Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Mac Pro 2011 "Sandy Bridge E" everything we know

nanofrog said:
I realized that, and I wish I was able to find that sort of testing (ideally a Tesla 2075 tested in x4, x8, and x16 electrical slots on the same system, running the same benchmark).

But it's not all that different when you look at the fundamentals (instructions + data passed to the GPU, GPU handles the processing, then returns the output). The main difference is the output is returned over the PCIe bus rather than over a graphics port such as DVI. Since the output data tends to be small (i.e. floating point value for each variable), it doesn't consume a lot of additional bandwidth on the PCIe bus compared to sending graphics output to a monitor.
Click to expand...

How much bandwidth was consumed overall?
 
Hellhammer said:
Prices are out: http://www.cpu-world.com//news_2011/2011102701_Prices_of_Xeon_E5-2600-series_CPUs.html

Looks like the lineup will be:

$3499 - E5-2620 (2GHz 12-core)
$4999 - E5-2650 (2GHz 16-core)
$5999 - E5-2665 (2.4GHz 16-core)
Click to expand...

If it were I feel that Apple going with the E5-2665 over the E5-2670 would be a bit insulting when the margin on the machine would be in the 30% range.

I'd prefer to see:

$3499 - E5-2620 (2GHz 12-core)
$4999 - E5-2640 (2.5Hz 12-core)
$5999 - E5-2670 (2.6GHz 16-core)
 
Umbongo said:
If it were I feel that Apple going with the E5-2665 over the E5-2670 would be a bit insulting when the margin on the machine would be in the 30% range.

I'd prefer to see:

$3499 - E5-2620 (2GHz 12-core)
$4999 - E5-2640 (2.5Hz 12-core)
$5999 - E5-2670 (2.6GHz 16-core)
Click to expand...

I would say it's insulting if Apple went with 2640 instead of 2650. $1500 for 500MHz increase in clock speed, whereas another $1000 gets you four extra cores and 100MHz frequency boost.

2665 is just a direct successor of X5670 in terms of price (both are $1440). I wouldn't rule out 2670.

Of course, our guesses are based on Apple keeping the same price points. The CPU prices are a bit different so there might be a change in the lineup as a whole (no need for two SP upgrades for example).
 
Hellhammer said:
I would say it's insulting if Apple went with 2640 instead of 2650. $1500 for 500MHz increase in clock speed, whereas another $1000 gets you four extra cores and 100MHz frequency boost.
Click to expand...

Well I guess this is more all on Intel's lineup and Apple's ideology of limited choice. I just see 12 x 2.5GHz cores being a lot more useful to Mac Pro users that 16 x 2GHz cores. I wouldn't consider anything below 2.5GHz personally, and 2.3GHz and 2.4GHz are pushing it for workstation usage.

Maybe upgrading to two $900 processors from two $400 processors doesn't need to be $1,500 either.
 
beaker7 said:
If that's the case I'll be doing a PC instead.
Click to expand...
Might be time to roll up my sleeves and just built some Xeon + Asus Workstation X79 board monster, because.

I will have to figure out how to make me current Core i5 750 a more portable system. 20 lbs. on the case alone gets tiresome to move.
 
Umbongo said:
Well I guess this is more all on Intel's lineup and Apple's ideology of limited choice. I just see 12 x 2.5GHz cores being a lot more useful to Mac Pro users that 16 x 2GHz cores. I wouldn't consider anything below 2.5GHz personally, and 2.3GHz and 2.4GHz are pushing it for workstation usage.

Maybe upgrading to two $900 processors from two $400 processors doesn't need to be $1,500 either.
Click to expand...

I wouldn't get too hung up on the GHz, especially when setting base lines for what constitutes a workstation. GHz only has meaningful, relative significance between chips of the same family. About 6 years ago Intel acknowledged that it had reached the GHz speed/thermal wall. Then there were faster factory clocked Intel Xeon chips than those for sale today, such as the Socket 604 Xeon chips which ranged from 2.8 to 3.6 GHz. [I still have some 2.8 and 3.0 GHz dual Xeon hyperthreaded (the feature's name differed as I recall) workstations that I'd gladly trade for 1.8 GHz dual CPU Sandy Bridge Xeon systems]. So Intel decided to go the core route. Later, as Intel introduced more on chip features, even a 3.2 GHz 8-core 2008 Mac Pro was made to feel the dominance of the Nehalem 2.9 GHz 8-core 2009 Mac Pro (with its hyper threading/turbo boosting/embedded 1066 MHz memory controller) and then the Nehalem 2.9 GHz 8-core 2009 Mac Pro had to kneel to the greater core count and other features (such as an embedded 1333 MHz memory controller) of the Westmere 2010 2.9 GHz Mac Pro. And so innovations continue with Sandy and Ivy Bridge.

