To extend some of my recent posts regarding quad core and above solutions for general purpose computing, consider the currently available quad core desktop solution: AMD's Quad FX. Intel partisans may claim that this is not a true quad core, but single die, dual die single socket, and dual die dual socket with HT are simply partitioning choices on the same beast. One can argue that Hypertransport beats the FSB, and equally that MCW is ahead of Athlon 64 X2 on raw performance, but this is beside the point. When you examine independent reviews of the Quad FX, they seem to boil down to one theme: It excels, but only when you run a lot of simultaneous applications on it. This is supported by AMD's marketing language:
"The Revolutionary AMD Quad FX platform is the first dual-socket, multi-core PC platform designed to take advantage of the latest enhancements in Windows Vista™ Ultimate and for megatasking PC enthusiasts and power-users who run the most demanding tasks simultaneously. The AMD Quad FX platform with Dual Socket Direct Connect (DSDC) Architecture is ready for the most demanding multi-threaded digital content creation software and for the next wave of incredibly demanding multi-threaded games... With the AMD Quad FX platform you are ready for the multi-core, multi-threaded future... Designed to perform with multi-threaded software."
The implication is that this "multithreaded future" is imminent, and that many of us will benefit from "megatasking", running lots of single threaded applications all at once.
The reality is that, barring servers, a few specialized content creation applications, and a small number of yet-to-be-released games, anything beyond dual core is going to be of limited utility for a long time to come. For the average individual, either at work or at home, whether mobile or desktop, quad core and above is goofy. Like the Megahertz madness, this will eventually become obvious. Intel would like us to miss this point for as long as possible, and emphasize a new round of "core wars", since if it is really still about single threaded IPC * clockspeed for single and dual core, parity is restored in Q3 2007 and there are no obvious magic bullets waiting in the wings. The litho barrier beyond 32 nm w/o EUV was bad enough, but that was still a few years away. If CPUs stop offering meaningful improvements to the vast majority of users starting mid next year, commoditization sets in, and the producer best able to support low ASPs gains share. Wonder who that is? |
