MikeB on 17 September 2007
@ Final-Fan
Here I see you making a "highway" analogy, but I don't see where you connect it to any facts in particular. What brings you to the 300+ conclusion?
"The command bus streams commands fast enough to support 102.4GB/s for coherent commands, and 204.8GB/s for non-coherent commands. The data rings can sustain 204.8GB/s for some workloads, with transient rates as high as 307.2GB/s."
http://www.ibm.com/developerworks/power/library/pa-fpfeib/
Isn't it true that, according to the 360's diagram, the GPU can give instructions to the logic center and it can work out the calculations at the 256GB/s rate, giving the data back to the GPU at the 32GB/s rate? I am not an expert.
The eDram internal logic to its internal memory bandwidth is 256 GB/s, important bottlenecks are eDrams memory size and shared (CPU/GPU) bus.
From the 360's diagram it looks like the RAM must go through the GPU to the CPU (at the 10.8 read/10.8 write rate). Is this correct? If so, it's pretty ironic that fazz would point out the "weakness" of the XDR's need to go through the Cell to the RSX at a 20 read/15 write rate.
Correct, in terms of bandwidth the XBox 360 design is bottlenecked.