 1 x16 and 2 x8 slots doesn't cut it. I understand you have Thunderbolt 2 ports on that board, but they aren't hot-pluggable in Windows or OS X (Hackintosh), you can quickly saturate that bandwidth, and sometimes they don't work. I'd rather see them add more PCIE chipsets to add more lanes and charge more money for them. IMO $500 isn't that much for a piece of hardware as important as the motherboard. I'd totally be willing to pay $700-800 for that board if it has the 2 TB2 ports in addition to 2 x16 double-width slots, and the rest x8... or even more x16.

I should have posted this one -

The Gigabyte Z87X-UD7 TH is built in the standard ATX form-factor. The board features five PCI-Express 3.0 x16 slots, of which three are gen 3.0 x16 capable, and four gen 3.0 x8 capable. A mechanism lets you bypass the bridge chip, and connect a single graphics card directly to the CPU. Two PCI-Express 2.0 x1 slots complete the expansion area.

http://www.eteknix.com/computex-gigabyte-show-off-z87x-ud7-thunderbolt-motherboard/

and this is coming soon -

http://www.anandtech.com/show/7325/gigabyte-tease-x79-refresh-x79-g1assassin-3

 

[egon touste]

You have me confused with someone else. I honestly have no idea what you're talking about. I don't know what "BBlist" is and have never posted anything about it.

my appologies, you're right about that.. that was someone else..i just dislike religion more than Apple, so you rubbed me up the wrong way with you're last post

 

[Antony Newman]

Mike,

I make may well have guessed wrong - maybe I didn't use the word 'guess' liberally enough in my post

If Intel's margin on the Xeon e5-2650L v2 (ie 10 core Xeon / 22nm process) was 70% - and is sold in single units at $1396 - then the cost per chip is around $420 for Intel.

If the cost of making chips is roughly proportional to die size, and that is in proportion to the number of transistors, and that is in proposition to the number of cores - then the Intel cost of a 12 core could be 20% more - say $500.

Hypothetical Sales
===========
Case 1) Intel sell the chips to Apple for 1k a pop = $500 a chip profit.
Apple entry level $3.5k / manages to sell 100,000 new mac pros (including some people who didn't intend to buy a desktop)
Intel make £50 million profit

Case 2) Intel sell the chips to Apple for 2k a pop = $1500 a chip profit.
Apple entry level $5k / manages to sell 20,000 new mac pros (mostly hard core people thinking about ROI)
Intel make £30 million profit


In 3 years time 1/2 of the people want to upgrade their machine ..... which version would the Intel shareholders most like to happen?

Without any hard facts, or knowledge of the deals being hammered out - this is of course complete conjecture on my part.

AJ

 

[Blair S. Paulsen]

Tactical view

Tactical view

There are many ways to slice this Apple in terms of volume vs margin for all their suppliers. In Intel's case, ROI on 32nm fabs before they migrate to 22 might figure in.

Whether or not Apple can use their clout to get killer deals from their vendors, I still think the bigger issue for the selling price is where Apple would like to be in the marketplace. If significantly larger market share, high enough volumes to lower unit costs and providing a platform for people to consume content/storage/etc is their mantra then a thin margin makes sense. If positioning themselves as a premium option for people who want a superior user experience is the market segment they seek, then making a fat margin on sexy new hardware - even if it means lower volume - fits that mold.

Cheers - #19

 

[M Most]

my appologies, you're right about that.. that was someone else..i just dislike religion more than Apple, so you rubbed me up the wrong way with you're last post

I feel the same way. I used the religion reference to describe decisions and conclusions that are made with no factual input or support, based purely on belief, heresay, and/or superstition, and therefore inarguable if one is to try to use any kind of rational, logical thought. You can't "win" any debate or argument when it involves religion because there is no factual or rational basis for any of the opinions. That describes conversations involving any religion, and it sadly often describes conversations about Apple and their products, both by those who "believe" and those who don't.

