No one is declaring this information pointless, but it's certainly a lot less relevant to Every Programmer than it was when this was written in 2007. I've actually got an okay understanding of this stuff, but the number of times that I've split
for(int i = 0; i < 10; i++)
{
j += i;
k += i;
l += i;
}
into
for(int i = 0; i < 10; i++)
j += i;
for(int i = 0; i < 10; i++)
k += i;
for(int i = 0; i < 10; i++)
l += i;
to reduce cache misses is very small because with today's hardware, it very often just doesn't matter. Obviously there are many programmers whose primary job is to do this, because they need every clock cycle they can squeeze out of the thing. They're just not representative of Every Programmer.
I'd argue it's more important now than when the white paper was first written. Where processor speed or memory size used to be the bottleneck, increasingly these days it's memory bandwidth instead, so managing that effectively is key to writing high performance programs.
I was going to say something almost precisely like this. The main thing I was going to add is that it is both memory bandwidth and latency that have become important bottlenecks, and often the latter is more important. You know how people were saying "disk is the new tape"? Guess what the new disk is? RAM. Much as with disk latencies, RAM latencies haven't really improved that much over the last decade or so, though memory bandwidth and CPU power have. Guess what that means? ;-)
..but you can only know this if you understand what's going on underneath the hood. So if you don't, no need to worry; you won't even know when the moment of your utter irrelevance arrives.
