Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Status
Not open for further replies.
I've been lurking here keeping up with this thread as often as possible, and although I haven't read every page, I have some thoughts on what we might expect from a Nehalem DP Mac Pro.

First let me say that perhaps this thread is getting too long? Can we not start posting threads in the main forum on different subjects rather than try to put everything in here?

Anyway, excuse me if this has already been discussed to a logical conclusion, but I notice in the first post, that current speculation has dual risers with 12 slots for ECC based DDR3 memory.

Based on what I know, riser based RAM was unique to FBDIMM technology. Risers cannot be used on regular memory such as DDR2 and DDR3 where each module has a physical connection to the memory controller. Trace length plays an important role in DDR memory performance and trace lengths must be maintained as equal as possible to avoid phase issues. While DDR3 technology compensates for this somewhat, I still don't think DDR3 could ever support something like a riser.

If this holds true, then I believe we will see memory directly on the motherboard. The Nehalem-EP dual processor architecture offers 3 DDR3 channels per processor with each channel supporting up to 3 physical DIMMS. That's a maximum of 9 DIMM slots per processor or 18 for a dual socket system. This is supported by Intel's roadmap slide and existing prototype boards that are floating around.

1002xeon_05.jpg


Supermicro.jpg


Another intersting fact is that early Nehalem-EP systems appear to support non-ECC memory which is good news for desktop users as it should improve performance (slightly) while reducing cost and increasing choice. (Link)

On the subject of I/O, I assume Apple will adopt Intel's ICH10 southbridge chip for I/O which will provide the following capabilities, although there's no guarantee that Apple won't ignore some of this capability or supplement it with additional capability.

small
 
Absolutely, but I don't think that Tylersburg lets you have two 16x and two 8x.

I could be wrong. I was under the impression that it could give four 16x or two 16x and two 4x, with the latter being the most logical choice of Apple.

I don't know. Someone step in.

Two PCI-E 2.0 x16 and one PCI-E 2.0 x4 per IOH (actually 36 lanes however is needed). The chipset can support two IOHs. The southbridge supports six PCI-E 1.1 lanes that can be configured as one x4 and two x1.

The current Mac Pro has two PCI-E 2.0 x16 and two PCI-E 1.0 x4 slots.
 
First let me say that perhaps this thread is getting too long? Can we not start posting threads in the main forum on different subjects rather than try to put everything in here?
The problem is that then we end up with the same discussion happening in many threads. Neither is really an Ideal solution.
Based on what I know, riser based RAM was unique to FBDIMM technology.
I've really no idea on this, Apple are the only ones I know using it so heavily. Other companies probably don't care about it except for having the extra DIMM count.
The Nehalem-EP dual processor architecture offers 3 DDR3 channels per processor with each channel supporting up to 3 physical DIMMS. That's a maximum of 9 DIMM slots per processor or 18 for a dual socket system. This is supported by Intel's roadmap slide and existing prototype boards that are floating around.
I am under the impression 18 DIMMs is intended for servers with 12 DIMMs being for Workstations due to the reduced physical space from extra PCI-E, PCI, PCI-X slots.
Another intersting fact is that early Nehalem-EP systems appear to support non-ECC memory which is good news for desktop users as it should improve performance (slightly) while reducing cost and increasing choice. (Link)
I think this may come down to how Apple (and maybe Intel) want to push things from a marketing perspective. I could see Apple not wanting people to figure out if they need need ECC or Registered memory and just forcing it, they will almost certainly be only offering one type on their store.
 
I am under the impression 18 DIMMs is intended for servers with 12 DIMMs being for Workstations due to the reduced physical space from extra PCI-E, PCI, PCI-X slots.

I think this may come down to how Apple (nad maybe Intel) want to push things from a marketing perspective. I could see Apple not wanting people to figure out if they need need ECC or Registered memory and just forcing it, they will almost certainly be only offering one type on their store.
In Apple's case, the current model uses an SSI CEB board (12" x 10.5"), so the risers were a slick way to offer that much memory with the space constrictions. Just not enough PCB real estate otherwise.

There's no way to be certain, but they would likely repeat it if they wish to use the same size board and similar sized case, if not just slightly modding the current one's internals. Placing the DIMM's directly on it, would require SSI EEB/E-ATX, and a different case to squeeze it in. (Unless they eliminate something like the PCIe slots). :eek: :p

Most WS boards aren't SSI EEB/E-ATX (12" x 13"), so 12 is the practical limit. Server boards are however, so 18 DIMM slots can actually fit. ;) (Then there's the fact that servers are expected to run massive amounts of RAM). :p

The Nehalem architecture will allow for 3 memory variations (unless intentionally disabled), but given the cost, Registered would be very unlikely. In keeping with the workstation/server classification, and Apple is truly serious about retaining this, Unbuffered is the most likely, given it's the least expensive form of ECC. Granted, DDR3 (non ECC) is possible, but I'm not sure if Apple would go that far. (At least I hope not). :p
 
I am under the impression 18 DIMMs is intended for servers with 12 DIMMs being for Workstations due to the reduced physical space from extra PCI-E, PCI, PCI-X slots.

