3 ms·
Hitting the Memory Wall: Implications of the Obvious (1994) [pdf]
- phkahler 12y agoThey don't consider a simple solution - static RAM. You can trade cost for memory performance at any time. Although interconnect is becoming a bottleneck, so we build bigger SRAM caches. They also don't mention how compute requirements often scale faster than the memory size. Some algorithms can be written in very cache friendly ways which take advantage of this.