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

Next Mac Pro to be 16-core?

Saladinos

Saladinos

macrumors 68000
Original poster
Feb 26, 2008
1,845
4
Hi,

Looking at Wikipedia's page on Nehalem, I noticed that only one of the processors supports FB-DIMMs - that is, the 8-core model (Beckton). If there are 8 cores in one package, Apple only has a few options for the base model:

1. Keep Harpertown. Increase the clock speed. Keep Nehalem for BTO configurations.
Pros:
  • Keeps the base-model cost down
Cons:
  • Nehalem has a radically different architecture (QuickPath and an Integrated Memory Controller). You can't just plug in a Harpertown. These would be two entirely different systems.


2. Have Intel make a custom Gainestown with FB-DIMM support, and use two of them.
Pros:
  • Both have an IMC and QP, so system architecture would be similar if not identical
Cons:
  • A custom processor would likely be expensive, and Intel might not be able to deliver


3. Use two Beckton processors, to make the base model 16-core.
Pros:
  • Clock speeds can be increased on BTO. Obvious upgrade path.
  • Don't need to rely on custom parts from Intel, at a time when their production is going to be at capacity
  • Increased power in the base model gives good value for buyers
Cons:
  • Could be prohibitively expensive


4. Use a single Beckon processor in the base model, offer upgrade to dual Becktons (16-core)
Pros:
  • Increases the power, without substantially increasing cost
Cons:
  • FSB (QP) could be a performance bottleneck
  • Traditional upgrade is through higher CPU clocks - this would complicate the upgrade path
  • Buyers would be tempted to upgrade the CPU themselves, which is not a user-serviceable part


In response to point 2, Apple could use DDR3 for the base Mac Pro (Wikipedia states modules with up to 16GB capacity, keeping the capacity benefit). I don't think they'll do this, as it would complicate the product range too much. Apple tends to like small upgrades across one product - they wouldn't like to change the CPU and the type of memory used by the machine.

I think option 4 is the most attractive from Apple's perspective. Yes, it would complicate the upgrade scenario somewhat, and bottlenecks may prevent the performance enhancements many are expecting, but it's the smartest business option to go for. Of course, if they can afford to cut their margins a little, option 3 would be most favourable to buyers.

I know we say this with every revision, but the amount of internal redesign that Nehalem brings could tempt Apple to perform some external design changes.
 
while 16 cores would be heaven, more than likely processors like this would have a low clock rate about 2 - 2.5 ghz , apple dont want to give up the 3 ghz title
 
Apple don't care about a 3Ghz base model - the current one is only 2.8Ghz. They're emphasising the number of cores, which, with Snow Leopard, should become more significant. 2-2.5 GHz is far too low - Intel won't make a Nehalem chip that low.

Scrapping FB-DIMMs is another alternative. One that I'd rather like to see. The problem with that is that you'd have to get rid of the 8 memory slots (I think it's 8). The intel controller likely won't support more than 4, which makes upgrading your RAM more expensive, as you'd need to buy pairs of higher capacity memory and throw out your old ones. Actually, that does kind of sound like Apple.

The number of memory slots is one of the best things about the MP. For scientific applications, capacity can take precedence over speed (to an extent, obviously).
 
Option 2 doesnt convince me at all...

I doubt intel would custom make a gainestown variant for the mac pro JUST for fb-dimm... wouldnt make any sense... also, i'm pretty sure intel is busy as hell pushing most of their staff into the nehalem project and getting it done ON TIME.
 
Guys...come on..they're going to use Gainstown.

clovertown/Harpertown/Gainstown

well, I obviously don't know. but seems Gainstown is the DP and Beckton is MP and MPs are known to use DP. and using Gainstown the MP would stay at 8 cores sure, but 16 threads.

But of course..I can see them using Becktown too.
 
Apple doesn't need FB-DIMMs since Gainestown supports ECC DDR-3 memory. We may see a Beckton Mac Pro down the road, but the first models will almost assuredly be dual-CPU units using four-core/eight-thread Gainestown CPUs.
 