I think this may come down to how Apple (and maybe Intel) want to push things from a marketing perspective. I could see Apple not wanting people to figure out if they need need ECC or Registered memory and just forcing it, they will almost certainly be only offering one type on their store.

I agree on the DIMM slot count and board real-estate relationship... 18 DIMM slots is the maximum... there could be anything from 6 to 18 with 12 being a good choice.

I also agree that Apple will likely promote only one kind of memory but regardless of what they advocate, it appears the IMC on the Nehalem die can support either giving owners the option of using whatever they choose unless there is some way for Apple to force this in firmware.
 
In Apple's case, the current model uses an SSI CEB board (12" x 10.5"), so the risers were a slick way to offer that much memory with the space constrictions. Just not enough PCB real estate otherwise.

There's no way to be certain, but they would likely repeat it if they wish to use the same size board and similar sized case, if not just slightly modding the current one's internals. Placing the DIMM's directly on it, would require SSI EEB/E-ATX, and a different case to squeeze it in. (Unless they eliminate something like the PCIe slots).

As I mentioned, I don't believe risers are possible with DDR3 memory. Risers were only possible on the current platform due to the unique architecture afforded by FBDIMM memory which Nehalem does not support (as of yet).
 
To start yet another thread within a thread... does anyone think it's possible Apple will offer a single socket Mac Pro? The reason I ask is that dual processor Nehalem's are simply way more computing power than most power users could every use. While they could duplicate what they are offering now with a dual socket board with only one processor for the low-end, would it make sense to further trim costs and offer a single-socket entry-level Mac Pro.
 
Two PCI-E 2.0 x16 and one PCI-E 2.0 x4 per IOH (actually 36 lanes however is needed). The chipset can support two IOHs. The southbridge supports six PCI-E 1.1 lanes that can be configured as one x4 and two x1.

The current Mac Pro has two PCI-E 2.0 x16 and two PCI-E 1.0 x4 slots.

When counting PCI-e lanes, I believe it's important to budget 4 of the 36 lanes on the Tylersburg for the interface with the ICH.
 
To start yet another thread within a thread... does anyone think it's possible Apple will offer a single socket Mac Pro? The reason I ask is that dual processor Nehalem's are simply way more computing power than most power users could every use. While they could duplicate what they are offering now with a dual socket board with only one processor for the low-end, would it make sense to further trim costs and offer a single-socket entry-level Mac Pro.

I think they will, not so much because we want one, but because there's a very good chance the dual socket machine is going to grow both in price and size to avoid losing features. Most Mac Pro users aren't going to be able to take another $2-300 price hike. The current case is too small for a dual Xeon 5500 machine with 12 DIMMs (can't be on risers like FB-DIMMs) and too big for an i7/Xeon 3500 machine with 6 DIMMs.


I prefer the term PowerMac replacement. The xMac idea was a high end professional workstation to match the xServe, which is what the MacPro basically is.
 
When counting PCI-e lanes, I believe it's important to budget 4 of the 36 lanes on the Tylersburg for the interface with the ICH.

This is what I originally thought the situation was, but was told otherwise by someone and I haven't discussed it with them since. That would mean the use of two IOHs to provide the same PCI-E functionality as the current Mac Pro has unless I am mistaken.
 
As I mentioned, I don't believe risers are possible with DDR3 memory. Risers were only possible on the current platform due to the unique architecture afforded by FBDIMM memory which Nehalem does not support (as of yet).
FB-DIMM definitely made it easier, but DDR3 on a riser is possible, requiring a custom socket.

It would likely even become necessary if they attempt to stay with SSI CEB boards. Otherwise, it will have to get bigger. No way around it, given the larger package (LGA1366) eating even more of the PCB.

----------

One a separate note, I finally located a listing for a DP board for the Xeon 5500/3500 series. (Going by the info posted).

SuperMicro X8ST3-F-O listed on provantage. It's not even up on SuperMicro's site ATM. Apparently SKU only, as it's listed as "Special Order". ;)

Kind of an odd-ball, as it has 16 DIMM slots, in an unusual configuration. 12+4. It also indicates it only supports Unbuffered and Non ECC DDR3.
 
This is what I originally thought the situation was, but was told otherwise by someone and I haven't discussed it with them since. That would mean the use of two IOHs to provide the same PCI-E functionality as the current Mac Pro has unless I am mistaken.

Intel PC motherboard manufacturers (ASUS, Gigabyte, MSI, etc.) often add additional PCI-e lanes using PCI-e switches which is just a small inexpensive chip. Apple could do the same to offer more expansion capability much cheaper than adding a second Tylersburg chip.
 
Intel PC motherboard manufacturers (ASUS, Gigabyte, MSI, etc.) often add additional PCI-e lanes using PCI-e switches which is just a small inexpensive chip. Apple could do the same to offer more expansion capability much cheaper than adding a second Tylersburg chip.
Though the 2nd Tylersburg would be nice, I wouldn't expect it.
 
