End Of Specialized Supercomputing
Intel's Itanium microprocessors were once a highly anticipated revolution in CPU technology. As far back as 1994, then codenamed 'Merced,' the Itanium processors were supposed to leave the baggage of legacy compatibility behind. It was designed to bring 64bit computing to the desktop, a market dominated at the time by DEC Alpha. In 2001, the first batch of Intel Itaniums ran at 733Mhz, addressed 16GB of RAM and were capable of processing a theoretical 6GFlop/s. These numbers are impressive for their era considering IBM's Power5 architecture targets those numbers today and is touted as super-computing for masses.
Today, Itanium CPUs generally only see use in supercomputer applications in massive Linux clusters. Beowulf clustering software has eliminated the need for custom, proprietary processors. Now, the top supercomputers are built from 10,000 to 100,000 linked CPUs that nearly break the 300GFlop/s barrier. Dedicated, non-mainstream processors seem to be losing market share, a steady trend since the late '90s. DEC Alpha, Sun UltraSPARC, Intel Itanium and even supercomputing giant Cray have all lost major representation against clusters of commonplace Pentium 4 and Xeon. These low budget Intel systems comprise more than half the Top 500 supercomputers. Cray has thrown in the towel and committed it's new series of supercomputers to run from clusters of AMD Opteron components. Given this trend, is there any real purpose for Sun to unveil the new, 8 core UltraSPARC T1 even after converting their corporate strategy to use x86? Should IBM continue their development of Power5 with Apple's support gone? Is there a benefit for Intel to pump research into Itanium?