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

dhazeghi

macrumors member
Original poster
Nov 2, 2006
89
25
One of the neat things about Haswell-E is that it brings down prices on higher-core-count CPUs considerably. The cheapest reasonably-clocked 8-core Intel CPU (the one in the 8-core nMP) had a TRAY price of $1700. Haswell-E brings that down to $1000. 6-core drops by a similar percentage ($580 to $390).

So what I'm wondering is whether these changes in pricing mean the next generation nMP will offer significantly better bang-for-the-buck? Today Apple charges $2k to upgrade from 4 to 8 cores, but it seems like that'll be a lot harder to justify going forward.
 
Could come down to $1,200 for the 8-core upgrade, but that's about it. Note there is a 3.2GHz 8-core part, that replaces the 8-core used in the Mac Pro that is more expensive that the 3GHz parts.

There is no cheap 6-core E5 Xeon like the i7-5820K, also if there was it doesn't have enough PCI-E lanes, with 28 instead of 40. Expect 4 core, 6 core same as now, 8-core for $1,200 and maybe a 14 core.
 
Last edited:
One of the neat things about Haswell-E is that it brings down prices on higher-core-count CPUs considerably. The cheapest reasonably-clocked 8-core Intel CPU (the one in the 8-core nMP) had a TRAY price of $1700. Haswell-E brings that down to $1000. 6-core drops by a similar percentage ($580 to $390).

So what I'm wondering is whether these changes in pricing mean the next generation nMP will offer significantly better bang-for-the-buck? Today Apple charges $2k to upgrade from 4 to 8 cores, but it seems like that'll be a lot harder to justify going forward.

Desktop CPUs are different from Xeon CPUs, in both price and features.
Haswell Xeons are not going to cost much less than Ivy Bridge Xeons(comparing similar products).
 
Last edited:
The pricing for the 2600 v3 series (used on 8/12 core) has been circulating for awhile and generally seems to be the same as v2 part-for-part, but with two more cores at lower clocks (and higher TDP). So Apple could maintain the same pricing but offer more cores like 10/14 cores (at the expense of single core performance) or choose to keep core count offerings the same and perhaps move all but the top end BTO to the 1600 v3 series. I haven't seen pricing on those yet.

Comparison between 2600 v2 and v3... http://www.cpu-world.com/news_2014/2014080502_Xeon_E5-2600_v3_CPUs_are_available_for_pre-order.html
 
The pricing for the 2600 v3 series (used on 8/12 core) has been circulating for awhile and generally seems to be the same as v2 part-for-part, but with two more cores at lower clocks (and higher TDP). So Apple could maintain the same pricing but offer more cores like 10/14 cores (at the expense of single core performance) or choose to keep core count offerings the same and perhaps move all but the top end BTO to the 1600 v3 series. I haven't seen pricing on those yet.

Comparison between 2600 v2 and v3... http://www.cpu-world.com/news_2014/2014080502_Xeon_E5-2600_v3_CPUs_are_available_for_pre-order.html

The 8-core part is the E5-1680 V2.

Pricing for the 3.7GHz 4-core and 3.5GHz are the same, they introduce a 3.0GHz 8-core for $1,000 and the E5-1680 V3 drops in price a bit.
 
The 8-core part is the E5-1680 V2.

Ok, I thought it might have been from the 2600 series.

Then the 4/6/8 configs will likely use these parts (no word on pricing)...
http://www.cpu-world.com/news_2014/2014083101_Some_details_on_Xeon_E5-1600_v3_CPUs.html

----------

Pricing for the 3.7GHz 4-core and 3.5GHz are the same, they introduce a 3.0GHz 8-core for $1,000 and the E5-1680 V3 drops in price a bit.

Do you have a link?

EDIT... Found this... http://www.chiploco.com/intel-haswell-ep-e5-1600-v3-35072/
 
Could come down to $1,200 for the 8-core upgrade, but that's about it. Note there is a 3.2GHz 8-core part, that replaces the 8-core used in the Mac Pro that is more expensive that the 3GHz parts.

There is no cheap 6-core E5 Xeon like the i7-5820K,

That's too bad. The 1620v2 and the 4820k are basically the same CPU. I'd hoped they would have continued the progression with Haswell-E.

also if there was it doesn't have enough PCI-E lanes, with 28 instead of 40. Expect 4 core, 6 core same as now, 8-core for $1,200 and maybe a 14 core.

I suppose Apple does things there own way, but an 8-core option for $1200 would be interesting when OWC could put together an equivalent upgrade kit for $1000 or less.

Desktop CPUs are different from Xeon CPUs, in both price and features.

