With such a lineup they would have to compete with 8c/16t zen against intel i7 4c/8t.
4c/8t Zen will not match (or beat) intels latest 4c/8t or they would have to sell low clocked 8c/18t at (below) intels i7 price. But then they will have low single thread performance, which they obviously don't want - bulldozer yet again?
If they leave such CPU with unlocked overclocking, they just obsoleted their high end products, which has nothing more than some additional cache and MHz that can be diminished with overclocking. Clearly not worth the price difference that will separate Workstation CPUs from enthusiast desktop CPUs.
meanwhile harvested or artificially gimped 8c/8t Zen will go circles around 4c/8t i7 and will be miles behind full 8c/16t Zen Workstation CPU, making a server premium well deserved.
It is very tricky to have a compelling product lineup that doesn't compete with itself, making obsolete the high profit part of the portfolio.
This is the thinking that prompted me to create the topic. The 8C/8T might offer:
I. The same core/thread count as Bulldozer but with much better efficiency/performance
a) making people who bought Bulldozer and were fans of it not feel unhappy for being pushed toward a smaller core/thread count
b) possibly having more parity with the console paradigm (lots of cores/threads), making porting games easier
c) impressing the tech press and others by offering the same core/thread count as Bulldozer but with higher efficiency/performance — this is good for marketing
II. A part with strong performance that doesn't overly "eat the breakfast" of the 8/18 premium part
a) some will want to buy the 8/16 part just because it has more threads, even if they don't need those extra threads
III. The extra CPU utilization caused by having SMT on will put more pressure on the TDP (related to 14nm LPP limitations), making 8/8 potentially more efficient for less threaded workloads since it may be able to complete tasks more quickly
The overall objective efficiency of disabling working mechanisms that add to the total efficiency of a part might be unfortunate but it's also, as I said, part of the two different types of efficiency (objective and business). Stepping aside from that debate, the point #III is interesting. One type of efficiency is how quickly work is completed. There is also how long a product lasts in the market. Higher single-thread performance could trump having 16 threads since parallelism is hard to extract for the general-purpose desktop.