Intel looks to software to lift Itanium

The company is taking a new look at the software behind its processors to increase performance for future chips.

John G. Spooner
John G. Spooner Staff Writer, CNET News.com
John Spooner
covers the PC market, chips and automotive technology.
4 min read
Intel is taking a new look at the software behind its processors to increase performance for future chips.

Intel's Microprocessor Research Labs (MRL) established an alliance Wednesday aimed at increasing the performance of compilers for its long-awaited IA-64 processors for servers. The compiler plays a vital role by allowing operating systems, such as Microsoft's Windows, to interact with processors in the most efficient way.

A well-tuned compiler is in Intel's best interest, since it can wring the highest performance out of new chips by allowing application software, such as Web servers and transaction processing systems, to take full advantage of a new processor's speed.

The MRL, which is responsible for researching new processor designs five to eight years in advance, enlisted the Chinese Academy of Science near Beijing for help with the compiler project.

The goal of the project is to augment existing compilers, such as Silicon Graphics' compiler for IA-64 processors, to increase performance of the chips.

Intel would like to loosen up the traditionally rigid compiler-design process by creating a new, freely available compiler development framework that would allow easy experimentation. The best of the new ideas, in theory, would go to market with future processors.

"We're at a point, particularly with IA-64, where we're looking at testing out a number of novel optimizations," said Wilf Pinfold, MRL's business director. "We're looking for a way to test novel ideas and add new techniques in a modular way."

Intel's first IA-64 chip, Itanium, is considered a crucial step in Intel's efforts to challenge Sun Microsystems in the lucrative server market, but the chip has been a long time coming. Originally due at the end of 1999, the chip has been delayed several times since, while Intel worked to complete its design.

The company later grappled with clock-speed problems but now plans to release 733MHz and 800MHz versions of the chip. While production-level chips are set to ship this quarter, big-name servers using the chips won't arrive until fall. Adding to the drama, most server makers have greater interest in McKinley, the follow-up to the first Itanium chip.

But Intel soldiers on with Itanium and its associated technologies, such as compilers. The compiler's work is long completed by the time computer users boot-up their PCs, but it affects people directly in terms of the performance it lends to or takes away from a processor, either in a desktop PC or in a distant network server. A poorly written compiler can sap performance from a chip, while a well-devised one can help squeeze out extra power.

"The continued ability to make processors run more quickly is related to a few things, and one of them is the compiler," Pinfold said.

One method for increasing processor performance, known as parallelism, tackles many jobs at the same time, instead of waiting for one to finish before moving on to the next. IA-64 chips already use another performance bump, processing data in 64-bit chunks instead of in the 32-bit chunks used by Pentium chips.

Just as a person might speed things along at work by making a phone call while waiting for a PC to boot, a processor would tackle several jobs at once. It could also break a large job into many smaller jobs, which could be processed at the same time. Here, the compiler helps to direct data as these jobs are split between parts of a processor.

Intel intends to explore these and other means of boosting processor performance by making the compiler more efficient at handling data.

"There are some very interesting and innovative ways to split up tasks" between parts of a processor, Pinfold said.

The MRL, which maintains an office in Beijing, will tap the academy's Institute of Computing Technology, where engineers are particularly talented in writing system software such as compilers, Pinfold said. There, a dozen or so engineers will work on the new framework in combination with Intel's MRL compiler-research team of 10 to 15 people.

The MRL expects to complete the compiler-framework project in about two years. However, it plans to regularly post developments, such as compiler-code modules, for the researchers to review.

Analysts said improving compiler performance could boost the performance of IA-64 chips over time.

"Certainly, Intel has enough support for IA-64, but the more ideas you can get behind it the better," said Linley Gwennap, principal researcher at The Linley Group. "In theory, if this compiler technology could give better performance, it would be a rising tide for all boats."

But until Intel can prove performance improvements, analysts will remain skeptical that the project is aimed at anything more than drumming up more support for IA-64 chips.

"Without knowing exactly what these guys are doing, it's tough to see the value," Gwennap said.

By making the compiler framework freely available, Intel believes it can benefit from thousands of extra sets of hands. Researchers would download the framework and its associated software tools and test new twists on compilers designed for Intel chips.

"At this stage, you learn a lot from breadth of research. You want to test out a lot of ideas, even if some of them are a bit wacky," Pinfold said.

Ultimately, however, Intel expects to directly benefit when new ideas for compilers are tested, proven and then written into production-level compiler software, helping to increase performance of future 64-bit chips.

Typically, it takes time to perfect the compiler for a new chip, Pinfold said. Now, "we're trying to do some of that work earlier in the game."