Server architecture fuels Intel growth
The new I2O standard, supported by Intel's i960 chip, brings minicomputer technology to servers.
Intelligent Input/Output, or I2O, creates a separate processor to handle so-called input/output (I/O) data transfers in server computers. This boosts server performance because it frees up the central processor from high-overhead, administrative data-handling tasks, say proponents. A similar architecture has been used in minicomputers and mainframes for years.
At the same time, the improvement will add little to the bottom line costs of manufacturers and, hence, customers. Intel's i960 R chip, the processor adapted for I2O use, derives from a class of chips that has been around for years, according to Tom Starnes, an embedded chip analyst at Dataquest. The i960 RD chip, running at 66 MHz, is the fastest of the two currently available R-class chips and costs approximately $80 in quantity, said sources.
"There's a little extra cost, but the benefits will far outweigh them," said Marty Strakhovsky, senior product manager for servers at NEC Computer Systems, who said some I2O benchmarks show improvement of up to 30 percent. "With I2O you don't have to worry about different driver [enabling software] for every I/O device. It's OS independent. Data transfers can go directly to memory" and no burden is placed on the main Pentium Pro or Pentium II processor, he said, which frees these processors up for tasks they are more adept at handling.
Compaq, Dell, IBM, Hewlett Packard, NEC, Gateway 2000, Mitsubishi, and others have announced products or announced support for the standard.
Generally, these servers are using 266-MHz or 300-MHz Pentium II processors with I2O functionality integrated onto a controller card. AST, which will be one of the first manufacturers to ship a product with I2O functionality when it releases its HS Premium Server this month, is utilizing two i960 processors.
The i960 chip class is the first to implement the I2O protocol, according to Wendy Vittori, general manager of Intel's Connected PC division. The chip is based on a fundamental architecture called "Data Flow," according to Vittori, which essentially creates fatter data "pipes" inside a server PC for data to flow through, increasing performance.
I2O will begin to take on more functions as it evolves, said Strakhovsky. By 1998, for instance, I2O-enabled devices will likely be able to conduct "peer-to-peer" communication without calling on the central CPU for help. Data backup to tape drives, for example, could take place without the main processor. ISDF has developed a specification for peer-to-peer communication and has placed it out for a vote among members.
Different manufacturers will also improvise on the basic architecture. Next year, NEC will use its own chipset and incorporate the Intel i960 on a newly designed server circuit board for I2O that will integrate RAID storage technology on the motherboard in its Express 5800 server. Currently, NEC uses Intel's chipset in its I2O designs.
In addition, "you will see vendors using other processors to do this function," Strakhovsky added.
Further development by Intel is inevitable, added Starnes, especially since it opens further markets for Intel. "I would expect them to do reasonably well on this," he said.
Intel is an investor in C/NET: The Computer Network.