InvestorsHub Logo
icon url

wbmw

11/27/02 11:37 AM

#2792 RE: fowler #2790

John, Re: The 1MB version of Hammer will likely sport a surface area of 180 square millimeters, said Kevin Krewell, managing editor of industry newsletter Microprocessor Report. The smaller version for desktops will take up around 105 square millimeters.

Intel's current Pentium 4 and Xeon chip for workstations take up 146 square millimeters of area, according to Intel. These chips contain 512KB of cache. A Xeon with 2MB of cache takes up 268 square millimeters.


Most of the article is a boring rehash of old info. Amazing that Michael Kanellos gets away with it. The above quote, however, confirms a few things for me, one that I remember hearing from Microprocessor Forum. That is that Opteron will be 180mm^2. Compared to Xeon, it is smaller, but I think Kanellos is using an obsolete figure, since Intel has shrunk their .13u chips by about 10% (making the 2MB Xeon about 240mm^2).

On the desktop side, a 105mm^2 Athlon 64 with 256KB L2 cache would compare with a Pentium 4 Northwood at 131mm^2 (again with the 10% shrink). Of course, with SOI yields, AMD may not be able to get as many good die as Intel can, even with the die size advantage. Of course, we've known about that, too, for quite some time.

wbmw
icon url

subzero

11/27/02 1:26 PM

#2797 RE: fowler #2790

The chip will debut at around 2GHz and come out with a performance rating number in the mid-3,000s, added Dirk Meyer,

Is AMD completely confused with their Opteron strategy?
I ask this because AMD keeps clinging to their Quantihurtz model numbering system - which compares an AMD CPU to a Pentium 4 running retail/consumer software applications.

Out of one side of AMD's mouth, they claim that Opteron will compete with Intel's Xeon.
Well, Intel's Xeon is not generally used to run office applications or games - it is used to run server-specific applications - or workstation applications.
More often than not, Xeons are used in 2-way or 4-way configurations, sometime 8-way or more.
Hence, why would any customer care about an artificial model number for an Opteron CPU based on desktop software in a single cpu mode, when their targeted application is 2-way (or higher) running server and/or workstation applications?

AMD keeps clinging to their retail white box fan base but this will be a pitifully small market for their Opterons.

Now, will AMD re-characterize their model hurtz rating using TCP- type tests for server-specific applications? After all, that, they claim, is their "market" goal.
Or will they just be laughed out of corporate accounts with their quake 3 model rating roots and numbers?

-SZ