Okay, I'm getting lost in product line names vs. processor code names...
It is very confusing, I admit.
So Beckton will be the Nehalem Xeon, and Gainstown will be... the highest-end version of the main processor line?
Beckton is a server processor, also known as "Nehalem-EX". It will be used in groups of four - so four physical CPUs. Each Beckton CPU can have eight CPU cores so you could have a total of 32 cores (and 64 threads). It will have 4 QPI Links - one for each CPU. It will also use FB-DIMM2 memory with four channels. It uses the LGA 1557 socket.
Gainestown is workstation and server processor, also known as "Nehalem-EP". It will be used in groups of two - so two physical CPUs. Each Gainestown can have four CPU cores so you could have a total of 8 cores (and 16 threads). It will have 2 QPI Links to link the two CPUs. It will use DDR3 memory with three channels. It uses the LGA 1366 socket.
Bloomfield is a high-end desktop processor. It operates as a single CPU. Each Bloomfield can have four CPU cores executing eight threads. It will have 1 QPI Link and will use DDR3 memory with three channels. It uses the LGA 1366 socket.
Lynnfield/Clarksfield is the "mainstream" desktop processor. It's effectively a Bloomfield except it only uses two DDR3 channels.
I don't get why Apple would drop the server chips that they've been using, though.
It is not a case of Apple dropping a server chip, but Intel changing the specification of that server chip. For dual-CPU workstations and servers, Intel uses three channels of DDR3. For four-CPU servers, Intel uses four channels of FB-DIMM2. I do not expect the Mac Pro or the X-Serve to move to four CPUs - at least directly. As such, they would be using the two-CPU chipsets which call for DDR3.
I want you to be right more than anything, but I just. don't. see them dropping FB-DIMMs and Xeons for Nehalem unless they plan on marketing the Mac Pro differently. You mentioned 16 to 32 cores. Apple cares about cores, as evidenced by Grand Central. With the Mac Mini stock left to run out and the Mac Pro being a niche, I can't see Apple caring enough to change their strategy to conform with what we want.
Regardless of what we might want, Apple could find themselves at a serious disadvantage in the high-end computing workspace if they don't move to Nehalem with expediency. If they're still running dual Xeon 5400s and the Windows world is offering dual Nehalems at 50% or better performance, major customers in Apple's core competency industries could move to Windows because time is money to these people.
Apple could very well move into the four-CPU space down the road. A 16-core/32-thread Mac Pro would indeed be a monster with Grand Central-optimized applications and system software. But I hope Apple doesn't wait for that before they make the move to improve the Mac Pro.
Again, I hope that you're right. I just need some concrete evidence to believe that Apple has an incentive for dropping FB-DIMMs other than the latency that could have been solved by not using Xeons in the first place.
There was no real incentive for Apple to use FB-DIMM - it was what they had to use to offer a dual-CPU platform on the Intel 32-bit architecture. The only incentive they have for not dropping it is because they want to only offer single-CPU and four-CPU options. And considering how folks carp about what a Mac Pro costs over an iMac, they're going to have a heart-attack at what a four quad-core Mac Pro is going to cost compared to a single quad-core CPU iMac.
Offering quad-core processors in single, dual and quad CPU configurations would allow Apple to maximize the benefits of Grand Central and tailor their models to more markets.