That's really not the case. The 'E' line of Core i7 chips have always been re-branded Xeons with unlocked clock multipliers and a few business features disabled. Branding aside, they are about as close as these things get. At least in the past, pricing has also been quite comparable.
 
I expect the $2999 model to have 6 cores. Whether they offer a 4 core for less than that or cut it from the line up is up in the air.
 
I expect the $2999 model to have 6 cores. Whether they offer a 4 core for less than that or cut it from the line up is up in the air.

Seriously, the base 2999 should have six. But even at current prices, they are selling well, so I can't see that happening soon.
 
One of the neat things about Haswell-E is that it brings down prices on higher-core-count CPUs considerably. The cheapest reasonably-clocked 8-core Intel CPU (the one in the 8-core nMP) had a TRAY price of $1700. Haswell-E brings that down to $1000. 6-core drops by a similar percentage ($580 to $390).

So what I'm wondering is whether these changes in pricing mean the next generation nMP will offer significantly better bang-for-the-buck? Today Apple charges $2k to upgrade from 4 to 8 cores, but it seems like that'll be a lot harder to justify going forward.

For the most part. I wouldn't expect any significant change in prices or performance differences for most OSX apps, except (1) the Haswell 4-core may either be a tad slower (E5-1620) or a tad more expensive (E5-1630); the 8-core Haswell (E5-1680) might be a tad faster; but if Apple goes with the 8-core Haswell (E5-1660), then there's lots of room for a potential price decrease there w/o loss in performance; and the Haswell 14-core (E5-2697) will be a faster in certain high core applications, than the Ivy Bridge 12-core (E5-2697), but the V3 will cost a little more. See my next reply herein.

Ok, I thought it might have been from the 2600 series.

Then the 4/6/8 configs will likely use these parts (no word on pricing)...
http://www.cpu-world.com/news_2014/2014083101_Some_details_on_Xeon_E5-1600_v3_CPUs.html... .

Here are CPU-World's listing of some introductory prices. All, but one - the 1660 V3, are on links on the upper right hand side of page that you cited. I just clicked on each of the individual listings to help make this chart. I pulled the info for the Xeon E5-1660 v3 by clicking on it's listing at the bottom of the page in one of the other E5-1600 V3 CPU pages. Usually, at the bottom of each page of an individual cpu description is a processor family comparison.

Xeon E5-1620 v3 4 / 8 3.5 GHz 10 MB 140 Watt - $294 vs. v2 (4 cores/3.7 GHz/10MB/130W - $294)
Xeon E5-1630 v3 4 / 8 3.7 GHz 10 MB 140 Watt - $372 vs. v2 (None)
Xeon E5-1650 v3 6 / 12 3.5 GHz 15 MB 140 Watt - $583 vs. v2 (6 cores/3.5 GHz/12MB/130W - $583)
Xeon E5-1660 v3 8 / 16 3.0 GHz 20 MB 140 Watt - $1080 vs. v2 (6 cores/3.7 GHz/15MB/130W - $1080)
Xeon E5-1680 v3 8 / 16 3.2 GHz 20 MB 140 Watt - $1723 vs. v2 (8 cores/3.0 GHz/25MB/130W - $1723)
Xeon E5-2697 v3 14/28 2.6 GHz 35 MB 145 Watt - $2702 vs. v2 (12 cores/2.7 GHz/30MB/130W - $2614)

Sales prices are, however, usually a tad higher than introductory pricing until a year or two passes.

Second Thought - Apple would not risk being banned if it offered the following options:

Xeon E5-2698 v3 16 / 32 2.3 GHz 3.6 GHz 40 MB 2 135W - $3226 (before Apple levy) and
Xeon E5-2699 v3 18 / 36 2.3 GHz 3.6 GHz 45 MB 2 145W - $4150 (before Apple levy).
 
Last edited:
Sales prices are, however, usually a tad higher than introductory pricing until a year or two passes.

These days you can find lower than Intel's pricing pretty quick. Superbiiz are an example of a company that have great prices on Xeon, R and LR DIMMs, Quadros and FirePro cards.
 
For the most part. I wouldn't expect any significant change in prices or performance differences for most OSX apps, except (1) the Haswell 4-core may either be a tad slower (E5-1620) or a tad more expensive (E5-1630); the 8-core Haswell (E5-1680) might be a tad faster; but if Apple goes with the 8-core Haswell (E5-1660), then there's lots of room for a potential price decrease there w/o loss in performance; and the Haswell 14-core (E5-2697) will be a faster in certain high core applications, than the Ivy Bridge 12-core (E5-2697), but the V3 will cost a little more. See my next reply herein.



