Re: AMD's Problem
AMD has many challenges but the days of substantial gains from a reduced node size are long gone for everyone. A node shrink can lower power consumption a bit and increase density, but the increased density makes cooling more difficult.
AMD's latest Carrizo chips are on 28Nm even though they were planned to be on 20Nm but none of the Fabs were able to get decent 20Nm yields. Despite staying on 28Nm AMD was able to make substantial improvements in power reduction and IPC, quite comparable to Intel dropping a node size. Any process below 32Nm yields very small performance gains. Zen will be on 16Nm FinFET which is a good step forward on many fronts.
As far as Jim Keller leaving, he did this before the K8 was released also. He's an interesting guy who gets bored easily unless he has a new challenge. After Zen comes Zen plus an evolution of Zen that is already in full development. Thus AMD is set for several years so it's no surprise that Jim has left for another challenge. He may return to AMD some day in the future. One never knows for sure. Either way Zen is a great new architecture core and it will deliver excellent performance.