Expect 8 Cores w/ 16 Threads from the Gainestown Platform.

Also don't be surprised to see 12 RAM Slots, for up to 48GB of Ram.

For some sort of idea about performance, here is a conclusion from Anandtech.

"The 4-core, 8-thread Nehalem posted performance results that were within 20% of a current top-of-the-line 3.2GHz 8-core SkullTrail system. This suggests that clock-for-clock, that Nehalem's 4-cores with 8-threads is comparable to a pair of Penryn's with 8-cores."

Tracer
 
Get rid of FB-DIMM support, problem solved. Its a dead technology and nobody wants it anymore.

I don't understand the irrational thinking that Apple MUST use FB-DIMMs in the next MacPro. They don't need it.
 
As for your BOT post, your right, they never would.. but wouldnt it be absolutely wonderful if they did?
 
Indiochano: I know Opt. 2 is unlikely. I was trying to list all possibilities. I noted in Opt 3 that using custom parts when Intel's going to be working close to capacity is not realistic.

Apple wouldn't ship it with open sockets, but then, there's no reason they can't leave sockets out of the motherboard design. You don't have to have a physical socket for every connection the chipset supports - for example, many laptop chipsets can support additional SATA drives and USB ports - they're just not implemented.

There are some advantages to keeping FB-DIMMs - capacity and ease of upgrading it is probably the most obvious. Besides, I don't expect Apple to do a U-Turn and start using DDR3 unless they have to. Not only is it bad for consistency, it's harder for support and upgrades.

Apple have pig-headedly defended the PPC architecture in the past, despite the engineering challenges; either they've learned, and will adopt DDR3, or they haven't, and they'll keep FB-DIMMs, leading to price/performance tradeoffs as above.
 
Saladinos said:
Indiochano: I know Opt. 2 is unlikely. I was trying to list all possibilities. I noted in Opt 3 that using custom parts when Intel's going to be working close to capacity is not realistic.

Apple wouldn't ship it with open sockets, but then, there's no reason they can't leave sockets out of the motherboard design. You don't have to have a physical socket for every connection the chipset supports - for example, many laptop chipsets can support additional SATA drives and USB ports - they're just not implemented.

There are some advantages to keeping FB-DIMMs - capacity and ease of upgrading it is probably the most obvious. Besides, I don't expect Apple to do a U-Turn and start using DDR3 unless they have to. Not only is it bad for consistency, it's harder for support and upgrades.

Apple have pig-headedly defended the PPC architecture in the past, despite the engineering challenges; either they've learned, and will adopt DDR3, or they haven't, and they'll keep FB-DIMMs, leading to price/performance tradeoffs as above.
Click to expand...

You are assuming Apple have any sort of investment in FB-DIMMs. I can't imagine they do. Apple have shown they aren't interest in being a memory vendor.

By the way, two Beckton processors and a systemboard are likely to start at around $4,000 ($4,500 if using a quad socket board), where as two Gainstown processors that will outperform them and a systemboard are likely to be under $2,000.
 
I don't see why FB-DIMMS are so bad. I think the only reason apple would move to 16 cores is if 1. they are readily available 2. Other Manufacturers are going to be using them and 3. they are reasonably priced.
 
judging by the introduction of grand central, im pretty sure apple and intel have been working hand in hand for the past year or so and should have a finished product by the end of the year. hence, since gainestown will be the first to hit the market, id go for that one instead of beckton... also, beckton's price tag will obviously be way higher than gainestown.
 
indiochano said:
judging by the introduction of grand central, im pretty sure apple and intel have been working hand in hand for the past year or so and should have a finished product by the end of the year. hence, since gainestown will be the first to hit the market, id go for that one instead of beckton... also, beckton's price tag will obviously be way higher than gainestown.
Click to expand...

If Beckton is a server part, it's possible that's the one Apple will be using. Also, since Beckton's IMC only supports FB-DIMMs, demand is likely to be much less than supply. An integrated memory controller introduces those sorts of pricing issues. Beckton's price could be lowered as a result.
 
I forsee a Gainestown Mac Pro and a Beckton Xserve.

