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

new intel chip! next mac pro maybe?!?

This technology is obviously aimed at the enterprise server (MP) market where systems with 64 cores are already being introduced. You can get something like this next year if you buy a bunch of Becton 8-core CPUs and build an 8 socket system around it. The problem is that such cloud computers are going to be designed in 45 nm technology and that they will suck huge power to do their thing.

The new tech mentioned by the beeb article unfortunately makes no reference to the manufacturing technology level. We can assume that it will be two or three levels advanced from now. The next Mac Pro will have 32 nm Westmere chips with six cores per CPU code named Gulftown. The 64 core chips are probably laboratory samples (potentially not even working) of 22 or 16 nm technology.

Every step takes approximately two years to execute. So if we want to speculate here about applicability we would see this technology arrive at a Mac Pro in 5 years (2014). Each generation of die shrink can add approximately 50% more cores to one CPU. So if we extrapolate to the workstation/DP market in 2015 we should have the potential of 12-16 cores on a typical socket or 24-32 cores in a Mac Pro. If the workstation/DP server market segment is still in existence as we know it now and if Apple is still in the market I might add for caveat.
 
The chip maker, which is slated to unveil six- and eight-core Nehalem chips next year, also noted that it expects to integrate key features of the research chip into a new Core line of commercially available processors by early 2010.
Click to expand...

That is obviously a hint at Becton, which indeed has a completely different architecture compared to the Nehalem Gainestown and Gulftown Xeons. Among other features it uses a central memory router and a ring bus on chip. Memory isn't directly connected to the memory channel but in groups via special control ASICS.
 
gugucom said:
This technology is obviously aimed at the enterprise server (MP) market where systems with 64 cores are already being introduced. You can get something like this next year if you buy a bunch of Becton 8-core CPUs and build an 8 socket system around it. The problem is that such cloud computers are going to be designed in 45 nm technology and that they will suck huge power to do their thing.

The new tech mentioned by the beeb article unfortunately makes no reference to the manufacturing technology level. We can assume that it will be two or three levels advanced from now. The next Mac Pro will have 32 nm Westmere chips with six cores per CPU code named Gulftown. The 64 core chips are probably laboratory samples (potentially not even working) of 22 or 16 nm technology.

Every step takes approximately two years to execute. So if we want to speculate here about applicability we would see this technology arrive at a Mac Pro in 5 years (2014). Each generation of die shrink can add approximately 50% more cores to one CPU. So if we extrapolate to the workstation/DP market in 2015 we should have the potential of 12-16 cores on a typical socket or 24-32 cores in a Mac Pro. If the workstation/DP server market segment is still in existence as we know it now and if Apple is still in the market I might add for caveat.
Click to expand...
Power consumption is 25W - 125W, and according to this article, is actually made with a 45nm process.

Size = 567mm^2 (just under 24mm x 24 mm).
Max RAM = 64GB

So they must have simplified the designs substantially, as they did with the Tera-Scale Polaris Project chip (3.16GHz, 80 core using 65nm process).
 
Nehalem figures for 45 nm CPUs socket 1366:
Die Size 263 mm2
# of Transistors 731 million

So they want 1,3 billion of transistors on a die just two times bigger than 1366 sockets. You don't have to go to a smaller lithographic tech to do this. But the size of the die indicates that we are talking MP as I have said before. To arrive at DP tech there will have to be some die shrinks.
 
gugucom said:
Nehalem figures for 45 nm CPUs socket 1366:
Die Size 263 mm2
# of Transistors 731 million

So they want 1,3 billion of transistors on a die just two times bigger than 1366 sockets. You don't have to go to a smaller lithographic tech to do this. But the size of the die indicates that we are talking MP as I have said before. To arrive at DP tech there will have to be some die shrinks.
Click to expand...
At 48 cores, it's definitely MP. :p I didn't think anyone would thing its DP level parts with such a high core count, as they're aimed at clusters/clouds. So the associated archtecture follows (memory, bus routing,...).

Definitely not a workstation parts, or meant to be, as they are. With a substantial redesign of the memory and bus arctitecture, yes. But without the software, why bother? There's just not enough need yet. If software can catch up, then maybe.
 
Intel does show x-core CPUs for some years during Intel Developer Forum (IDF) events. They already had 80 core and 50 core chips there in the last years. Look for "Tera-Scale" and so on.

Also, search for "Transputer" in some spare time. It´s old but still fancy from a technology point of view. :D


And for the good news: Channelweb got a leaked roadmap showing Westmere-EP 6 core CPUs launching at March 16th, 2010.
This also fits into the update cycle of Apple, so there is a possibility that a new MP will come in March.

Westmere-EP 6core launch March 2010

Nice fact: The Westmere won´t require new Motherboards nor Sockets, so it may be possible to upgrade 2009 MPs.
 
TheSilencer said:
Intel does show x-core CPUs for some years during Intel Developer Forum (IDF) events. They already had 80 core and 50 core chips there in the last years. Look for "Tera-Scale" and so on.

