Replies to post #28250 on Advanced Micro Devices Inc (AMD)

[jhalada]

Does anyone have any idea's about the likely speed grades bump that K8 will get from their transition to 90nm ?

30% to 35% improvement is a good ballpark.

Joe

[sgolds]

out to make money, welcome aboard! As to speed grades with 90nm, that is a big question mark. AMD has not disclosed expected speed grades so we have to fall back on the two examples already in production.

Prescott has its famous power and heat issues, so this 90nm product would make it seem that 90nm is very difficult. However, that has to be tempered against the fact that Intel and AMD's processes have diverged so far. What lessons can one take from bulk 90nm to SOI? It is not very clear.

The other example is a lot closer - IBM's Power microprocessors. IBM and AMD have very similar technology at 90nm, they collaborate. (65nm and 45nm are being codeveloped by the two companies so that process will be virtually identical for both.) IBM's product does appear to be much lower power - I recall reading that a 90nm PowerPC chip, 65mm2, uses 39W power maximum at 1.5GHz. The power is low, but IBM has not shown its hand yet in scaling it up. Still, it is a hopeful sign because low power is one of the requirements for good scaling, and this is where Prescott fell flat on its proverbial face.

Thus there is good reason to expect that AMD's experience will be better than Intel. However, we will not really know until some samples are in the hands of overclockers (and only the FX supports overclocking among the K8 chips).

[CombJelly]

"Does anyone have any idea's about the likely speed grades bump that K8 will get from their transition to 90nm ?"

Traditionally? At introduction there often isn't any initial speed bump, but with in a month or two, speed starts to go up. Over the life of the process, it is generally at least 50% and often gets to 100%.

AMD might not do that at 90nm. For one, there might be some severe problems with static current leakage, Intel seems to be having a problem with this. But if it doesn't, there might be an opportunity for AMD. They might want to emphasis low power over high clock rate, at least for the first generation of the processors on 90nm. Given that Prescott doesn't seem like it will be a barn burner in anything but the literal sense, AMD could be following a low power path for this year, moving to a higher power consumption and/or dual core next year before they start the transition to 65nm...