| PLEX86 | ||
winscape 2364winscape 2365 It is more a hybrid of the paths Cray and the Transputer has taken. The heavy graphics tasks are pretty well... I think this is a little imprecise. In terms of instruction types, x86 has been usably extended to 64 bits by AMD64. In terms of instruction throughput (IPC), x86 has been only slowly improving since the PPro & K6 triple-issue cores 10 years ago. AMD has been making more incremental improvements. Intel made a radical departure, reverting to a simpler-deeper pipelined core for the Pentium4 in the hopes the clock would cover. It didn't, and they are rolling back to the PentiumM. But x86 has always counted heavily on process improvements. These definitely have slowed recently. Leakage has prevented voltage from being decreased with feature size. What was an n-squared improvement in heat-power has become merely linear. In a very real sense, I'd say all CPUs have run out of steam. Seymore Cray's solution. It works when the software is parallelizable or the tasks inherently are (servers). Unfortunately, many software and tasks are not. Lots, but much of the current dies is used for cache. Linux runs on ARMs just fine, but I think ARMs would need significant expansion to handle multiprocessing, particularly for MOESI cache coherency. Shared cache (a'la Pentium I) is possible, but it will run out of bus. For these tasks, certainly. But many users are just waiting for MS-Windows to repaint the screen faster. Or reboot. Unfortunately, this is where the volume market is. -- Robert
|
||||
Alt Folklore Computers from Newsgroups The #1 Usenet Provider on the Internet
|
||||