If Zen will be based on "off the shelf" 14nm GF / Samsung variant it will be the end of AMD.
The 14nm processes available at GF / Samsung are LPE and LPP, neither suitable for a high-performance desktop CPU.
Zen needs to have at least 50% higher IPC compared to Carrizo in order to be even remotely competitive with Intel's offerings in 2016.
Since I find it highly unlikely this to happen in one generation AMD also needs to have either an clock advantage or at least 1.5x more cores over Intel designs.

Zen will be using 14nm LP (low power) process and no one knows for sure if the process is suitable for a high performance design or not.
Based on the currently known facts I wouldn´t expect Zen to reach any higher than 3.5GHz even in optimal conditions.
For all we currently know it could be restricted to frequency ranges similar to Mullins or other felidae (Cat) based designs.

Since nobody knows for sure how the Samsung developed 14nm LP/LPP FinFET bulk process works on a design which is nothing like it has been tested on before (ARM) I have to be rather pessimistic in order to avoid any further dissapointments.
I expect Zen to have initially at least 25% hit in the highest shipping frequency compared to current Piledriver based mainstream FX models (shipping up to 4.2GHz base freq.).
With the frequency hit included I expect Zen to be initially around 27% faster than the current mainstream Piledriver (up to FX-8K series) FX models. Around 58% greater IPC than in Piledriver.

AMD is a fabless company and they have no control on how the final 14nm LPP manufacturing process will turn out. How will the low power targeted manufacturing process suit a high power design, which is much larger, much more complex and targeted for higher frequencies than anything previously tested on the early versions of the process. That is and will be the unknown for a while still.
I´d say the chances for Zen´s success lies almost 50/50 on AMD and GlobalFoundries / Samsung.

Haswell already has around 45% higher avg. IPC in INT and 55-80% higher IPC in FP than Steamroller.
Zen (assuming it´s in the schedule) has to compete with Cannonlake DT chips which are around >20% faster from the IPC alone than Haswell.
Improvements (in frequency) from the newer nodes are not even taken into account (22 -> 10nm) as they cannot be predicted at this point.
In the best case Zen will be able to provide around 85% of Cannonlake DT´s performance. In the worst case the 14nm LPP is found unfit for the design and Zen will be limited to <3.5GHz frequency range.
The IPC improvements in Carrizo come mostly from the increased L1D and lower latency L2 cache.

As long as the 14nm LPP process is providing enough of frequency scalability and the leakage characteristics are as good as GlobalFoundries has advertized for their sample designs, in that case I expect Zen to be a pretty solid design. However if AMD is unable to initially* ship Zen at frequencies above 3.5GHz then their story as a CPU manufacturer ends right there (* within a year from the launch). Intel´s P1272 process variant doesn´t seem to scale good above 3.8GHz. The frequency scaling of the P1273 process used in Skylake is yet to be seen, however in order to reach the frequencies AMD needs Zen to hit in order to succeed, they need to outdo the frequency scaling of Intel´s initial 14nm P1272 process. A pretty damn tall order if you ask me.
What troubles me most is the fact that AMD has no "exit plan" for scenario where the 14nm LPP process is found unsuitable for larger / higher power / higher clocked designs than ARM. Their life depends on a process which has been developed by a third party and the process has never been tested in practice on anything even remotely similar design. Doing that requires true faith...

Just not going to happen.
- 4C/8T SMT (DT)
- IPC >30% behind Skylake
- Close to zero clock margin
The clocks will be low (~ 3 - 3.5GHz) and most likely there is nearly no margin left.
AMD has to push the clocks to the limit right at the time of launch, since otherwise it won´t be even remotely competitive.
GlobalFoundries won´t have even the 14nm LPP ramped up in time for Zen so it will be made on 14nm LPE instead (in Fab 8, Luther Forrest).
Currently the 14nm LPP is TBA/TBD at GlobalFoundries so basically there is no change it to be in time.
The clocks would have been low even with the 14nm LPP let alone on LPE, which according to Samsung is 10% worse in every aspect.

It is completely unrealistic to assume that AMD will ever again be able compete with Intel in raw CPU performance.
In order to do that AMD must be aqquired by a company with both the possibility and the will to initially spend an unlimited amount of cash.
There are only few companies with such resources and even fewer of those who are willing to do so, if any.

If Zen lives up to it´s promises and delivers the => 40% higher IPC than the 4th Bulldozer Generation (Excavator core), then it´s likely that things will improve for AMD. However the improved IPC alone is far from enough. Zen must be able to reach frequencies high enough and outdo Intel in number of cores in the same segment and price point.

If AMD fails to meet any of these Zen will be a failure and AMD in it´s current form will perish.

- > 40% average IPC improvement over BD Gen.4
- > 3500MHz frequency at launch & process / design frequency ceiling initially at > 4200MHz *
- > 6C/12T configuration
- =< Intel 4C/8T price point
* (this is the largest single pitfall, since the process characteristics are neither optimal or fully known at the moment).

If AMD would have announced that Zen will have 80% higher IPC than Excavator, this discussion would be completely unnecessary. It is certain that there will be a impact in the frequencies when compared to the latest AMD desktop offering. Either from the design itself or from the manufacturing process, but most likely from both. It´s not AMD I am worried about. I´m worried about Samsungs process and it´s suitability for a desktop / server CPU design. AMD is getting into a gunfight without knowing if their powder is dry or not.

I´ll start hoarding AMD stocks once they announce that Zen will have >=80% greater IPC than Excavator or they cancel Zen on Samsung´s 14nm LPP and port it to either GlobalFoundries 22nm FDX-UHP or 32nm SHP SOI HKMG instead. Wishfull thinking