R C Macaulay wrote: > Howdy Horace, > Mentioned quads for use to unmask encription technology. 25 years out > of game is eons. Rice U giving up on parallel computing research didn't > mean the end of the work.
Current high end systems typically run between 4 and 16 processor cores per board, closely coupled (single system image, memory shared among all processors, and cache coherency). Lack of cache coherency was a major bugaboo of tightly coupled systems 30 years ago. The shared memory bus of a tightly coupled system puts an upper end on how big you can make it. Beyond something like 16 processors you start seeing the performance nose over due to contention on the memories. Faster memories and faster buses don't really help much with this, because the performance loss is relative to what the same processors would be able to do without contention, and in a faster system with faster buses and faster memories the "no-contention" performance is also correspondingly better. For loosely couple systems, of course, the sky's the limit. IBM's Blue Gene, for example, can be purchased in configurations with many thousands of processors (contact your IBM sales rep for details). Communication bandwidth between processors is enormous but none the less they're not single system image, and communication is by message passing, not shared memory. ... >> Sounds like somebody is giving you some overblown hype on quad >> processing. I hope it is not a stock broker. It is true a lot of >> graphics horsepower (and fast monitors) will be required when >> operating systems go 3D, but this is not a technological breakthrough >> AFAIK, except maybe for price. I kind of wonder if the new Sony >> Bravia XBR7 with 240 Hz refresh rate is possibly designed for 3D >> game playing. Sony uses some very bizarre processors. See, for instance http://www.blachford.info/computer/Cell/Cell0_v2.html