Here are CPU-World's listing of some introductory prices. All, but one - the 1660 V3, are on links on the upper right hand side of page that you cited. I just clicked on each of the individual listings to help make this chart. I pulled the info for the Xeon E5-1660 v3 by clicking on it's listing at the bottom of the page in one of the other E5-1600 V3 CPU pages. Usually, at the bottom of each page of an individual cpu description is a processor family comparison.

Xeon E5-1620 v3 4 / 8 3.5 GHz 10 MB 140 Watt - $294 vs. v2 (4 cores/3.7 GHz/10MB/130W - $294)
Xeon E5-1630 v3 4 / 8 3.7 GHz 10 MB 140 Watt - $372 vs. v2 (None)
Xeon E5-1650 v3 6 / 12 3.5 GHz 15 MB 140 Watt - $583 vs. v2 (6 cores/3.5 GHz/12MB/130W - $583)
Xeon E5-1660 v3 8 / 16 3.0 GHz 20 MB 140 Watt - $1080 vs. v2 (6 cores/3.7 GHz/15MB/130W - $1080)
Xeon E5-1680 v3 8 / 16 3.2 GHz 20 MB 140 Watt - $1723 vs. v2 (8 cores/3.0 GHz/25MB/130W - $1723)
Xeon E5-2697 v3 14/28 2.6 GHz 35 MB 145 Watt - $2702 vs. v2 (12 cores/2.7 GHz/30MB/130W - $2614)

Sales prices are, however, usually a tad higher than introductory pricing until a year or two passes.

Second Thought - Apple would not risk being banned if it offered the following options:

Xeon E5-2698 v3 16 / 32 2.3 GHz 3.6 GHz 40 MB 2 135W - $3226 (before Apple levy) and
Xeon E5-2699 v3 18 / 36 2.3 GHz 3.6 GHz 45 MB 2 145W - $4150 (before Apple levy).

Great summary. If Apple does a Haswell update, clearly there will be no change in price on the 4/6/8 core. And if they offer either of those last two on the high-end, they will set a new record for "Apple Tax" ;)

I'm still skeptical of a Haswell update... No price savings, barely any performance gains in creative apps, and apparently they run hotter. :confused:
 
Great summary. If Apple does a Haswell update, clearly there will be no change in price on the 4/6/8 core. And if they offer either of those last two on the high-end, they will set a new record for "Apple Tax" ;)

I'm still skeptical of a Haswell update... No price savings, barely any performance gains in creative apps, and apparently they run hotter. :confused:

The one price saving can come from a cheaper 8-core, not that I expect it. As we've gone back and forth over before though, unless Apple come out and say "yeah not worth the update", which will never happen, then it looks like they really don't care for Mac Pro users and they get people waiting for the update and not buying current models too.
 
Seriously, the base 2999 should have six. But even at current prices, they are selling well, so I can't see that happening soon.

They could take the opportunity of a refresh to make a 4C entry-level for $2499 or so.
 
For the most part. I wouldn't expect any significant change in prices or performance differences for most OSX apps, except (1) the Haswell 4-core may either be a tad slower (E5-1620) or a tad more expensive (E5-1630); the 8-core Haswell (E5-1680) might be a tad faster;
....
Here are CPU-World's listing of some introductory prices. ...

Xeon E5-1620 v3 4 / 8 3.5 GHz 10 MB 140 Watt - $294 vs. v2 (4 cores/3.7 GHz/10MB/130W - $294)
Xeon E5-1630 v3 4 / 8 3.7 GHz 10 MB 140 Watt - $372 vs. v2 (None)
Xeon E5-1650 v3 6 / 12 3.5 GHz 15 MB 140 Watt - $583 vs. v2 (6 cores/3.5 GHz/12MB/130W - $583)
Xeon E5-1660 v3 8 / 16 3.0 GHz 20 MB 140 Watt - $1080 vs. v2 (6 cores/3.7 GHz/15MB/130W - $1080)
Xeon E5-1680 v3 8 / 16 3.2 GHz 20 MB 140 Watt - $1723 vs. v2 (8 cores/3.0 GHz/25MB/130W - $1723)
Xeon E5-2697 v3 14/28 2.6 GHz 35 MB 145 Watt - $2702 vs. v2 (12 cores/2.7 GHz/30MB/130W - $2614)


The 1620-1630 being potentially slower looks more like a factor of Intel having no competition than some technological flaw in Haswell. The 4 core models look to simply be kneecapped on Turbo. Lower turbo rates on 4 & 6 v3 core models than v2 ? Smacks of using the IPC improvements to be offset by clock reduction to keep performance the same. Most likely while the TDP cap hasn't moved ( jump by merging voltage regulation in ... that's is just a system TDP ballon squeeze) they will probably generally operate even lower than before for same (or better ) performance.

