AMD Athlon (Carrizo)
x4 845 Interrogation
You could call it a performance review, I guess...

Results Overview
The chart on the right represents the performance of each system tested (minus memory and gaming results), relative to Kaveri. All systems were clocked at 3GHz, and each benchmark was run a minimum of three times - the Kaveri results were run about six times due to an odd BIOS upgrade issue which gave misleading and confusing results until a reflash.

What may not be so apparent from this chart is that Excavator, from all of these results, averages just 7.64% faster than Steamroller... and just slightly slower than the K10 core in the Phenom II.

These are not particularly great results, taken at face value, but they don't tell the whole story, either.

The x4 845 is essentially a mobile Excavator-based CPU brought over to the FM2+ package for some unknown reason - presumably to harvest dies with defective GPUs or other logic not used on the desktop form-factor. It has half the L2 cache of Steamroller without improving its performance anywhere near enough to make up for the reduction in capacity... yet it's still faster.

This reduced L2 cache size (1MB vs 2MB) has a measurably negative impact on performance in certain cache-heavy workloads - particularly multi-threaded workloads. This means the IPC increase is higher than the aforementioned 7.64% and that the actual core improvements will be higher still. And that doesn't even take into account that the benchmarks in that list are over-emphasizing certain aspects of performance.

One thing is very clear: Sandy Bridge remains well out of reach for anything AMD has currently on the market. In fact, Zen's claimed 40% greater instructions per clock (IPC) over Excavator will not guarantee a CPU faster than Sandy Bridge let alone Ivy Bridge, Haswell, Skylake, or any upcoming Intel architectures, if this chart is to be taken at face value... but things are rarely so simple.
Results Relative to Kaveri
In order to compare CPUs "properly," you, first, need a lot of benchmarks. Then you need to know which ones to compare, or maybe even discard, due to their failure to be representative of real-world programs - or their particular favoritism towards a certain capabilty, architectural feature, or instruction set.

To see how I address this, just continue to: Carrizo vs Kaveri