 

[paul schefz]

I guess that
+) An entry level 8 core Mac Pro will appear at $3k (32GB of RAM + 512GB Flash).
+) An entry level 12 core Mac Pro will appear at $3.5k (32GB of RAM + 512GB Flash).
+) A maxxed out 12 core Mac Pro will appear at $8k (128GB RAM + 4TB Flash)


AJ

as much as i would love to see that, i doubt it will happen....
first i would be surprised if a bare system will ship with more then 16gb ram....and what is the point of 4gb flash? although the internal drive speed of these will make it interesting to actually use any internal drives for anything other then system (in which case 256 or 512 is plenty) i doubt there is room for more then 2 drives and i believe that the max capacity/drive is 1TB right now, which would put it at 2TB....and considering that the 1TB option for the imac is 1000$....
i think the real unknown is the options/prices for GPU(s)....
i still think that apple would like to have a bare bones system coming in right under 4000$....

 

[paul schefz]

It's been a long day, so I'm probably not explaining myself very well. Simply put, the new Mac Pro feels like it goes to 8, and I want it to go to 11. I want a faster machine to run OS X, and the future is bleak.

according to these benchmarks, the low end mac pro will be as fast as the last high end mac pro....not really taking into consideration all GPU options....or TB2 expansion

we all want faster and more for less money....but apple will never make an empty box....
this new mac pro will cover 99.9% of all computing needs and keeping things contained from a hardware standpoint has always helped apple squeeze the most out of their hardware by not having to bloat their software to accommodate more hardware....
in a way it works the same with the iPhone....is it the best phone, camera, browser with the most and fastest options? no...but the overall package is hard to beat...for consumers and pros....but if you need a very specific option, you might have to go with something else (and live with those shortcomings, but maybe they don't matter to you anyway, so....)

 

[Mike Halper]

I should have posted this one -

The Gigabyte Z87X-UD7 TH is built in the standard ATX form-factor. The board features five PCI-Express 3.0 x16 slots, of which three are gen 3.0 x16 capable, and four gen 3.0 x8 capable. A mechanism lets you bypass the bridge chip, and connect a single graphics card directly to the CPU. Two PCI-Express 2.0 x1 slots complete the expansion area.

http://www.eteknix.com/computex-gigabyte-show-off-z87x-ud7-thunderbolt-motherboard/

and this is coming soon -

http://www.anandtech.com/show/7325/gigabyte-tease-x79-refresh-x79-g1assassin-3

That UD7 board looks promising, but it will probably like all the other Gigabyte boards where if you use more than one x16 slot then they drop down to x8, and if you use more than one x8 slot they drop down to x4. Nice to see an updated X79 board, but it seems to have too few USB3 ports (only 4 when other newer boards have as many as 10 or even 12) and there's really 4 greater than x1 slots and maybe no Thunderbolt. If you want dual GPU, capture card, RAID, and HBA, you're SOL on something.

Edit: I also don't understand some of the choices of chipsets in a lot of these boards. The manufacturers have to know that there's a huge number of people wanting to run OS X on these, yet they include chipsets like Creative audio that aren't compatible with OS X. They could sell a lot more products if they purposely made it so they're easier to set up with OS X. They don't need to advertise it as OS X compatible but the Hackintosh community would catch on very quickly that they are.

 

[Mike Halper]

Mike,

I make may well have guessed wrong - maybe I didn't use the word 'guess' liberally enough in my post

If Intel's margin on the Xeon e5-2650L v2 (ie 10 core Xeon / 22nm process) was 70% - and is sold in single units at $1396 - then the cost per chip is around $420 for Intel.

If the cost of making chips is roughly proportional to die size, and that is in proportion to the number of transistors, and that is in proposition to the number of cores - then the Intel cost of a 12 core could be 20% more - say $500.

