The "always" accurate folks at the Inquirer say Penryn will have hyperthreading in addition to 6 megs of cache, SSE4, Low/Hi K dialetrics/gates, and a 45 nano process.
There has always been "rumours" that Core2duo actually had HT circuitry (much like Willamette and early Northwoods did) but that Intel never turn it on.
HT was never worth a whole lot in the desktop universe, a couple % at best, it did however make 25% + differences in selected server type applications.
The big problem was that Netburst could only RETIRE two micro-ops per clock, so regardles of how full the pipeline got stuffed, what came out the other end was always limited.
With Core doing 4 (or kinda 5, depending how you want to argue and/or count things) instructions per clock, hyperthreading may actually work this time.
A more interesting possibility is that Intel may actually have Penryn use hyperthreading to (under some conditions) exectute BOTH haves of a conditional in the code, hyperthread both branches, and then actually use the one that turns out to be correct - greatly reducing the penalty of having to flush the pipeline in the event of a branch prediction error.
There has always been "rumours" that Core2duo actually had HT circuitry (much like Willamette and early Northwoods did) but that Intel never turn it on.
HT was never worth a whole lot in the desktop universe, a couple % at best, it did however make 25% + differences in selected server type applications.
The big problem was that Netburst could only RETIRE two micro-ops per clock, so regardles of how full the pipeline got stuffed, what came out the other end was always limited.
With Core doing 4 (or kinda 5, depending how you want to argue and/or count things) instructions per clock, hyperthreading may actually work this time.
A more interesting possibility is that Intel may actually have Penryn use hyperthreading to (under some conditions) exectute BOTH haves of a conditional in the code, hyperthread both branches, and then actually use the one that turns out to be correct - greatly reducing the penalty of having to flush the pipeline in the event of a branch prediction error.