Linus Torvalds (torvalds@linux-foundation.org) on 2/6/09 wrote:
---------------------------

>I think it does.
>
>In fact, it's even easily explained: one counts processors
>as "cores", and another counts processors as "sockets".
>
>I would expect that most of the Itaniums sold would be
>dual-core. And suddenly the numbers don't look all that
>different any more. The 150k "processors" in 2007 were
>just 70k sockets, and if that's a third of the whole, then
>you had 200k sockets for the whole year.
>
>Then, the 2008 numbers say 160k sockets for the first
>three quarters. Looks like the same number, doesn't it?
>Certainly gives room for a slight single-digit increase
>too.
>
>And the 7-vs-4.1 is the same thing. An average of seven
>cores doesn't sound totally unlikely. In contrast,
>averaging 7 sockets sounds ludicrous. Sure, there are big
>machines, but the bulk of Itanium must still be quad socket,
>with probably a healthy sprinkling of duals.
>
>As to the $9k/core price, that just means that the hardware
>would be in the $70k range for a 4-socket IPF machine. I
>can't make hp.com give me a quote, but that doesn't sound
>too odd. But Intel won't be getting anything like that.
>They'll get $1.5k per socket or whatever.
>
>So I think your numbers line up fine, without looking any
>deeper at them.
>
>Linus

About the socket vs. cores ; In HPC I'd expect Itanium to target large single image shared memory systems with 8-16 and more sockets.Buying clusters of 2s/4s IPF servers for HPC seems doubtful considering its performance vs. x86.