I submit that most will not be disappointed with the workstation performance of a refurb 2.0 GHz 16 core 2012 Mac Pro for the following reasons:

First, a 2.0 GHz Sandy B Mac Pro system is at least 10% faster (on a core per core basis) than a 2.0 GHz Westmere Mac Pro system would have been.

Second, (I recall reading that) most of the Sandy B DP Xeons will have turbo boost bins of 6+. Thus, one would have a Westmere equivalent of, at least, a 2.2 (2.0 + 10% of 2.0) GHz system which turbo boosts to 2.8 GHz (Westmere equivalent).

Third, there will be 16 cores instead of 12 (and 32 logical cores rather than 24).

Fourth, I don't believe that the CPU's will be welded into the system and the same technique that allows Westmere Xeons to be swapped for Nehalem Xeons should allow Ivy Bridge Xeons to be swapped for Sandy Bridge Xeons. Swapping CPU's makes more economic sense with bottom of the line refurbs.

Thus, this lowly 2.0 Ghz (unmodded) should hold its own with the current top of the line Mac Pro for multithreaded apps and will turbo boost to where the current middle of the Mac Pro line does for singular/low threaded apps. That'll be satisfactory workstation performance for many for the price of the current bottom of the Mac Pro line, unless, of course, one were to roll his own. The 2.3 - 2.5 GHz Sandy B Mac Pros (should Apple give such variety) have the potential to crush all of their DP Mac Pro Westmere and earlier ancestors, not to mention those Socket 604 3.6 Ghz speed demons on a core per core basis, but one will have to pay so much for them that it'll be years before CPU swapping doesn't cause trigger the gag reflex. The increased core count, turbo boost potential and memory thru put of Sandy and Ivy Bridge Mac Pros are the new big game changers for those willing to pay Apple's prices and looking for enhanced computational performance in highly threaded apps. But remember as with all things Apple, quarterly they have got to boast about their margins (and that's where the rub lies).
 
Last edited:
Tutor said:
I wouldn't get too hung up on the GHz, especially when setting base lines for what constitutes a workstation. GHz only has meaningful, relative significance between chips of the same family.
Click to expand...


I'm comparing two processors within the same family, we are talking about which would be a better option. I propose that two processors that cost $480 more (each) and are 25% faster in clock speed will be better for most Mac Pro user/buyer's work flows than having 33% extra cores and cost $700 more. Very few creative applications don't benefit from clockspeed. Many components, functions and even some full applications don't benefit from more cores, let alone 16 over 12.

If in all that stuff you posted you are suggesting that 16 2GHz cores will operate as fast or faster when operating with the same cores as the faster clocked, lower cored CPU then perhaps and that would be great, but I wouldn't count on it.
 
Last edited:
From this.

Model -> iMac 2010 -> iMac 2011 Processor Price
21" $1199 -> i3 540 $133 -> i5 2400s $195
21" $1499 -> i3 550 $138 -> i5 2500s $216
27" $1699 -> i3 550 $138 -> i5 2500s $216
27" $1999 -> i5 760 $205 -> i5 2400 $184

Model -> MBP 2010 -> MBP 2011 Processor Price
15" $2199 -> i5 620M $322 -> i7 2720QM $378

I think.