But wait, isn't one of the advantages to FB-DIMM the maximum RAM, or does Nehalem/DDR3 get rid of that?
 
Saladinos said:
The number of memory slots is one of the best things about the MP. For scientific applications, capacity can take precedence over speed (to an extent, obviously).
Click to expand...

And it was one of the best forward moves in the G5 as well (everything but the base single 1.8gHz model had 8 slots). I mean, considering they had to be matched pairs, it's really just like having 4 slots and larger chips, but in terms of affordability of incremental upgrade, it's much much better.
 
ZeelessOne said:
I forsee a Gainestown Mac Pro and a Beckton Xserve.

But wait, isn't one of the advantages to FB-DIMM the maximum RAM, or does Nehalem/DDR3 get rid of that?
Click to expand...

Xserve will be Gainstown too.

FB-DIMMs are good for supporting very large amounts of memory, but I wouldn't think the move to registered DDR3 modules when Nehalem comes is something Apple or the vast majority of Mac Pro users should be concerned over. Nehalem Mac Pros will probably have 12 memory slots and 4GB modules should be available, probably 8GB too.
 
Do your homework

Umm, actually with Beckton, there'll be a four-socket configuration on the top end.

Beckton is a Nehalem-based processor with eight cores and uses buffering inside the chipset to support up to 16 standard DDR3 DIMMS per CPU socket without requiring the use of FB-DIMMS.[20] It has four (or more?) QuickPath interfaces, so it can be used in at least quad-socket configurations. It is expected to be launched in Q1 2010.[21]

Beckton is the code name for an 8 core processor based on Nehalem. It is also referred to as Nehalem-EX (EXpandable server market). Beckton is being designed by the Digital Enterprise Group (DEG) Santa Clara Xeon Design Team. Beckton will be manufactured on the P1266 (45 nm) technology.

Snow Leopard's press release makes a big deal about Grand Central and Open CL.

Both of these efforts are at consolidating the OSX environment to consider its resources as ubiquitous commodities, GPU and CPU can be used interchangeably.

This will mean that whatever your load, you get aggregate performance from the system.

Currently, unless the software is built multi-threaded, then you don't get that much extra performance from multiple cores.

With Snow Leopard, apple will have a central system scheduler which dispatches instructions fluidly to both GPU and CPU, meaning that you can just add GPU and CPU power in chunks, and get the performance no matter how the software is written (more likely... if the software is built out of the newest cocoa system libraries. Photoshop... naughty naughty.)

Still, bottom line is that if apple are building a top-end workstation, it'll most likely feature 4x8 cores, and appear Q1 or Q2 2010.

That's gonna be a brutal mac pro. Maybe a case redesign. Expect the mac pro to be the first recipient of Beckton, along with the Xserve.

Maaaaaybe... if my hunch serves me correctly...

Apple in Q2/3 release a new Xserve based architecture of stackable compute clusters, 16 cores in a 1u case, and make the iMac a GPU and Dual-core blind terminal, based on a thin client, cloud, or one might even suggest mainframe type model. Return to NIX style. Iphones at the ready! You buy imacs and Xserves, and connect them with some sort of switched fabric. Xgrid meets Snow Leopard.

Look out for Quicktime X actually being a desktop OS-grade web plugin, allowing VNC type access in the browser. A future competitor to flash (hence no flash on the iphone, yet quartz composer with OpenGL works a treat.)
 
bishopdante said:
Still, bottom line is that if apple are building a top-end workstation, it'll most likely feature 4x8 cores, and appear Q1 or Q2 2010.

That's gonna be a brutal mac pro. Maybe a case redesign. Expect the mac pro to be the first recipient of Beckton, along with the Xserve.
Click to expand...

Not going to happen. Beckton is not for workstations, it uses FB-DIMMs, there is no room in the current Mac Pro case, a 32 core system at 2.66Ghz will likely run $20,000. You've miss understood the intended uses of the Nehalem EX platform.

It could go in to Xserves but that would require a major shift in how Apple view their server business and I doubt that will happen either.
 
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