Well, I see that some benches have been leaked. Granted, most everything is in Chinese, so one has to go by the graphics. Now, not to rain on the parade, but while it's great to see that the Northwood can run at 2.6GHz (the cooling needed is unknown, though), since the part was speced for 2.2GHz and that a 2.6GHz part probably isn't due out for 4-6 months, it's not a good comparison to current processors. Also, since Sandra either doesn't have, or the "reviewer" chose not too, XP was not there for comparison. But, compared to the "old" P4, there isn't much of an increase. According to my calcs (which might not be that good, I admit) the 2.6GHz Northwood is only about 1% faster in MIPS. Also, basing even more off extrapolated numbers (this assumes a 20% increase for ever 400MHz increase, which is the difference in all three Sandra benches I use, based off the Sandara numbers), the integer iSSE2 numbers are less then 1% higher than the theoretically clocked 2.6GHz "old" P4, and the floating point iSSE2 is 3% higher.
Now, I admit, these numbers might be horribly off, for many reasons. But, if these numbers are close to the actual results of the Northwood, I'm very disapointed in the way it performs. I would expect the increase in L2 cache and FSB speed to yield an increase at least on the order of 6-10%. Hopefully for Intel, this sample was older, and much has been revised since this (The WCPUID date is 11/17/01). Otherwise, the die shrink accomplished only a wattage output decrease (and subsequently, the ability to scale further without needing exotic cooling). That's one point it could have going for it, as we all know that current processor operate at a wattage output that suggests there is much wasted energy. At least they make good space heaters.
-SammyBoy