TDP across a range of plug-in compatible models makes for one design to fit multiple configurations. It isn't that every one of those configurations run at the exactly the same temperature.



compare ______ base/turbo(average) drop [ if 10% IPC improvement with v3 ]
1620 v3 to v2 _____ 3.5/3.6(3.55) <- 3.7/3.9(3.8) [ -6.6% drop in average. stlll up 3.4% with 10% bump]

1630 v3 -- 1620 v2 3.7/3.8(3.6) <- 3.7/3.9(3.8) [ -5.2% drop in avg. still up 4.8% on bump ]

1650 v3 -- 1660 v2 3.5/3.8(3.65) <- 3.7/4.0(3.85) [ -5.2% drop in avg still up 4.8% on bump ]

1660 v3 -- 1680 v2 3/3.5(3.25) <- 3/3.9(3.45) [ -5.8% drop in avg still up 4.2% on bump]

1680v3 -- 1680 v2 3.2/3.8(3.5) <- 3/3.9(3.45) [ -1.4% drop in avg still up 8.6% on bump ]

For 4 core Intel set the base even lower and split range with two products at different price points. A 100Mhz wide base-turbo zone seems more indicative of market segmentation than of technology. 6 core narrowed to just one model at about the performance top end on mixed workloads. Almost zero competitive pressure from AMD and milking the margins. v4 (Broadwell) isn't going to particularly make that any different.



Sales prices are, however, usually a tad higher than introductory pricing until a year or two passes.

What Apple is paying for their volume purchase contract is not what these are but Intel's standard pricing is far more useful when comparing across generations and what Apple's price targets are that likely aren't changing across generation.

Pricing that sags over time because of bloated inventory are highly unlikely to impact Apple at all since they are highly likely not going to keep bloated inventory long term.


Second Thought - Apple would not risk being banned if it offered the following options:

The likely approximately 30% "apple tax" is going to put those is the "I have money to throw away zone". If Apple does the round-robin memory allocation (to even out NUMA hiccups but tax everything with the overhead) and non-OS aware resource allocation, that is alot of money for something that the firmware/OS/infrastructure is going to partially piss away.

I'm sure there would be some buyers who are just purely chasing corer counts but "max core count" isn't effective if don't match up infrastructure to support it.

----------

They could take the opportunity of a refresh to make a 4C entry-level for $2499 or so.

Only if they gutted the GPUs. The CPU costs are about the same. Memory costs are incrementally up. SSD and chipset costs are probably roughly the same. Where is this the ~$380 in hardware+software component savings coming from??? Discrete USB 3.0 chip is at most $10-25. That's an order of magnitude off.

Frankly the $2999 price only came with gimped VRAM ( only 2GB versus W7000's 4GB ) and RAM ( only 12GB DDR3 ).

Similar with making 6 core minimal entry point. The extra $200+ cost from from other components is coming from where?
 
I'm still skeptical of a Haswell update... No price savings,

Like Broadwell is going to be cheaper. Has Intel made the Xeon E5 (previously 3x000/5x000) equivalents substantially cheaper over time? Not really, The 4 core entry point has been around $300 since 2008 (at least ). That's back when AMD was trying to competitive.

barely any performance gains in creative apps,

Software ignoring the instruction set improvements is about as much a contributing factor there as the new hardware. This is also only myopically on a CPU dimension also. There is more than just a CPU package in a Mac Pro.

Intel chopping the clocks and doing even deeper market segmentation isn't going to get any better on v4 (Broadwell).

and apparently they run hotter. :confused:

Still confused even after pointed out that Haswell moved some voltage regulation onto the chip? Every Haswell model desktop/laptop/etc has an incrementally higher TDP. The associated chipsets all went down in TDP (since moved up to better process among other factors ). The voltage regulation is present on the older systems also. It is simply a system wide ballon squeeze to move it from one location on the logic board to the CPU package. Overall system TDP went down across the board on the other Mac Haswell updates. It would likely go down with E5 v3 upgrade also.

Frankly, on chip the TDP contribution from voltage regulation probably more cleanly gets feed into the Mac Pro's central thermal core. The operational thermal clearance should be even better.

This whole FUD card on TDP problems doesn't any technology merit to it.

v4 Broadwell isn't going to get any better.

image.php

Intel slide from xbitlabs article on 18 core Broadwell EP roadmap. (http://www.xbitlabs.com/news/cpu/di...icroprocessors_for_Launch_in_2015_Report.html )


The TDP line in the chart? Matched the same for both tock/tick versions. The differences are highlighted in gray. Waiting for v4 isn't going to bring any strategic/tactical benefits for TDP. It is just purely a waste of time.
Broadwell will have an incrementally bigger operational enveloped under a wider variety of workloads; nothing revolutionary.