Another intersting fact is that early Nehalem-EP systems appear to support non-ECC memory which is good news for desktop users as it should improve performance (slightly) while reducing cost and increasing choice.

I'm sorry but ECC memory is a good thing for us professionals, which the Mac Pro is aimed at. Not regular desktop users...

This isn't about a diminishing 3% performance increase but about stability. Can't have our calculations crash midway because of snotty memory.
 
Apple has single CPU Mac Pro's now. It leaves 1 socket open and it costs $500.00USD less than the "standard". So one question is: Will they continue with a 2 socket MP with one filled and one vacant? (Um, I think I read somewhere that's how they do it - correct me if I'm wrong - which I frequently am:p)http://store.apple.com/us_smb_78313/configure/MA970LL/A?mco=NzQ3Njkz I could get that, and be a 4 core whore for awhile - then toss in the other CPU when the price and my wallet converge.:D
 
I could get that, and be a 4 core whore for awhile - then toss in the other CPU when the price and my wallet converge.:D

You could but it may cost you more than what you saved to get the other CPU later - you should check before you make this decision.
 
http://www.theregister.co.uk/2009/02/10/intel_confirms_nehalem_imminent/

"We are currently in production, and we expect to have a system introduction later this quarter. So it is imminent." :)

And the reliability of the Register is... what?

Seriously, I don't know anything about these guys. What's that one site that seems to report on absolutely everything months in advance in great detail and then only gets it all half right?

Am I thinking of the Register?

Also, the first post has been updated with the Tylersburg DP info.
 
And the reliability of the Register is... what?QUOTE]

The Register are hardly a rumor site, they generally report on news and provide hardware reviews. A lot of people in the technology industry subscribe to their news letter as it reports on useful, genuine industry information.

This article begins with "During a briefing with reporters this morning in San Francisco" and then continues to quote what Stephen Smith, vice president and director of operations at Intel's digital enterprise group was report to have said.

Nothing we haven't heard third hand already really but this apparently comes from a reputiblesource - Intel.
 
And the reliability of the Register is... what?

Seriously, I don't know anything about these guys. What's that one site that seems to report on absolutely everything months in advance in great detail and then only gets it all half right?

Am I thinking of the Register?

Also, the first post has been updated with the Tylersburg DP info.

Yeah, the Register generally isn't the most reliable site in the IT/Computer world, that's for sure.
 
I read The Register daily and its basically a news site with one or two rumours thrown in as well as a lot of opinion pieces (Uncov's Ted Dziuba writes for them).

This article seems to directly quote Intel and is a news rather than rumour article.

Some of their Apple rumours do leave a lot to be desired: they usually quote something from this site ;)
 
To start yet another thread within a thread... does anyone think it's possible Apple will offer a single socket Mac Pro? The reason I ask is that dual processor Nehalem's are simply way more computing power than most power users could every use. While they could duplicate what they are offering now with a dual socket board with only one processor for the low-end, would it make sense to further trim costs and offer a single-socket entry-level Mac Pro.

Never say never my friend. When I got my Rev. A Dual 2.0 G5 it was more than I possibly could have needed. Now 5+ years later with the work that I do it is just starting to falter. So the need will eventually catch up to the users computer.

But that is the beauty of macs that the trolls never seem to understand. Yes I buy an overpowered machine at a premium but it lasts me far longer than most off the shelf PCs I could have bought 5 years ago. Macs amortize their costs quite nicely. The trick is to buy early and buy the best.
 
Apple has single CPU Mac Pro's now. It leaves 1 socket open and it costs $500.00USD less than the "standard". So one question is: Will they continue with a 2 socket MP with one filled and one vacant?

What they do depends on price and size. With the current Mac Pro, Apple didn't have a whole lot of options. The dual socket boards used 771 pins. The single socket boards used 775. You couldn't mix the Core 2/Xeon 3000-series and their 975x/x38/x48 motherboards with the 5000 series xeons and the 5000/5400 motherboard. This time around, the socket and the motherboard chipset is the same, like it was with the G5/U3 combination. When Apple does the driver work for the dual socket Mac Pro, they would have already done the work for the single socket machine.

Never say never my friend. When I got my Rev. A Dual 2.0 G5 it was more than I possibly could have needed. Now 5+ years later with the work that I do it is just starting to falter. So the need will eventually catch up to the users computer.

But that is the beauty of macs that the trolls never seem to understand. Yes I buy an overpowered machine at a premium but it lasts me far longer than most off the shelf PCs I could have bought 5 years ago. Macs amortize their costs quite nicely. The trick is to buy early and buy the best.

How many programs are designed to take advantage of 8 cores, even the professional ones? What people like you seem to forget is that there was a thriving lower range of PowerMac user for years and we got along just fine until Apple pulled out the rug. My B&W G3, which was also made by Apple, lasted 5+ years as well and without the hefty $2300 price tag Apple now commands.
 
Status
Not open for further replies.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.