Advanced Micro Devices Inc (AMD)

[Elmer Phud]

Paul -

In any case, I intended my main point to be that it was Tom talking, so I don't lend it a lot of credence.

I don't give it a lot of credence either. I was under the clear impression that September meant retail. Others here disagree and they may be right.

What are the implications of no September retail rollout? Someone here suggested that AMD didn't want to hurt Barton availability by using fab space for Athlon64 but if AMD has a production stepping available today, and used only 1% of their capacity to produce Athlon64, they would be producing about 55 wafers per week. At even a poor yield of 175GDPW** that's ~10K Athlon64s per week. With 6 months until rollout they could easily produce 250K Athlon64s with only 1% of their fab capacity. While some would still be in the pipeline, some would be coming out now as well, so it is easy to show that AMD could deliver 150K Athlon64s to customers for retail distribution in September. More than enough for retail availability. That's with less than 1% of their fab capacity going to Athlon64 if they had a working version today in production.

So it appears to me that, because 1% of their fab capacity is negligible, if AMD cannot deliver enough Athlon64s to assure retail rollout in September then we are left with several possibilities:

#1 They don't have a working, production version today.

#2 Yields are so terrible that the existing device would require a much larger percentage of capacity to produce. (I already assumed poor yields in my above estimates.)

#3 The die size estimates are no longer valid. This could be the case due to poor performance which requires a much larger cache to compete with Intel's offerings in Q3. The 1Meg Athlon64 has an estimated die size of 180mm2*** which would give a yield of about 80 GDPW with the same poor defect density I assumed above. This would mean AMD would need about 2% fab capacity to meet the same delivery numbers already discussed.

I don't believe AMD is doing this badly and I still expect to see Athlon64s on store shelves in Q3.


* die size estimate of 104mm2 for 256K L2
http://www.geek.com/procspec/amd/k8.htm

** defect density estimate of .35/cm2 (world class = <0.25/cm2)
http://www.icknowledge.com/misc_technology/die_calculator.xls

*** http://endian.net/details.asp?ItemNo=2490