Hypothetical Sales
===========
Case 1) Intel sell the chips to Apple for 1k a pop = $500 a chip profit.
Apple entry level $3.5k / manages to sell 100,000 new mac pros (including some people who didn't intend to buy a desktop)
Intel make £50 million profit

Case 2) Intel sell the chips to Apple for 2k a pop = $1500 a chip profit.
Apple entry level $5k / manages to sell 20,000 new mac pros (mostly hard core people thinking about ROI)
Intel make £30 million profit


In 3 years time 1/2 of the people want to upgrade their machine ..... which version would the Intel shareholders most like to happen?

Without any hard facts, or knowledge of the deals being hammered out - this is of course complete conjecture on my part.

AJ

I'm not sure where the profit margin figures are coming from, or the assumption that die size and manufacturing cost are proportional (if anything I'd expect smaller die size to be more complicated and thus more expensive), but something to consider is the yield. Intel may not be able manufacture that many CPUs to allow them to lower the cost. That was a major issue for IBM with the PowerPC G5 CPUs. Also, I may be wrong, but isn't there a legal issue with Intel offering considerably lower prices to Apple and not others? And why would Intel diminish the reputation of their Xeon processors like that? They didn't do anything like this in the past and there's no reason they should now.

Apple will likely sell multitudes more new Mac Pros than they've been selling the old Mac Pro in the last 2 years, which already means higher volume for Intel.

 

[Jeff Kilgroe]

I know Lougi, i'm stoked for mavericks

true gtx 7xx and titan support as well. Right now ML doesn't really know how to take advantage of it, and open cl doesn't really work with gtx 6xx and above cards. Mavericks should fix all of that. Gonna be a whole new gen of performance. I'm also seeing very little performance increase with haswell processors over sandy bridge. Isn't even worth updating the processors. Haswell is all about a little better performance than Sandy, but better cooling and power management. Which really doesn't matter when you have liquid cooling and a 1200w power supply... great for Surface Pro 2's though

OpenCL works fine with GTX680/690. Pretty well has since the 680 Mac Edition card hit the market. Now the 670 and lower is another story... You do need to have the video cards flashed with a proper EFI, can't just pick at the kexts and enable an off the shelf PC GPU. There are some necessary EFI hooks for full OpenGL support with these newer GPUs.

I wouldn't hold my breath for complete Titan support. 780 support seems likely, but a lot just depends on what happens with Mac Pro GPU options moving forward and who decides to support what. One ray of hope here is that the new GPU form factor for the Mac Pro is actually an open design and more of an Intel thing than an Apple thing. AMD is already on board, no word publicly about nVidia just yet. But this isn't a design specific to the new Mac Pro, it's a smaller form factor design that moves all the interface connections away from the GPUs so that GPUs can be easily leveraged in greater quantities as both display and compute processors. I'm really hoping the rest of the PC industry adopts it. Although, the rest of the PC industry is in a huge funk and can't seem to break away from 1997 design theory. Every time I build a PC I'm more pissed off at the industry as a whole. We haven't seen a new form factor or new innovation in the hardware arena (outside of evolutionary process shrinkages and increased speed, more cores) for the past 20 years. People complain that Apple can't innovate anymore. WTF? At least their trying something new. Because, you know, I just love building a PC in the same old box with 1960's tech used for the power supply and power subsystems. And aside from CPU and GPU, it doesn't get a whole lot better from there...

Now there's another angle from which to look at all this. Software. Aside from highly serialized or nodal tasks -- like 3D rendering and such that can be broken down into specific chunks -- current software just can't use all the available cores in a system. You can build a monster that will score 50000 in GeekBench, but when it comes to actually running your applications, is it really faster? My overclocked 6-core i7 shreds through Photoshop, REDCINE-X, and most other apps a lot faster than my 16-core Z820 does. The Z820 has the edge in Premiere because it can use a few more cores before it tops out -- 16 threads or so seems to be where you cross the point of diminished returns. Resolve is better on the Z820 and all my rendering applications of course love the extra cores and RAM.

