Michael S (already5chosen@yahoo.com) on 3/30/09 wrote:
>
>So, starting from today, quad-socket x86 servers are
>officially dead. Possibly, we will see a resurrection of
>this segment when (early next year?) Beckton comes
>to market. Or, may be, it would be too little to late.
>Anyway as it stays now, anybody who buys 4S x86 just
>proves that he has no clue.

Well, I think that's mainly because Nehalem SMT works so
well on OLTP workloads like TPC-E, and because nobody is
doing 4S Nehalem boxes (and it does look like Intel won't
support it at all until the Nehalem-EX octocore monster).

So I don't think it's "4S x86 is dead" as much as it's just
a matter of Nehalem being pretty good in general, and then
especially good for server loads that take advantage of
SMT.

And yes, obviously right now it's limited to 2S, and then
other non-Nehalem 4S x86 boxes then end up looking a bit
weak in comparison.

But other workloads won't get as much SMT improvement, and
will prefer 4S for extra memory bandwidth and full cores.

I do believe that Nehalem-EX will revitalize the 4S x86
market. And when that happens, the thing to compare against
will be 8S POWER, I suspect. 4S/32C/64T will do some real
damage.

Linus