Intel did the same thing with their first Core 2 Quad processors. The Kentsfield generation may have had more limitations to it's multithreaded performance than the single-die Yorkfields but the performance penalty was so negligible that dual-die chips like the Q6600 stayed relevant as long as the next generation of single-die successors.
I forgot the Q6600 was dual-die, probably the best bang for buck CPU I've ever purchased (first popular quad-core & huge overclocking potential).
I think AMD reusing the same dies is quite elegant (everything* scales as you add more cores). The inter-die bandwidth/ latency shouldn't really be a problem, because if more than 8 cores are dependant on the same data - locking issues would make any additional cores useless anyway. If they can get NUMA tuned correctly things should work nicely.
*it's strange that they all have the same amount of L3 cache.
Does anybody have contacts with AMD's marketing/ engineering department who can provide test units?
Maybe AMD pulls something akin to the "Iris chip" from Intel and makes L4 standard in a few years. Speaking of which, never understood why Intel wouldn't push a bit more for Iris, when used as L4 it gave close to 20% increase in performance (due to lesser cache mises I believe).
When Intel did the dual-die thing with Kentsfield, AMD mocked them for not being "true quad-core" and refused to follow suit. They had quad-core dies back then, so they could have easily hit back with an 8-core part if they swallowed their pride and put more than one die on a chip! I think it was a terrible move, just when AMD was starting to become irrelevant again.
10 years later, Intel sticks to single-die chips, and AMD has no qualms stuffing a gazillion dies on a chip. Looks like they've learned a lesson or two :)