I just sent out the door a brand new workstation with dual E5-2697 V2 CPUs (almost $2800 each!), 256GB RAM and dual Titan video cards. It's quite possibly the most powerful workstation I've ever built. Running through burn-in and some benchmarking, other than apps that could scale to all those 24 cores / 48 threads, it wasn't any faster than many of my other systems. See, there's a problem here that a lot of people are not taking into consideration... CPU core count vs. speed. The 12-core Xeons sure are awesome... But I truly feel they're clocked too slow to be taken seriously for most workstation setups in terms of utilizing all the cores. If you want the CPUs to run their fastest, and many times you do, many cores have to be deactivated. As more cores are introduced into the process, clock speed has to lower.

For most practical purposes, I think most users will do far better with an 8 core system, maybe a 10 core, that operates at a higher overall clock speed across all cores. It better suits most applications and general workstation activities. This also really makes sense as to why Apple is focusing on a smaller single-CPU system. For the majority of users, it makes the most sense. The new Mac Pro should drop right into the Mac ecosystem quite well if you look at their product line as a whole.

Based on my own workflows and workloads. I could see having a 10-core Mac Pro on my desk driving dual 4K displays. Then having two to four more of them in the equipment room with 12 cores each for rendering and offloaded compute tasks. I really think Apple is on to something here.

 

[Jeff Kilgroe]

I should have posted this one -

The Gigabyte Z87X-UD7 TH is built in the standard ATX form-factor. The board features five PCI-Express 3.0 x16 slots, of which three are gen 3.0 x16 capable, and four gen 3.0 x8 capable. A mechanism lets you bypass the bridge chip, and connect a single graphics card directly to the CPU. Two PCI-Express 2.0 x1 slots complete the expansion area.

When you write it like that, it makes it seem as if there's all sorts of PCIe slots and lanes available. But there's not. If you want to run all 5 PCIe slots, you can only run one of them at X16 and the other four of them at X4. Quad GPU support disables slot 5 and operates the other 4 slots each at X8. Any way you slice it, there are only 32 lanes of available PCIe v3 on this board. And it doesn't look like arbitrary configurations are possible either. Not sure what people use PCIe 2.0 X1 slots for these days. I guess if you need to add an extra USB port or something. Not trying to pick nits or anything, just being depressed over the reality of current PCIe support on the Intel Core platform...

http://www.eteknix.com/computex-gigabyte-show-off-z87x-ud7-thunderbolt-motherboard/[/quote]

and this is coming soon -

http://www.anandtech.com/show/7325/gigabyte-tease-x79-refresh-x79-g1assassin-3

This board looks great! Of course, still the PCIe shortcomings, but a lot of nice features there in a reasonably tight package. I guess I'd like to see a Thunderbolt2 option for both of these, but the reality of it is that doesn't fly in PC land. It's not that people don't want it, but from a product appeal and marketing perspective it's a non-starter. Motherboard makers, and consumers alike, both prefer the thought of "Supports 4 GPUs!!!" over "Hey, we can do 3 GPUs... But we also have this Thunderbolt port! "

...Still wish they would start including it and give us the option to disable it and reclaim some PCIe lanes.

 

[Blair S. Paulsen]

Reading Jeff's notes just reinforces my sense that Apple can exceed spec based expectations for MacPro performance via better utilization of resources. For example, a Titan GPU does not exist in a vacuum, it needs PCIe lanes and software written to exploit the resources available.

I am also keen to see what kinds of tasks could be leveraged in a multi MacPro topology using TB2 I/O - render farm 2.0 if you will. IMO, its easy to reach the point of diminishing returns when shoving more horsepower into a single build as some bottlenecks are inherent. If programs can apportion compute intensive tasks amongst several units efficiently, it should prove faster and more cost effective - plus a lot more flexibility in deployment.

Cheers - #19

 

[paul schefz]

if there wasn't already a thread to thank jeff, this would be one of those moments to start one...