Same article on performance
image.php

Sure v4 is incremntally higher than Haswell and Ivy Bridge, but the real issue is that the older Mac Pros are at least as far backwards from v3 as v2 is from v4. The object isn't to sell Mac Pros to folks who just bought them 12 months ago. The primary objective should be to sell to folks who did not move in last 12 months.
 
Last edited:
Still confused even after pointed out that Haswell moved some voltage regulation onto the chip? Every Haswell model desktop/laptop/etc has an incrementally higher TDP. The associated chipsets all went down in TDP (since moved up to better process among other factors ). The voltage regulation is present on the older systems also. It is simply a system wide ballon squeeze to move it from one location on the logic board to the CPU package. Overall system TDP went down across the board on the other Mac Haswell updates. It would likely go down with E5 v3 upgrade also.

Frankly, on chip the TDP contribution from voltage regulation probably more cleanly gets feed into the Mac Pro's central thermal core. The operational thermal clearance should be even better.

This whole FUD card on TDP problems doesn't any technology merit to it.

Thanks to the VR on chip, it's widely reported that desktop Haswells are running 15-deg hotter than Ivy. This is heat that, as you point out, is now central to the CPU and it's HSF which means it translates directly into higher fan speeds or higher CPU temperatures, or both. That's not FUD just fact. Now you may argue that the nMP Thermal core is up to the task by a large margin, but we all know that it's not limitless. A pair of W9100 equivalent GPUs that are also reported to be running hot (by 15-deg) combined with a hotter running Haswell CPU may very well have a negative impact on noise or thermal limiting conditions. Or require so much down clocking on the GPUs as to erase any performance benefits.

I could imagine Apple engineers putting together a couple of test systems and concluding it's not worth it... Hotter/noisier and offers no appreciable benchmark or real-world application performance so why bother doing a rev? They may choose to wait for Broadwell and 20nm next-gen GPUs instead. I'm not saying that will happen, but it's possible. We will know soon I suppose.
 
I don't think their profit margins on the Mac Pro are that high. It'd probably take at least 2.

They are higher than they were previously, which was a big part of moving it to the US - where the majority of sales are. So yeah its around that number. Actual hardware costs are like half of the retail price.
 
Thanks to the VR on chip, it's widely reported that desktop Haswells are running 15-deg hotter than Ivy. This is heat that, as you point out, is now central to the CPU and it's HSF which means it translates directly into higher fan speeds or higher CPU temperatures, or both. That's not FUD just fact. Now you may argue that the nMP Thermal core is up to the task by a large margin, but we all know that it's not limitless. A pair of W9100 equivalent GPUs that are also reported to be running hot (by 15-deg) combined with a hotter running Haswell CPU may very well have a negative impact on noise or thermal limiting conditions. Or require so much down clocking on the GPUs as to erase any performance benefits.

I could imagine Apple engineers putting together a couple of test systems and concluding it's not worth it... Hotter/noisier and offers no appreciable benchmark or real-world application performance so why bother doing a rev? They may choose to wait for Broadwell and 20nm next-gen GPUs instead. I'm not saying that will happen, but it's possible. We will know soon I suppose.

This is exactly how Apple typically opperat(ed)es. Throw a prototype together a lot of ways, but if it sucks or isn't enough of an improvement - do nothing.

***Why do you think the 2009 Mac Pro stayed on unchanged for so long?*** (new MP performance isn't any better, that's why.)
 
What have you been smoking? Since when has apple passed savings onto their dominions who worship apple? Reason why I built my own dual system. Don't get me wrong. Enjoyed OSX but the hardware to run it is way to much.
 
New Info on Haswell-E with DDR3 options!

Hi Folks,

I am eagerly waiting for the 21 Oktober event, but so far my hope for an updated nMP with a minor speed bump through Haswell-E V3 was low.

However, I found this article today. And now I am more then ever thrilled about Haswell-E.
In my eyes it basically gives Apple the chance to really reduce the development workload for that update. Theoretically they would not even require to change the memory architecture for an update. Just put the new CPU with a new socket in, keep all the rest of it and that's pretty much it.

http://www.anandtech.com/show/8536/intels-haswellep-xeons-with-ddr3-and-ddr4-on-the-horizon

It would be minor, but better then nothing! I mean -do they want us to wait another 4 years for the next speed bump again?

What do you think?
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.