no specs yet, can't say

The instant problem of doing physics in the console GPU is, of course, that your GPU no longer has enough power to "make pretty pictures". If you want 1080P/60Hz as your target, you'd better put a high-end GPU into your console... Technically, an A10/HD7770 combo would still be ok, do the physics on the A10 and forget all about that pseudo-xfire stuff.

That's why I said it depends on how much the offload would be, for example, Wii U setup with a supposed weaker CPU. Nvidia does just fine with their stupid physx bullshit, I say stupid because they won't make it open to AMD GPUs and Hybrid PhysX has been put on hold since the dev's old ass Nvidia card broke. the APU+GPU setup would be a great setup though, then like you said, they can do physics on the APU's GPU and use it to assist the more powerful GPU in other areas as devs see fit, we just have no details yet.

Because while the CELL is 10x better at the brute force floating point calculations needed for modern rendering than a traditional CPU. A modern GPU is over 100x better at that type of workload. 

The CELL is basically like a half way point between a modern CPU and GPU in terms of design. But while it may offer high parallelism and massive amounts of floating point math compaired to a  traditional CPU and much better interger performance, memory managmeent, branch prediction, flexibility and per thread performance compaired to a GPU. The CELL also offers poor floating point performance and parallelsim compaired to a GPU and poor memory management, integer performance, flexibility etc compaired to a traditional CPU. It's a jack of all trades master of none.

Now there is a trend towards heterogeneous processors being the future of computing. But that relies on the need for flexibility to outway the need for efficiency. A specialised peice of hardware will always be able to outperform a general perpouse one for the same price, or even a fraction of the price. And it seems the future will be more like a modern GPU than the CELL anyway, even for heterogeneous computing. 

The best you can throw at an A10 APU is a second HD7760, you get about 150% performance of the single A10 (on PCs).  You will fall waaay short of a high end PC. Most game benchmarks I've seen can't reach the targeted 1080p/60Hz goal, sometimes considerably. It is the old problem of crossfiring two GPUs. Usually you are better off buying the better single GPU than Xfiring two weaker GPUs (and pseudo-xfiring is not nearly as good as "true" xfiring). This all applies to PCs, of course. Direct drivers on the register level will considerably improve speeds on A10 consoles, but high quality 1080p/60Hz games.. I just don't see it.

I lost interest in following the PhysX stuff. It started with dedicated PPUs (Aegis or whatever it was called), dedicated graphic cards, then went to doing it in existing GPUs. The former solutions are too expensive for average Joe, the bechmarks I'd seen of the latter had (sometimes very) dramatic decreases in fps (no surprise here, and I think only a few games actaully tried it).

But like all CPU's and GPU's that go into gaming consoles (and now APU's), they're all modified slightly somehow to benifit the console. So a modified A10 and modified HD7760 should provide us with graphics we have yet to see on a PC right?

Oh also it's also worth noting Intel was once developing a CPU similar to the CELL in many ways codenamed Larrabee. Which was meant to have 32+ stripped down x86 CPU cores (with a dedicated texture unit in each core, 4 way multithreading with in order execution, and a very wide SIMD unit for floating point math) back in 2010, in 2009 a prototype offered 1 TFLOP at 1GHz well over 4x that of the CELL (graphics performance would likely have also been much better than the CELL thanks to the dedicated hardware as well) after sinking $4-4.5 billion into the project they decided to roll the tech into their traditional CPUs and intergrated GPU projects and skrap the project as performance was not competitive with modern GPUs.

No (and BTW it's 7670, not 7760) - combined performance of that combo (A10-5700 with 7660D@760MHz) is 12.5GPixels/s; 37.4GTexels/s; and 1352GFLOPS. No amount of "modification" can put that to level of 7970 with 33.6GPixels/s; 134.4GTexels/s and 4300GFLOPS (which is, if we take a look at ATI GPU history and recent rumours, almost what will 8870 will be pushing when it launches probalby early next year). Check out Superchunk's thread, I posted table there several days ago for quick referencing:

http://gamrconnect.vgchartz.com/thread.php?id=136756&page=87

I'm afraid all next-gen consoles will be vastly inferior to high-end PCs, even at launch, not to mention few years from now. Which for PC crowd could be either very bad or very good thing, depending on what approach (some) developers take.

i think developers aim only for one thing: money. yes, a high-end-pc is most probably even better by now than what we can expect from next-gen-consoles but i am sure games will be developed for consoles due to very good sales-numbers. well, pc-sales are hard to count nowadays with digital distribution but i highly doubt that assassin's creed 3, for example, sold nearly the numbers it did on consoles (not to mention both consoles combined...). i think the situation will be like it was the last years: the pc *is* of course technically more advanced but 1. it is not really used by games and 2. it is quite expensive to develop a game especially for consoles and again for pc so i think pc will still get ports the next years, with polished graphics.