One of the cuda support documents I read yesterday stated this:
For a 384 bit memory interface which works at 900Mhz, the maximum speed data transfer speed obtained is 384 * 1800/8 = 86.4 Gbps.
What is the reasoning behind this calculation ?
384 bits * 900 Mhz * 2 (DDR) / 8 (bits per byte) / 1000
What does 2(DDR) mean…
And why divide by 8 bits per byte when the answer (86.4Gbps) is itself in bits per second and not bytes per second.
DDR memory transfers 2 times per clock cycle, so at each clock tick, high and low flank of the signal. Thats why you take the basic clock rate times 2 for DDR.
I think the 86.4Gbps is a type. You have 86. Gigabytes per second on an 8800 GTX
Johannes
and that is giga as in 1e9 to be completely accurate ;) (1e9, not 10e9 as I so optimistically wrote :P)