Advanced Micro Devices Inc (AMD)

[Petz]

A critical look at some other aspects of the xbitlabs article at http://www.xbitlabs.com/news/cpu/display/20040302091145.html

Here's a summary of what it says:
25-May: unveiling (introduction) of socket 939

Q2: 50K socket 939 CPUs (Athlon 64 and Athlon 64 FX)+700K S754 Athlon 64. He states that FX chips will remain 1M L2 even on S939, therefore this implies that the 50K chips are split between the two types.

Q3: 260K S939 + 970K S754. PROBLEM: Even if you add in 200K Opterons, the total 1.4M is way less than half a fab should be able to produce. At the very least, over half of AMD's production capacity is supposed to be K8s by the end of Q1.

But the main problem I have with these projections is that it would make absolutely no sense for AMD to not switch over to the smaller die S939 CPUs rapidly. S939 CPUs will give better performance using less silicon.

Q4: 540K S939 + 900K S754. These numbers directly contradict what Rivet said on March 1:
When crossover to K8? A: Sept/Oct/Nov
See http://www.siliconinvestor.com/stocktalk/msg.gsp?msgid=19864175

If AMD truly changed their guidance since March 1, it would be a violation of the quiet period to reveal that to anyone.

Q1'05: 1.2M S939 + 250K S754. Again, no matter how many Opterons you add, this would imply that Rivet's "K8 crossover statement" was in error by at least 6 months.

The comments on 90 nm are flat out incorrect in reporting what Rivet actually said.

xbitlabs: Advanced Micro Devices said its first commercial 90nm microprocessors will show up in "Septermber-October-November” timeframe.

No, this was the date range given for K8 crossover. It was combjelly who, on SI, said that early April wafer starts meant September/October system availability, and he later moderated that statement:
http://www.siliconinvestor.com/stocktalk/msg.gsp?msgid=19864134
http://www.siliconinvestor.com/stocktalk/msg.gsp?msgid=19864273

So, w.r.t. 90 nm rollout, Anton is remembering what a poster on SI said, rather than what Rivet said.

More importantly, w.r.t. the projected numbers, they are inconsistent with what Rivet said 3 weeks earlier. Has anything changed negatively since then?

Petz