Prof Aims to Rebuild Google With Stuff In Desk Drawer
http://www.wired.com/wiredenterprise/2012/01/wimpy_nodes/
I have studied process optimization for many years, almost exclusively as a hobby because there is so little need for that skill that I can’t find any jobs (at least jobs that pay well; why is it that interesting work pays least and boring work pays most?). I have felt for years that the blazing fast CPUs that are churned out today are counter productive except in some narrow situations (highly tuned optimum programs are still in a wait state some 70% of the time; routine programs might only utilize 10% of the cycles available) and it would be way better to have the CPU slowed down to the point where it can access RAM at the same speed as registers. Not only would that lead to vastly simpler chips (no need for all these levels of cache everywhere and why bother with pipe-lining or out-of-order instruction execution), but probably faster throughput for most routine programs. The logical extension of that is ‘compute RAM’ where vastly simpler CPUs are interspersed within in RAM modules so there is no latency in access. To my knowledge (sadly several years out of date), no one has tried to commercialize compute-RAM, but it would be an interesting thing to study (if, of course, I could get paid well to study it, instead of being paid buckets to do crap work). I actually started down that path of reading because for a couple of years at the turn of the millennium I was marketing a business plan to produce molecular scale computing devices (I was going to start with the equivalent of a solid-state disk). I was really interested in the idea of building a molecular scale computer and due to the complexity was studying minimal instruction sets, the RISC paradigm, etc. I felt at the time (and still feel, though as I say, my knowledge has got stale) that having reduced latency between the CPU entity and the storage (RAM) entity would be the key to vastly accelerated throughput. I remember many enjoyable hours learning and thinking about CPU design and how it could be manifest in the world of biological molecules. Sadly (as is so common in my life), no one was interested in backing my plan (I stupidly chose the height of the dot.com hysteria to market it), so I have moved on to research and prepare other proposals to be ignored by investors.
