You're still largely dealing with large, linear data rather than hopping randomly around RAM. Automatic prefetching should remain typically effective. The difference between 32 MBs cache and 8 MBs is probably going to be a few % faster performance*. You have to draw the line somewhere.
* Well, we know it is, because if, say, 32 MB meant double the performance, the design wouldn't drop it as it'd be the most cost effective way to up performance.
* Well, we know it is, because if, say, 32 MB meant double the performance, the design wouldn't drop it as it'd be the most cost effective way to up performance.