Replies to post #81991 on Advanced Micro Devices Inc (AMD)

[Elmer Phud]

Wouter

unless they only want to focus on HPC

With the yields implied by their published defect density number and the huge die size, they would do well to limit the supply, but they're not going to recoup their investment by only servicing the HPC market. Add in Phenom and at 283mm2, poor yields and performance problems I'm still wondering what it is that's going to get them back to profitability? Dual core with be just as much of a challenge against Penryn and they'll have to drop prices again just to unload those DCBBs.

[chipguy]

IMO AMD will definitely need to lower their prices while Penryn ramps, unless they only want to focus on HPC (which might not even be a bad idea).

I wonder how that will work out in practice. Barcelona has
twice as many cores and twice the peak flop/cycle per core
as K8. OTOH it initially shares the same socket so has the
same bandwidth limitations. Add in the fact that it clocks
at only 2/3 as fast as high end K8s and it isn't clear to me
how attractive it will be for HPC vs cheap K8s for many
classes of HPC applications.

[Hans de Vries]

The Spec2006Int_Rate score for a two socket Opteron 2358 2.4 GHz
system will be around 102 with the newer compilers compared
to 106 for a two socket 3.0 GHz Clovertown Xeon system. So the
initial pricing seems to be justified, giving the large advantage in HPC
applications away for free.

What the increase from 3.0 to 3.16 GHz Penryn based systems will
do for Server (rate) applications isn't clear yet, 10-13%?
AMD indeed might have to adjust prices when these processors
hit the market.


Regards, Hans

[tecate]

Which would be what I believe they should have done from the beginning.

[Hans de Vries]

Your Harpertown pricing seems to be lower than that of Intel
published numbers them self however.

http://www.dailytech.com/Intel+Sets+Penryn+Launch+Date/article8451.htm


Regards, Hans