Intel Corp (INTC)

[mmoy]

Always good to have a refresher on Nehalem.

Slide 23, in particular, caught my eye in that it's the slide on fast unaligned cache accesses as that's what I'm playing with right now (not the cache part but memory accesses in general). The slide indicates that unaligned access performance in Nehalem will improve over Core 2. But it doesn't say by how much and the slide implies that the performance will not be that of aligned instructions. This will make life easier for programmers (yes, programmers are lazy) so that they don't have to align structures but the best performance will probably come from aligned structures.

There's no indication on whether there will be unaligned performance improvements with integer instructions. AMD's unaligned access on integer instructions (I assume FP instructions too) is very good which is why VS2005 uses 64-bit integers for bulk memcpy operations.