[blt]

> Has the bottleneck increasingly become memory rather than CPU?

Yes.

Besides the basic memory latency vs. CPU clock speed issue, there's also:

- CPU caches have gotten bigger, so the payoff of making an program cache-friendly is larger

- CPUs have gained SIMD vector ALUs and increased superscalar behavior, so the CPU is capable of processing more data in per clock cycle

- A cultural shift happened after the freewheeling Moore's Law era ended in the mid-2000s, and people started examining performance more closely

http://gec.di.uminho.pt/discip/minf/ac0102/1000gap_proc-mem_...

[sanderjd]

Thanks, this is what I had suspected, but it seems to be left unstated in everything I've read.