I'm pretty close to Althornin, too. The P4 prices in the low-mid range have a serious bottom under them (namely Celeron...hmm...perhaps I should have said "a serious ass under them"?), while the AXP prices just go into freefall. From a price/performance standpoint this is really screwed up, because the lower clocked P4s are all 400 MHz FSB, while down in the high teens the Quantispeed ratings really meant something. (After all: 1400+ was extremely conservatively rated, and the daft 66 MHz = 100+ formula distorts more the higher the rating goes.)
The P4 starts becoming a good value when the 533 MHz FSB kicks in, for one thing because as per Intel policy, the FSB increase doesn't change the list price at all (i.e. 2.4 GHz P4 costs the same as 2.4B; street price there's like a $5 difference). Meanwhile, the performance value of a Quantispeed point continues to plummet, just as the prices get closer to parity. As I noted before, street prices reach parity at 2600+ (interpolated between 2.53 and 2.66 GHz P4s) and stay there all the way on up. (Including the new 3000+ Barton, which is actually showing on pricewatch! Reading Anand's review now, maybe if I feel masochistic I'll post on it...tomorrow.)
Meanwhile, performance "crossover"--i.e. when 1 AXP point actually equals 1 P4 MHz--probably occurs around 2400 when the P4 is on 400 MHz FSB, and would occur around 2000 for 533 MHz FSB (of course the directly comparable chips don't exist, unless you can get an unlocked (engineering sample) P4).
To sum up: the AXP would appear to have a very large price/performance lead up through 2200 (+/MHz), and the P4 has a large price/performance lead from around 2600 (+/MHz) onwards. The AXP price/performance sweetspot is 1800+ to 2100+, while the P4 price/performance sweetspot is 2.4 GHz (533 FSB) to 2.66 GHz.
Given how different the two IHV's price/performance curves are, it seems safe to say that those price/performance sweetspots very closely mirror speed bin sweetspots at the fabs.
The P4 starts becoming a good value when the 533 MHz FSB kicks in, for one thing because as per Intel policy, the FSB increase doesn't change the list price at all (i.e. 2.4 GHz P4 costs the same as 2.4B; street price there's like a $5 difference). Meanwhile, the performance value of a Quantispeed point continues to plummet, just as the prices get closer to parity. As I noted before, street prices reach parity at 2600+ (interpolated between 2.53 and 2.66 GHz P4s) and stay there all the way on up. (Including the new 3000+ Barton, which is actually showing on pricewatch! Reading Anand's review now, maybe if I feel masochistic I'll post on it...tomorrow.)
Meanwhile, performance "crossover"--i.e. when 1 AXP point actually equals 1 P4 MHz--probably occurs around 2400 when the P4 is on 400 MHz FSB, and would occur around 2000 for 533 MHz FSB (of course the directly comparable chips don't exist, unless you can get an unlocked (engineering sample) P4).
To sum up: the AXP would appear to have a very large price/performance lead up through 2200 (+/MHz), and the P4 has a large price/performance lead from around 2600 (+/MHz) onwards. The AXP price/performance sweetspot is 1800+ to 2100+, while the P4 price/performance sweetspot is 2.4 GHz (533 FSB) to 2.66 GHz.
Given how different the two IHV's price/performance curves are, it seems safe to say that those price/performance sweetspots very closely mirror speed bin sweetspots at the fabs.