"The idea is that hyperthreading will mask latency problems."
It will help, just like pre-fetching did. And larger caches, especially a larger trace cache in combination with hyperthreading will help a lot.
But, DDRII, assuming it lives up to it's billing which is always a tricky thing, should combine both lower latency and near equivalent bandwidth. Even without DDRII, plain old 126 bit wide DDR will do that. And that doesn't make the motherboards all that expensive, look at nVidia.
While there may be some 'Droids, or even many, who expect Intel to be standing still, there are a large percentage of 'Bees that expect the same of AMD. Neither company is standing still, but both are hampered by the long design cycle of processors. When you throw in the added factor that a Timna-like processor at the right time will be a goldmine, but the wrong time is a disaster, it makes it all the more interesting...
Note: I have been predicting that the industry will wind up with fully integrated processors for more than 15 years. And at some point, they will be all the rage. Fortunately, I am not charged with strategic direction with any of the companies... |
