There will also be a quad-CPU enabled Xeon E5 Socket 2011 version, a quarter later from now. How much memory can it take? How about a terabyte?
Two QPI links on Socket 2011 CPUs do enable connecting each processor to TWO other ones, if you accept lesser inter-processor bandwidth compared to just dual-link pairing of two CPUs alone. There will be a separate Xeon E5-4600 series covering that particular market segment, which sits between the high end dual-socket E5-2600 and the enterprise RAS-enhanced even higher end E7 series (those fancy 10-core 30 MB L3 cache Westmere-EX chips and their followons).
These are good for uber-rendering workstations with many threads, as well as SMP server jobs where high per-core performance is more important than RAS capability or huge interprocessor bandwidth – E7 series have, after all, four QPI links on each CPU for that purpose.
Now, since the maximum memory load on each Socket 2011 processor is three DIMMs per channel, or 12 in total, it would allow for 48 DIMMs altogether on a quad processor board, as you can see here on the 'Lizard Head Pass' S4600LH board expected from Intel. A truly lovely platform which, if good liquid cooling system was there, could actually fit into 1U platform! Else, it's 2U, still not bad at all.
Now, remember those fancy Samsung – and other vendors' – 32 GB LR-DIMMs at DDR3-1333 speeds and taking just like 5 watts? Well, put 48 of them into this 'little' mobo, and you got… umm… 1.5 TB RAM, yes. Now, inexpensive large memory data mining and analysis ops, not to mention simulations like, for instance, impact of MRT line failure on all the car and bus traffic in Singapore down to every person, vehicle and path – an increasingly more often required app here, it seems – become very real, especially when coupled with the current low memory pricing.
And, oh yes, there are 160 PCIe v3 lanes in total on such platform. Now, how about an extender box for an 8 x dual-GPU setup as an accelerator? A eight-set of liquid-cooled OC AMD HD7990 dual GPU cards could give you up to 15 TFLOPs of double precision power to assist those Xeons on board…