Single Processor
$2499 - E5-1650 (3.2GHz 6-core) 130W
$2899 - E5-1660 (3.3GHz 6-Core) 130W

Dual Processor
$3499 - E5-2630 (2.3GHz 12-core) 95W*2
$4999 - E5-2660 (2.2GHz 16-core) 95W*2
$6199 - E5-2670 (2.6GHz 16-core) 115W*2
 
If we used 2006 schema we will get

Dual Processor
$2199 - E5-2609 (2.4 8-Core) 80W $294*2
$2499 - E5-2630 (2.3 12-Core) 95W $612*2
$3299 - E5-2640 (2.5 12-Core) 95W $884*2
$3999 - E5-2660 (2.2 16-Core) 95W $1329*2

If we used 2008 schema we will get

Single Processor
$2299 - E5-1650 (3.2 6-Core) 130W $583
(2008 quad used 1*E5462@$797)

Dual Processor
$2799 - E5-2630 (2.5 12-Core) 95W $884*2
$3599 - E5-2650 (2 16-Core) 95W $1106*2
$3999 - E5-2660 (2.2 16-Core) 95W $1329*2
 
Tutor Says - Think differently when it comes to clocking newer processors.

Umbongo said:
...
If in all that stuff you posted you are suggesting that 16 2GHz cores will operate as fast or faster when operating with the same cores as the faster clocked, lower cored CPU then perhaps and that would be great, but I wouldn't count on it.
Click to expand...

Exactly my suggestion which I've learned to count on daily for workstation apps, i.e., those multithreaded like Cinema 4d. As the lone wolf in the forest, I'm howling that we embrace a seeming paradox - a pack with many members moving slowly and methodically will seize the prey's entire clan faster than a small pack with greater speed. Underclocking a many-cored system, while positively biasing turbo boost ratios, yields greater speed advantages for less costs than simply goosing up the 24/7 speed of a few. This is in line with the methods I began employing of late and they are embraced fully within the Sandy Bridge E5 line to take advantage of inefficiencies in current processing and coding methods. Note that the result shown below comes from running dual W5680's in my MutliPro at 2.48 GHz and exaggerating their turbo ratios to amount to DDDDEE (or 13,13,13,13,14,14) [shown on every boot screen by verbosing (-v) to see native power management initialization]. I'd love to have 2, or even 4, more slow 2.48 GHz cores on each of my chips, not to mention better memory management.* (N.B. - Also the Geekbench 11.5 score in my sig resulting from this paradoxical clocking technique.) Cross pollination [mac -> hac -> mac -> hac ->mac ~] truly enriches us all.

* Here's some more stuff to chew on and fully digest - Lets do some math:

(1) Geekbench2 - If 40,100 is the yield from 6 cores, then 8cores/6cores [Sandy Bridge advantage] = 1.33 or 10cores/6cores [Ivy Bridge advantage] = 1.67; 1.33 x 40,100 = 53,333 or 1.67 x 40,100 = 66,967, assuming linearity and not accounting for better memory performance of Sandy [then later Ivy] Bridge, and
53,333 / 24,218 [http://www.primatelabs.ca/geekbench/mac-benchmarks/#64bit] = 2.20 times faster than current top of the line, assuming linearity and not accounting for better memory performance of Sandy Bridge
66,967 / 24,218 [http://www.primatelabs.ca/geekbench/mac-benchmarks/#64bit] = 2.76 times faster than current top of the line, assuming linearity and not accounting for better memory performance of Ivy Bridge

(2) Cinebench 11.5 (almost perfect proxy for Cinema 4d) - If 24.70 is the yield from 6 cores, then 8cores/6cores [Sandy Bridge advantage] = 1.33 or 10cores/6cores [Ivy Bridge advantage] = 1.67; 1.33 x 24.70 = 32.85 or 1.67 x 24.70 = 41.25, assuming linearity and not accounting for better memory performance of Sandy [then later Ivy] Bridge, and
32.85 / 15.45 [http://www.cbscores.com/] = 2.13 times faster than current top of the line, assuming linearity and not accounting for better memory performance of Sandy Bridge
41.25 / 15.45 [http://www.cbscores.com/] = 2.67 times faster than current top of the line, assuming linearity and not accounting for better memory performance of Ivy Bridge
 

Attachments

  • cb40k+.png
    cb40k+.png
    82 KB · Views: 88
Last edited:
I know you've answered this before, but is your system a standard Mac Pro with X5680s installed and then settings altered? Is this just EFI modifications that any Mac Pro owner can easily do?

