Well, a 256-bit bus will give 192GB/s with 6.0Gb/s (highest manufacturer rated/power consuming at this time) RAM. Getting that high clock rates will impose a certain complexity to the physical I/O though, as we've seen on Tahiti/Kepler.
Hey thanks Al, now I wonder where I got ~220+GB/s bandwidth for GDDR5, maybe I just misread some figures somewhere.
Also I thought I read they reached 7Gb/s, no? Thought I saw some article when I was looking at memory months ago. I guess it doesn't really matter if complexity becomes an issue. =p
I guess that supports my assumption that we'll probably not see BC on the next xbox either.
I see 256-bit GDDR5 at around 130GB/s on some PC cards while the eDRAM is 256GB/s. But I assume it is not simple 1:1 math to emulate it.
I always thought that if you matched the bandwidth, that's all they needed on that end. You could be right though that it isn't that straight forward.
From what I understand, DX11 allows for more bandwidth efficient techniques, so developer can take better advantage at the bandwidth they do have.