It was stated as early as
August 2010 by AMD themselves that a" Bulldozer module can achieve 80% of the performance of two complete cores of the
same capability."
Of course most people ignored this and tried to come up with very complex explanations of 1*100 + 1*80% for 2nd core....etc.
The statement was actually very clear to me. I always took it as 2 Bulldozer cores in 1 module will only provide up to 80% of the performance of 2 Bulldozer cores on their own (i.e., dedicated cores). From this statement,
on several occasions I stated that if in theory 1 Bulldozer core had an IPC increase of say 20-25% over Phenom II, at the same clock speeds, then Bulldozer would still
not be any faster than Phenom II since it would need to overcome the 20% penalty (i.e., 0.8 x 1 Module (2 BD cores) x 1.25x IPC increase = 1 x 2 Phenom II cores). I even took a best case scenario and used an average of 10-20% penalty (i.e., assumed 90% base case not 80%).
So you can just imagine that if 1 BD core was slower in IPC than 1 Phenom II core, subtract another 20% for every module and performance starts to fall drastically --> hence the requirement for 4.0ghz+ Turbo just to maintain Phenom II levels....
Technically, JF-AMD can still be right. If BD core is 10% faster per clock than Phenom II, if you have a program that runs 2 threads on 2 BD cores and on 2 Phenom II cores, the 20% module penalty would still make the Phenom II faster despite BD's increase in IPC. Unfortunately we don't have a Bulldozer CPU with dedicated cores, which makes it impossible to test the accuracy of this hypothesis. I suppose we can test some single-threaded benchmarks and extrapolate.