Xbox One (Durango) Technical hardware investigation

Status
Not open for further replies.
Are we looking at only 16 ROPs if we assume each block represents 4 actual ROPs? That would be half of the assumed Orbis number...
 
Certainly has similarities, diagram seems to imply 64 wide SIMD but what about the scheduler embedded inside the SC, and why the additional division into 16 wide units?

GCN works on 64-wide registers with 16-wide SIMD units, with 8-cycle long barrel processor. So, each cycle the scheduler pics a runnable thread from the pool, and issues it to a simd unit. Then, on the next cycle, it cycles to the next simd unit and issues another thread to it. It comes back to the simd unit it started with after 4 cycles, and issues yet another thread to it. It can only issue instructions the same thread it started with after 8 cycles.

I feel that someone who has heard this explanation and understood it "somewhat" could come up with that diagram.
 
Are you saying a CB/DB make for one RBE?

Well, an RBE has a little bit more:
render-backend-unit-360.gif
 
I was just about to post that picture! I couldn't find anything more recent than that R770 diagram for a RBE You can see how one might consider the Z-cache and depth stencil a "Depth Block" while the color cache, MSAA and blend hardware was part of a "Color Block" but that's semantics.
 
You know, there could just not be any 'secret sauce' to make it somehow perform outside of its power grade to begin with, and this is a semi normal customized console GPU at 1.2 tflops with GCN architecture :p


I guess we'll find out soon enough.
 
Last edited by a moderator:
Possibility that all tid bits of info we have been getting is stuff from multiple devices.

Of course that is if you believe that there are multiple Xbox devices coming like Paul Thurrot has mentioned a few times in the last two weeks.
 
Status
Not open for further replies.
Back
Top