I guess I don't see how what you are doing is any different than over-clocking, but then my knowledge of that is limited. From what you've posted to me it seems like your processors run at 2.48GHz and turbo up very high giving you great geekbench results? How is this different from running at 4GHz+ with speed step enabled? I'd love to see all this explained :)
 
Tutor said:
Exactly my suggestion which I've learned to count on daily for workstation apps, i.e., those multi-threaded like Cinema 4d.
Click to expand...
Your premise/logic is correct, but only under specific circumstances.
  1. The software must be properly multi-threaded before it can utilize those cores.
  2. Most users will use the system bone-stock (no EFI mods to change the CPU's frequencies - this is way over most users' heads, so they won't attempt to experiment).
 
If Apple would just sack up and offer the 2687W or 2690 then there wouldn't be any problem deciding.
 
beaker7 said:
If Apple would just sack up and offer the 2687W or 2690 then there wouldn't be any problem deciding.
Click to expand...

Either of those are basically $2K each. So roughly $4K in processors alone. Keeping it to the 2670 is going to be about $3100 for both processors. At least then you have a chance to maintain roughly the same price levels. I don't think too many people are going to nuts over having a 2.9 clock speed over 2.6 if it means at least an extra $1K after the apple mark up.

Also, I'm a little confused by the complaining about the sacrifice in clock speed vs. cores. That's why we have options people. You can't have both without a large increase in price. Plus, we should be able to pretty comfortably assume about 10% increase in speed with turbo boost. So, if you want a DP system because of the increased RAM and through put for some heavily multi-threaded tasks, also have programs that can only use <6 cores, the 2670 could very well turbo boost up around 2.9 GHz. And with the way these second gen. Sandy Bridges have been going on the low GHz models, with some tubro-ing from ~2 GHz to ~3 GHz, even the 2.0 GHz 2620 could get up close to 3 GHz. I think its just a little early to get all bent out of shape because these clock speeds seem a little low. Plus the memory controller will go up to 1600 MHz, and SATA III ports. The **** will be fast people, calm down.
 
Turbo Boost doesn't only kick in with massively threaded apps.

Umbongo,

I built that system from scratch, personally choosing and assembling every part carefully. First, I tried to overclock it (i.e, I tried to get it to run at 4.6 GHz 24/7 [and I disabled turbo boost, hyper threading and all native power management features, including but not limited to Speedstep, and C,P and T states, as recommended by overclockers]), but Cinebench 11.5 topped out at about 20 to 21 and Geekbench2 topped out at about 29,000 to 30,000. What I got was a system that continuously run warmer than I wanted, requiring that I crank all of my fans up to 100%, adding significant noise and consuming many watts (through continuously high Vcore) all of the time. Then, luckily I happened upon a great thread elsewhere by d00d who advocates that we should make self-built systems behave on a software level like the systems we're trying to emulate (Knowledge of Mac gives to [->] Hac). First, following all of his guidelines I adjusted all of my software exactly as he recommends; then, I went one step further - taking his philosophy's central thesis that less is more I did the counter-intuitive and began to under clock. I rely solely on turbo boosting for the heavy lifting (i.e., I made my system run slower and thus cooler, except when the load triggers turbo boosting) (CPU must have unlocked multi's for this to work correctly if done by enduser). I use hyper threading and every native power management feature available, and set my Vcore below the top end of the range (Spec VID) specified by Intel, running a cooler and thus quieter system and using less electricity. Because I have enforced the environment that turbo boost needs to kick in (i.e., not too hot {I run it slow, thus cooler than normal speed of 3.33 GHz} and not at top end of Spec VID which is 1.35 volts), I get it to turbo boost way past 4.6 GHz whenever it determines the need for heavy lifting. The turbo boost kicks in with apps ranging from Photoshop CS 5.5 and Logic 9 (such as when hoggish filters/plugins are applied), Final Cut Pro 7 and X (such as when rendering and applying filters), Motion (such as when applying effects or pre-vuing multi-layers), Compressor, and Cinema 4d (such as when rendering or manipulating complex models) [and obviously also with benching apps like Geekbeech2 and Cinebench 11.5], but not with apps such as Safari, Mail, and Word. This appears to be the path chosen by Intel for future Xeons, such as the E5's, though they don't exaggerate the turbo boost as much as I do and they have better memory management and thru put than on my system. All in all, I like the direction in which Intel is going and look forward to getting my hands on some Sandy and/or Ivy Bridges. This time, allow knowledge of Hac to give back to (->) Mac. This is what I meant by my statement, "Cross pollination [mac -> hac -> mac -> hac -> mac ~] truly enriches us all."
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back