Dale, Re: "My SWAG is that they will achieve at least a 20% increase in peak clock rate, slightly higher average IPC for single tasking, and significantly higher IPC for multitasking and multiprocessing."
Regarding 20% higher clock rate, I doubt it, simply because there is not a linear relationship between the number of pipeline stages and the extra clock frequency that can be obtained. AMD may have extended the pipeline by 20% (12 stages vs 10), but that doesn't mean that frequency will improve that much. BTW, you are speaking with respect to Barton, right?
Regarding IPC in single threaded apps, I agree, since an improved branch prediction rate, and an improved TLB system will no doubt give Hammer incremental increases in performance, overall.
Regarding IPC of mutlthreaded apps in multiprocessor mode, I think that depends on your definition of "significant". I think that, like the 760-MP, AMD will have a more linear increase in performance by adding more processors than Intel can with a shared bus. However, how much more is questionable at this point.
wanna_bmw |
