The Itanium saga has gone on for more than a decade. In December 1998, Hewlett-Packard began work on a new chip architecture called EPIC, said Dick Lampman, leader of the effort and now director of HP's Labs. The project was cloaked in secrecy, with a "black budget" that made funding difficult to trace. Around the same time, Intel was planning its future. The company was making chips that crunched data in 32-bit chunks, but saw that it would need 64-bit chips. In addition to processing twice as much data at a time, 64-bit chips also let a single processor access data from larger amounts of memory -- an important consideration for running databases and other large applications. The 32-bit chips, such as the Pentium II, III and 4 and all the existing Xeons from Intel, can juggle only 4GB of memory. Additionally, 64-bit chips can handle tremendous amounts of memory. Some Intel design teams advocated getting to 64-bit chips by merely stretching the existing 32-bit architecture, sources said, a relatively easy leap in terms of engineering. And as an added bonus, both chip families would be able to handle the same software. Intel performed this sort of stretching magic when going from 16-bit to 32-bit chips in the early 1980s (and likewise when it moved from 4-bit to 8-bit to 16-bit chips in the 1970s.) But for the 64-bit chips, the company decided to work with HP on building processors around the EPIC idea. A new software base would have to be created for EPIC chips, but the switch would also let Intel break free of many of the inherited twists in the "IA" architecture, the design behind all the other the Intel processors. By the mid-1990s, the future EPIC chips, known by the code name Merced, loomed as a danger to Sun Microsystems, Digital Equipment, IBM and other server makers that also made their own chips. With its mass manufacturing capabilities, Intel would be able to pump out low-cost, high-speed 64-bit chips faster than these individual manufacturers, analysts said at the time. Making Merced, however, proved difficult in practice and, after several lengthy delays, the chip went from being a giant killer to a conceptually interesting test vehicle. The chip finally came out commercially in May 2001. Performance was relatively anaemic, and sales were even worse. In the third quarter of 2001, less than 500 Itanium servers were shipped, according to IDC. So what's new? But although the first Itanium was a flop, newer versions could succeed. For one thing, Intel still wields a larger manufacturing base than any other chipmaker, giving the company an inherent advantage when it comes to lowering costs, accelerating speeds or developing multiple products simultaneously. "In the long run -- even in the short run -- our capabilities in manufacturing will keep us ahead," said enterprise-chip marketing chief Hambrick. A four-processor Itanium 2 server with 8GB of memory will cost around $41,000, less than competing Sun servers. In contrast, a Sun Fire 3800 with four 900MHz UltraSparc III chips with 4GB of memory goes for $133,000, according to Sun's Web site. Hambrick says that on benchmarks the Itanium 2 system will outperform this machine. Lower prices, combined with improved performance, will certainly make Itanium 2 more attractive than its predecessor, said TimesTen's Shetler. "It looks more promising," Shetler said. "With the recession, it becomes even more of a viable option." Sun disagrees, saying Itanium will be the more expensive option. "The migration costs are going to be very big, there is very little software available and it's an unproven architecture," said Martin Chorich, a Sun spokesman. "This is the second time out with a demonstration vehicle." Another factor is that Intel has put tremendous amounts of energy and money into building a world around Itanium. The IA-64 fund invested approximately $250m in start-ups to get them to develop Itanium applications and worked extensively with established hardware and software manufacturers to get them to make Itanium-based projects. Consulting and testing labs were set up around the world so large corporate buyers could tune their applications for the new chip. Intel will even build complete servers for manufacturers who can't spare the engineers. This option is expected to be popular with manufacturers in developing nations, such as Russia's Kraftway, said Phil Brace, a marketing manager for Intel. The company has also straightened out many of the kinks that held back the original Itanium. The photos of the core of the original Itanium looked like a "collage" of chip components rather than a tightly integrated processor, Krewell said. "They slapped the first one together to get it out the door after it slipped so badly," Krewell said. Itanium 2 is vastly improved. It contains a 400MHz bus that's 128 bits wide, compared with the 64-bit wide bus on the original Itanium, a change that allows for greater data transfer rates. The level three cache, a reservoir of memory for rapid data access, goes up to 3MB. The "pipeline," which serves as a kind of assembly line inside the processor, is also shorter. In addition, improvements have been made to the compiler, software that organises and schedules how different resources on the chip will get deployed, Krewell said. "In the technical computing space, they need the extra performance," said D.H. Brown's Ghatpande, predicting that scientific institutions and life sciences companies will adopt the chip relatively early. Security will also improve. Like other chips, Itanium 2 can wrap data within varying levels of security. Itanium 2, though, can hold more data at the highest security levels because of data management techniques developed at the University of Washington, HP and Intel, according to Jerry Crawford, Itanium's chief architect. Despite the improvements though, the fact remains that performance and security considerations are secondary for many corporations, said Dan Kuznetsky, an analyst at IDC. And even cost isn't necessarily the deciding factor. Companies also take a serious look at the headaches involved in incorporating new breeds of hardware and software into their organisations. Even if Itanium 2 servers cost less, businesses aren't inclined to begin the arduous process of installing them for real-world use. "Experimentation with new hardware, I suspect, is a luxury that few companies can afford," Kuznetsky said. "If they do anything at all, they will buy a small number of systems to test." And besides, existing hardware is sufficient to solve most people's problems. "The current RISC systems...provide the power we need," said Nate Robertson, of Clark County, Washington's government technology purchasing office, which opts for HP Unix servers.
See Chips Central for the latest headlines on processors and semiconductors.
Have your say instantly, and see what others have said. Go to the Chips Central Forum.
Let the Chips Central editor know what you think by email. And sign up for the weekly Chips Central newsletter.
