Replies to post #29161 on Advanced Micro Devices Inc (AMD)

[smooth2o]

wmbz: What if by the time Intel adds all that stuff, AMD has dual core A64s? Actually that brings up a good question. Will Dothan and its successor support Hyperthreading or is that going to die along with Netburst?

What if? Intel can do dual cores also. If all mentioned here comes true, I suspect Intel will be further ahead of AMD in profits, marketing and features. Obviously, these architectures are pushing toward limits (that would include AMD, BTW). The only way to get far more performance is to add dual cores. I suspect that in Intel's architectures, HT will continue on each core.

Things to note are that dual cores take more silicon. Power is a problem so one would think that dual cores will eventually fall the way of Dothan. Dual uP have always enjoyed higher prices (unlike going from 32 to 64 bits, eg) so one would think that the dual core products would be more expensive and ASPs would be higher, much like the top frequencies enjoy very high price premiums today. Marketing will be very key here, there are lots of ways to further differentiate. Dual cores could bring more silicon into play adding functionality to the CPU something Intel has always done. Profits and market share go to those who have the added capacity. Added capacity comes from having added profits.

Smooth

[Petz]

Interesting question, "Will Dothan & its successors support Hyperthreading?"

Dothan -- very, very doubtful

Successors -- Not sure it makes sense on a shorter pipeline architecture like Banias/Dothan. Basically, HT is a way of ameliorating the penalty of a lot of pipeline stalls. If pipeline stalls are not occurring, there is no advantage to HT.

A shorter pipeline architecture has less stalls because
1. There are fewer instructions in flight, capable of being stalled.
2. Fewer instructions in flight are competing for the same resources

Also, the stalls are not as catastrophic when they DO occur. Finally, Dothan has that HUGE 2M cache, which should nearly eliminate one of the prime reasons for stalls -- waiting for data from memory.

BTW, almost all these arguments apply to Athlon and Athlon 64, except cache. But memory stalls on Athlon 64 are much shorter than either Dothan or Prescott, because of the lower memory latency from integrated controller. So memory stalls don't waste as much time, reducing the potential benefit from switching to another thread.

So I think it is more likely that Intel will rename some other microarchitectural feature and call it hyperthreading, even if it has absolutely nothing to do with P4 HT. This will keep customers happy, thinking they haven't "lost" a feature.

Petz