The end of "one architecture to handle them all"
Paul Murphy summarize the <a href=http://blogs.zdnet.com/Murphy/?p=524&part=rss&tag=feed&subj=zdblog”>comparision between Niagara and the rest of the world</a>:
So why is this interesting? For two reasons: first because that 2:1 ratio for Xeon to PPC crops up a lot in other benchmark results, and secondly because this illustrates the utter dominance of the "slow" CMT approach over higher megahertz on multi-threaded task
I thougth a little bit about the article and the comments to it while shaving: The era of general purpose processor architectures has come to an and. Why ? Okay, simple example: Niagara and Cell. The both are like Arnold Schwarzenegger and Danny de Vito in “Twins”. Arnold as Niagara, de Vito as Cell (hey, i work for sun, i have to say it this way). Diametral twins. Niagara: 8 Cores, 1 FPU. Cell: 1 Core, 8 FPUs called Synergistic Processing Unit. Both developed for a clear workload in mind. Niagara for server computing, Cell for games. Two designs to solve certain problems unique to their targeted workload. An x86-Design cannot compete with both of them in their respective targeted workload. So we see more and more CPU architectures specific to a certain workload.
It is like in twins: You would not ask Danny de Vito for lifting your car when you have forgotten you car jack. But having Arnold in reach in such a moment, seems to be quite handy.