Also, search for "Transputer" in some spare time. It´s old but still fancy from a technology point of view. :D


And for the good news: Channelweb got a leaked roadmap showing Westmere-EP 6 core CPUs launching at March 16th, 2010.
This also fits into the update cycle of Apple, so there is a possibility that a new MP will come in March.

Westmere-EP 6core launch March 2010

Nice fact: The Westmere won´t require new Motherboards nor Sockets, so it may be possible to upgrade 2009 MPs.
Click to expand...

It requires a logicboard firmware upgrade and that is unlikely to be available as there is no incentive for Apple to do it and they haven't in the past.
 
TheSilencer said:
Channelweb got a leaked roadmap showing Westmere-EP 6 core CPUs launching at March 16th, 2010.
This also fits into the update cycle of Apple, so there is a possibility that a new MP will come in March.

Westmere-EP 6core launch March 2010

Nice fact: The Westmere won´t require new Motherboards nor Sockets, so it may be possible to upgrade 2009 MPs.
Click to expand...

Apple usually gets Xeons before the launch, so the new Mac Pro could still launch in February.

Suspiciously absent from the launch are most 3600 CPUs except the 3680 at 3,3 GHz. There were previous reports that Apple will get exclusive use of the 3600 series for some time. This may indicate that the hex core Mac Pros will top out at the 3670.

A surprise are the 56X7 quad CPUs in 32 nm. I did not think that Intel would do them. I think that Apple will not have quad core models but they may offer these octads.

I do not believe in drop in replacements given the previous track record of EFI upgrades.
 
gugucom said:
Apple usually gets Xeons before the launch, so the new Mac Pro could still launch in February.
Click to expand...

Only happened once out of 4 Mac Pro launches and it may not have been an amicable agreement between Intel and Apple. If we go on history it could be anywhere from mid February to mid may. Should start hearing about how Intel's production is going soon, and then pricing will leak and we can make some educated guesses on what the next line up will be like, unless Apple throw a curve ball.
 
Unless I'm a bit mistaken Apple got product early for the 2007 and 2009 Mac Pro ahead of other vendors. 2006 will not count because it was the switch over. So I will stick to my guns and expect Feb. ;)
 
This is nuts!

48 cores on a single chip! You could run a country on that! Uh if only companies spent as much time on software as they do hardware. Powerful programs and games use what? 4CORES! If programs were made that NEEDED even 6 cores it would blow everyones mind!
 
gugucom said:
Unless I'm a bit mistaken Apple got product early for the 2007 and 2009 Mac Pro ahead of other vendors. 2006 will not count because it was the switch over. So I will stick to my guns and expect Feb. ;)
Click to expand...

What appears to have happened with the 2007 Mac Pro was Apple didn't want the confusion of 2.66GHz 8 cores and 3GHz quad cores, so they waited (5 months) for enough 3GHz parts. They then launched a new Mac Pro at their next event. Some smaller companies already were offering them. +- 1 month from the launch date seems a good estimate. Let's not forget long shipping extensions for certain components in the past too (graphics and RAID cards).
 
Those quads really put a monkey wrench into guessing the specs on the new models. The fact that the leak is missing the tray prices makes this especially hard. I wonder if Apple might go dual-socket for the whole line, since there is only one logic board to deal with that way. Thus, they save money on development, production, and repairs by not having to shuffle two daughter cards around like they do now. The quads look like they go pretty cheap, and the e-p3nis that comes with dual-socket boards helps them look nicer than the iMacs. Maybe a 2.13Ghz or 2.4Ghz (or 2.53Ghz) 8-core machine priced just north of $2k for a good entry level model. Has good enough ad copy "$1000 less than the previous 8 core, yet up to 20% more powerful!".

They could either keep it as two lines like now (8 core and 12 core instead of 4 and 8), or merge them back into one and either have multiple stock models like the G5, or just make all the hexa-core chips BTO options like the Clovertown MP. (Which I don't think was technically a "stock" model despite having it's own model ID of MacPro2,1)


I have too many guesses to fantasize now. I need more leaks. :p

EDIT: If the base model is 2.4Ghz, I'm waiting for Sandy Bridge. :p My MBP and my gaming rig already have 2.4Ghz processors. I am not buying ANOTHER 2.4Ghz computer, I need variety in my numbers. :p
 
J the Ninja said:
Those quads really put a monkey wrench into guessing the specs on the new models. The fact that the leak is missing the tray prices makes this especially hard. I wonder if Apple might go dual-socket for the whole line, since there is only one logic board to deal with that way. Thus, they save money on development, production, and repairs by not having to shuffle two daughter cards around like they do now. The quads look like they go pretty cheap, and the e-p3nis that comes with dual-socket boards helps them look nicer than the iMacs.
Click to expand...

While the single and dual socket platforms continue to be the same, expect Apple to have single and dual socket options. Apple aren't going to be able to buy Xeon 5000s for the price Xeon 3000s, and the difference in their price means seperation of the line. They don't need to have a huge disparity between the iMac and a dual socket Mac Pro, at least not when margins on their base Mac Pro are over 50%.
 
Each core in the 48-core chip was weaker than an Atom processor. It wouldn't have the power of even a quad iMac.
 
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