@jxx1998 I think there is a fixed amount of bits for each transfer, so if we request less bits than that, it is likely that those extra bits will be wasted.
I thought that burst mode is helpful only if we're requesting large amounts of data from the same row at a time. That way, we won't waste bits, and in fact we will improve on the previous slide since there is more frequent load latency.
Burst mode is an optimization over the common case. It amortizes the precharge and row activation steps for large transfers. It is a type of prefetching that hides latency. In a high frequency core, the hit we take on potential wasted cycles is much less than the hit we take on potential extraneous precharges/activations (which take >> 1 cycle).
Please log in to leave a comment.
I am not sure that I quite understand the idea behind "burst mode". So what if an application only requires 8 bits of data at a time, at different rows? Then if DRAM bulk transfers the bits, wouldn't those extra bits be wasted since the application doesn't need them in the first place?