YKhan said:
My feelings too, but I'm not sure a new core is all that important to
them. It seems all of their performance improvements were as a result
of stuff that hangs off of the core, but isn't really a part of the
core, like HTT and memory controller; making further improvements on
those seems to be the path of greater return.
The K8 had several improvements that were not related to their memory
controller or HT that were significant. Certainly, the most noticeable
changes were those two, but they also did some interesting stuff with
distributing and slightly enlarging instruction queues.
I think that depends if you are talking about MP performance or single
socket performance. I don't see a big advantage in the case of single
socket systems. In fact, I think AMD will be at a slight disadvantage
when they are still using shared interface systems and Intel is using
shared cache (again, for the 1 socket scenario, unsure about MP).
They can probably make
even lower power AMD64's by adding circuitry into the existing core
like Intel did with the Pentium 3 to come up with the Pentium M, which
would be the intelligent power management stuff.
I think you are trivializing what Intel did with the Pentium M. The
Pentium M is a radically different chip from the P6 or Pentium 3.
Sure, there was some different circuit design stuff done, but there was
also quite a bit more. Intel has been steadily improving their
decoders, so that some operations which used to be complex are now
simple, etc. etc.
If you want to get a grasp on the differences between the P!!! and the
PM, check out:
ftp://download.intel.com/technology/itj/2003/volume07issue02/art03_pentiumm/vol7iss2_art03.pdf
I'm assuming you are familiar enough with the P!!! uarch to make the
implicit comparisons...
David