Even when you consider the 100MHZ increase in clockspeed and an extra 4 threads the Core i7 2600K is under 10% faster than a Core i5 2500K in PCMark Vantage:
http://www.neoseeker.com/Articles/Hardware/Reviews/Intel_i7_2600K_i5_2500K/8.html
Yeah, that's kind of what I was getting at about PCMark. Wait for other benches!
As for those who think the Bulldozer module is closer to hyperthreading than to two cores, think about what task the Interlagos processor (from which Zambezi is designed from) was designed for. In the server space, Integer performance is more valuable than floating point performance, and while the module may only have one fetch/decode unit, it has two integer execution units. Further, each has their own L1 Data cache, and the L2 cache is able to be dynamically divided amongst the cores in a module (even unfairly). As such, for integer processing, it performs about as well as two cores without any components shared.
As for floating point, there may be only one FPU, but with each cycle, either core can operate on 256 bits of parallel data via two 128-bit instructions or one 256-bit instruction, OR each of the integer cores can execute 128-bit commands simultaneously. This is actually an advantage over Sandy Bridge, as while Sandy Bridge may also support 256bit AVX instructions, should a program not be programmed to take advantage of those, it is limited to a single 128bit FPU. Further, bear in mind that a 256bit FPU takes up a huge amount of die space, and so sharing it is an important way of saving chip real estate (put it towards the massive L3 cache these chips will have).
AMD "averages" their module design performance by saying that if both cores in a module are active, you'll get around 80% the throughput performance of having two separate cores without any shared resources, but at a significantly lower amount of power consumption. On the other hand, looking at Intel's Hyperthreading, you're lucky to get more than 15% throughput increases on an out-of-order execution processor, but still with a noticeable increase in power consumption. Hyperthreading is much more beneficial on in-order execution processors like the Atom or the Xbox 360s PowerPC CPU, as the pipeline is stalled significantly more often.
For Zambezi at the desktop, one also has to take into consideration how the power saving settings are in the OS (likely configurable by an Administrator). Say a Bulldozer's cores are set up like this (grouped in twos):
12|34|56|78
One can choose from a maximum power saving profile, or a maximum performance profile. Let's say we are executing a program that takes advantage of a quad core, and no additional cores (like many games). If we're interested in maximum power saving, we'd see these cores activated:
12|34
On the other hand, if we're more interested in performance, we'd see this:
1|3|5|7
In the latter case, no resources would be shared (other than L3), and we'd have each individual core running with a full 2MB L2 cache and complete access to the FPU. As most games don't take advantage of more than 4 cores, and because we are enthusiasts here, we'll probably select that option. But others (especially mobile users) will likely opt for the first.
