I've been thinking about this. Could MS release a 360 version with a simple ARM (or Jaguar) SOC + a 32nm XCGPU (with integrated eDram) + ~ 2 GB of DDR3 RAM. Let the XCGPU run all the games, but move the OS functionality and new windows apps to the new SOC.
That sounds awfully costly to me, 2 SoCs: one likely off the shelves but still money, the other sound like a significant effort, both would have there own pool of ram.
I'm not sure it worse it, especially to be sold at 99$. If Msft were to redesign the 360 in a way they can sell it for cheap, I wonder if moving to 32 nm is the way to vs finding other way to lower costs.
I'm not sure the daughter die is worse integrating in the design, it could be shrunk though to 40 nm.
For the main chip, I'm not sure it is worse the effort to do anything this late. At least I would not try to shrink it, I wonder if they could modify the design so they can use cheap DDR3 instead of GDDR3. In the process they also could bump the amount of memory of the system. I don't know if it doable. They would need to do a lot of tweaking to make sure that the new ram behaves exactly as the old one (a bit like they recreated the latencies induced by the communication between the CPU and GPU in their last revision).
Then they may have to remove the optical media, I may have to remove the HDD slot and any related hardware. Ultimately you end up with something really "bare" even if they were to port windows RT on the devices I wonder about how useful it would be. No matter how I look at it and even taking in account lithography progress there is a lot more in a 360 than say in a Wii.
There are significant issue with the 360 design that prevent it to end in a tiny box, I think the CPU being a speed demon burn a bit too much power, even using more advanced process I would suspect that those 3 cores at 3.2 GHz may end up quiet warm => significant cooling solution.
Overall it doesn't make sense to me. At this point if they want a really cheap 360, or something compatible with the 360, they may as well design something new with BC in mind. In that case it would have make sense to stick to PPC for durango too.
If they emulation ninja can do the trick and they have the R&D budget to design a system compatible with the original 360, I think they should try to use TSMC 40nm as a basis.
The transistor density in their last revision is pretty dreadful, I don't know if it has something to do with the high clock speed at which the CPU operates but still not sexy.
Long story made short I would ask the "emulation ninjas" if they think it is doable to emulate the 360, with more xenon cores running at significant lower clock speed, idem for the GPU (add SIMD array and drop the clock speed), etc. Having followed closely the insightful pov presented during the discussion about the WiiU hardware.. I would not hold my breath...
Again sounds like a massive pain in the a** especially as they dropped the ball on PPC, adding a ARM Soc+RAM to the 360 doesn't sound doable for me at 99$.
EDIT
Another consideration is that I wonder what would be the performance of a good ARM SOC with a good GPU, running at significant higher speed than in phone and tablet, to a point where I wonder if the 360 part of the system would be that interesting / competitive.
