Replies to post #50207 on Advanced Micro Devices Inc (AMD)

[chipguy]

I look at the common platform as part of an exit strategy of IPF: It's the golden hardware brigde to migrate to X86.

I think you are right. But *only* if IPF MPU designers
cannot distinguish their products from Intel x86 products
in performance and saleable features for a commercially
significant cross section of the server market. I think of it
as Intel taking the training wheels off the IPF gang. They
have until the end of the decade or so to demonstrate why
IPF should continue to exist alongside EM64T. But IMO
Montecito will show that quite readily before the end of
*this* year. ;-)

[jhalada]

Klaus,

I look at the common platform as part of an exit strategy of IPF

I was thinking along the same lines. If Intel creates a scalable interface (ditching shared FSB) for processors, it will open the doors to Xeon to higher end (which now is problematic with shared FSB and awkward connections between groups of 4 Xeon). Once Xeon can scale to higher performance levels, the need for Itanium end up lessened.

As far as emulation of IA64 on iAMD64 Xeons, it could be a face saving measure and some defense against litigation, it would be just a stop-gap measure for Intel to put the Itanium chapter behind.

The bottlenecks for Xeon were lack of 64 bit linear addressing, shared FSB and difficult (expensive) way to build systems with massive amount of memory.

iAMD64 solves first problem, the rumored "links" architecture will solve the second and FBDIMM will solve the third.

There will be a 4th benefit - increased parallelism through multicore chips. So the x86 server performance is about to make a leap, extending commodity pricing regime to much higher end systems.

The result is no need for Itanium, and a carnage for high-end niche server vendors.

The customers who are willing to spend big bucks are commercial customers desiring fast database performance. x86 vendors will fully satisfy this need with upcoming leaps in performance.

What's left is the HPC (scientific / technical computing, server rendering farm etc). These customers have no money to waste. They are looking for price performance, which again leads to the king of price performance - x86.

Joe