what the hell are people on, for one thing the k8l chips are coming out in 6 months on am2, am3 is pin compatible with am2(i think the idea is am3 chips will work with ddr2 in am2 boards, or am3 boards with ddr3). k8l will also be what quad cores launch in and judging by very early numbers on the kentsfield(intel quad core) even with something that "should" fit in cache it takes a 25% hit in speed per core when all 4 cores run due to a very inefficient connection to the memory. a k8l quad core is faster clock for clock than what a current k8 quad core would work like, added to its already far more efficent connection to the memory and quad core k8l's start looking VERY good in theory.
k8l's will undoubtedly be better, by a decent amount, than the k8.
as for the fx62, its end of life, theres only going to be a few left after the fx64 launches its dropping price slightly to sell the last few thats it. the fx64 won't be faster than a conroe, but in the same sense a 6700 is only marginally slower than a x6800, yet the price difference is vastly more. fx64's will still sell, they sell a tiny amount and are there largely to encourage buyers that there is something really great, and getting a chip largely similar for half the price is a good deal. 99% of people won't know the x6800 is faster than the fx64, we do, most of the world doesn't.
jadeskye, before a new architechture is launched there will already be teams working on the one two in the future. k8l has been a long time in the works, as has conroe, as has kentsfield, and amd quad cores, and the mobile specific amd chips due out end of next year. this is investment companies have to spend to not die off. conroe launching now won't make amd suddenly think they need a new design and spend loads making one, they've been spending for years and will continue to do so forever.
i can't remember all the stuff k8l is bringing, a larger and much better branch predicter, improved mem controllers, larger sse2/3 units i believe and just a bunch of stuff really all which will improve clock for clock performance.