a reader (a@b.c) on 3/19/10 wrote:
---------------------------
>David Kanter (dkanter@realworldtech.com) on 3/19/10 wrote:
>---------------------------
>>a reader (a@b.c) on 3/18/10 wrote:
>>---------------------------
>>>>Magny-Cours is 692mm2 of silicon. Westmere is 248mm2 of silicon.
>>>>
>>>
>>>that settles it. all hail intel.
>>
>>heh, i dunno know about that, and I hope you were being a little facetious : )
>>
>actually am a little worried about of lack of effective
>competition. otoh, intel earned it fair and square.
>good for them.

Yeah, Intel has been on a roll lately, they've managed to hit *most* important schedules.

Nehalem-EX is a little late, but that's not b/c of Intel.

>>It's not clear at all how AMD will price Istanbul and Magny Cours. It seems like
>>AMD ought to prefer pushing MC into the 4S server market, where the margins are
>>sufficient and the competition less intense. Not to knock Nehalem-EX, but it's still a 45nm product, not a 32nm one.
>>
>
>well, nehalem ex + ex5 can steamroll pretty much everything
>in its path. both intel and ibm got a winner here.

That's what I expect, although I don't know exactly what they are doing in EX5. For some reason, the folks in IBM's PR group ignore me...

My guess is that IBM will include a pretty hefty snoop filter, remote directory and cache for remote data (just like X3)...likely in eDRAM.

I'll be curious to see:
1. How big the local cells are (4 or 8 sockets)
2. How much I/O per local node
3. Whether the remote data cache is still in memory
4. Will it still be limited to 4 cells
5. Bandwidth between EX5 chips
6. Itanium support (just kidding here)

It'll be interesting to see if any of the IPF vendors decide they want to reuse their chipsets for Nehalem-EX though. I would think most of them could come up with a credible Nehalem-EX system...

David