when you talk about on-die framebuffer are you talking about esram? because it should be NOT a framebuffer and should be a separate block (as separate block was and still is the eDRAM on 360)
It's not a separate block, it's going to be integrated into the die itself. And yeah, it'll have the framebuffer (and/or various render targets, Z, G-buffer, shadow buffers etc) stored on it, of course it will. It offers the highest bandwidth of all memory in the system, and rasterization is the largest consumer of it. Texturing is also high, but 32MB is too small to fit all textures for a scene these days. It's what makes sense.
and yes the 12 CU's block should be the exact 12 CU's block on 7790, exact, same type, same technology, all the same.
"Should", yeah, just like the wuu's GPU should have the exact same shader units out of the radeon 4000 line since that's what it's based on (except it doesn't), so how can you be so sure? How many console GPUs have you designed in order to know this?
In any case, the shader block isn't the only component of the GPU of course and just because the shaders may be capable of clocking at 1+ GHz doesn't mean the rest of the logic on the GPU does. AMD GPUs use one single core clock, likely the eDRAM (SRAM, whatever) runs at core clock too, as well as any other custom logic designed specifically for durango. Is all of that going to be capable of reliably clocking 20% faster than the suggested speed? Unknown, yet you seem very certain for some reason, which is strange since there are literally no fact to back that up.
The higher you clock things the lower functional yields are going to be, and trying to "solve" it hobbyist-overclocker fashion by whacking up the volts isn't going to be a reliable solution for a company like microsoft that will sell millions of units. They've already gone through one red-ring fiasco that cost them multiple $billions already, I'm confident they never want a repeat of that again. (Also, bunches of original xboxes caught fire and so on, due to poor quality power supplies and whatnot.)
And if you don't like speculative discussions, you're in the wrong thread.
However you're not speculating, you're throwing out rigid assertions without any basis in fact, which is something different than just speculation.
Even more, a 1 GHz 12 CU's block should be less hot than a 0.8 GHz fat 18 CU's block
Depends on the voltages you need to pump through the chip. If the 18CU chip runs at 1V@800MHz and the 12CU chip at 1.175V@1GHz it's not as clear-cut anymore. Anyway, hot-spots on a chip is more of an issue generally than the total heatload of a chip, and as you ramp the volts heat quickly ramps up.