Intel's Itanium can't seem to get a firm toehold anywhere.
The Itanium chip family, which Intel has relegated to high-end servers, has rapidly declined on the Top500 Supercomputer list. In November 2004 the list had 84 computers with Itanium 2 processors. By June 2005, the number shrunk to 79.
Now only 46 computers contain Itanium 2 chips, according to the latest list, released on Monday.
Meanwhile, the number of supercomputers using AMD's Opteron chips has increased. A total of 55 Opteron-based computers made the list, up from 25 in June. (Opterons were found in just 29 computers on the November 2004 list.)
The number of supercomputers using Intel's 64-bit Xeon chips grew slightly to 81 on the latest list, from 79 in June. The chips only came to the market a year ago.
Although the list is fluid and changes considerably every few months, Itanium's dip on the list is bad news for Intel. The decline could reinforce the gloom that surrounds the chip and be seen as a harbinger that computer architects are losing interest in it.
An Intel representative cautioned about reading too much into the numbers. A new, albeit delayed, version of Itanium 2, code-named Montecito, is coming soon, which could reinvigorate Itanium's position.
"Overall Intel's presence on the list remains strong. We remain the predominant processor provider on the list," the representative said. The company's processors run 333 of the top 500 machines.
"It is more symbolic than anything else, but the top 500 does give you an insight into where the mainstream architecture is going," said Gordon Haff, an analyst at the Illuminata research firm. "When there so many questions raised around Itanium, there is more of a tendency to take a glass-is-half-empty point of view."
The decline of Itanium can be chalked up to a few factors, Haff stated. Silicon Graphics, a major backer of Itanium, is having difficulties. And HP, which helped design Itanium, has been shifting its weight to Opteron; in fact, many of the new Opteron systems on the list come from HP.
Itanium hasn't had a strong fit with how researchers are building supercomputers, Haff said. The chip is best suited for heavyweight clusters, with servers containing 4, 8, 16 or more processors each.
However, system architects now prefer to build large clusters from more lightweight servers with one to two processors connected to each other over high speed links.
The Itanium saga can be seen as a tale of soaring ambition versus commercial practicality. In the late '80s, Intel and HP outlined plans for Itanium, in part to correct some of the limitations of the x86 chip architecture, which was the basis for Intel's chips at the time. The new Itanium chip would also allow Intel to get into the market for server chips.
By the mid-'90s, however, Intel improved the performance of its x86 chips themselves; in fact, the company enhanced the older chips enough to start selling them to server makers. In a few years, the company's market share of server chips went from under 3 percent to over 80 percent.
And, during that time, Itanium suffered delays. The first version suffered from performance problems. Subsequent versions improved on performance, but customers and developers remained lukewarm about taking on the task of converting their software applications to run on Itanium.
By 2003, Itanium prospects began improving. In June of that year, 19 of the top 500 supercomputers ran on Itanium, a figure that jumped to 32 by November.
But then AMD introduced its Opteron chip, which weakened one of Itanium's big advantages — running 64-bit software and enabling computers to contain tremendous amounts of memory. Opteron runs 32-bit software, found on desktops, as well as 64-bit software. Intel followed with its 64-bit Xeon chip.
Over the years, Intel began to circumscribe Itanium's potential market. The company planned to sell the chip for a wide variety of servers and workstations. Then Dell and HP dropped Itanium workstations, so Intel stopped plugging it for those tasks. At the same time, Itanium went from being a server chip to only a high-end server chip. Sales remained low.
"With the benefit of 20/20 hindsight, there would not be an Itanium today," Haff said.
