| PeterSilenced said: But it's the same 32 mb ...how the hell did it went from 102 gbs to 192 GBs it doesnt make any sense... |
That is the whole point. We have a journalist that got some words from a developer that got some words from some MS party that suddenly they discovered "the gpu can read and write to esram at the same time". The last part is complete and utter bullshit, of course. Reading means waiting for pulses and writing means sending pulses, down the same data line. Also you need a legal address during read/write. Even if you COULD simultanously r/w on the same data line (I wonder what happens if two pulses "meet"), you'd do it to the same legal address which is an utterly pointless exercise.
What I think (I have speculated a bit in another thread) is the possibility that the esram is dual ported. One address/data bus connects to the gpu (hence in the article they mention the "102G/s still is the gpu tranfer rate"), the other (physically separated) address/data bus connects to something else (the four dma controllers most likely). If the gpu accesses a ram bank, the move engines can access another ram bank at the very same moment. Still this doesn't answer why the number 192G/s turns up, in the best of all worlds, it would add 102G/s + whatever GB/s the move engines can max out (assuming accessing always another ram bank concurrently).
Unless someone from the design team shows up and really explains how the esram is organised (cache or addressable ram or both), and how it is exactly connected to what, all we can do is speculate and marvel at the rumours that pop up daily to make the XBox One faster (apparently "The Cloud" is already on the way out in that regard...).
