Huge 3-D graphics emergence seen

Electronic News, Sept 2, 1996 by Jim DeTar

Santa Clara, Calif.--Although microprocessors continue to follow Moore's law to higher levels of performance in terms of operating frequencies and integration of features, 3-D graphics stands out as one of the applications that will gobble up all of the MIPS available and cry for more, according to Richard Wirt, an Intel Fellow and director of the Intel Microcomputer Labs.

In an interview, Dr. Wirt said that the labs, formed a year ago and focused on development of new MPU architectures, are bidding to assume the mantle of MPU development leadership that once was on the shoulders of mainframe makers such as IBM and minicomputer vendors such as Digital Equipment.

"In the past we've had higher-end systems above us and many of the ideas such as cache and dynamic execution have been tested in high-end mainframes," Dr. Wirt said. "We sort of knew the direction ahead. Today, most of the mainframe and supercomputer business has gone away. The responsibility for changing the course has fallen more on our shoulders."

Intel's Microcomputer Labs was initiated a year ago to focus the company on innovating new MPU architectures. "We currently have 70 people and that will continue to grow. We have not published a target figure we expect to grow to but we felt we needed 100 people to get to critical mass," Dr. Wirt said. "We have hired about 50 percent externally. One thing we've done intentionally from the start is to build our efforts close to the product teams. Like some other research labs, we have to affect the products. We're not off in a corner; we're effectively tied in with the products."

Although Intel has extensive research facilities in Santa Clara and Oregon at its Intel Architecture Labs (IAL), this is the first facility chartered to look at next-generation architectures. Intel's flagship x86 architecture was originally based on a Digital design, Dr. Wirt said.

"Prior to the Microcomputer Labs, I was part of IAL. We did some reorganization a year ago when we created this lab. IAL is focused on the PC and this (group) is focused on the platform. They were never on the next-generation microprocessor, strictly the platform. They've also been involved in trying to develop standards. We have a process research lab in our Process Technology group, distributed in Santa Clara and Oregon," he noted.

Now, as the x86 nears the end of its cycle with the planned introduction in a few years of the so-called Merced chip--which will combine the x86 with Hewlett-Packard's PA-RISC architecture--Intel is looking for other architectures to drive emerging applications, such as 3-D.

"We think 3-D graphics is the one that is going to consume a lot of MIPS. The Toy Story movie took 120 workstations each computing typically five to eight hours a day, all for a 1-1/2-hour movie. And that's only surface rendering. In medical image rendering, you want to go through the entire body," which will eat up even more processing power, he noted.

Parallel computing is one of the areas that the lab is researching to meet emerging computing challenges. "Ultimately it's very important," Dr. Wirt said, adding: "It's more a question of time. We can put hundreds of millions of transistors on a chip; you have to look at how you organize these. Today, people have one stream of execution. They are using pipelining and superscalar to get more and more parallelism. Ultimately, that has a limit determined by application and compiler technology."

Compiler technology will play a ever-increasing role in the architectures of the future. "The compiler is becoming more and more part of the architecture," Dr. Wirt said.

"We started compiler work back in '89, trying to get highly optimized compilers. We worked closely with the Pentium team. The concept got us more parallelism. From that, we said we need to step that up and begin looking at it from other application views. Now a big one is graphics and visualization. Graphics can consume as much computer power as you can throw at it today. We have built specialty teams in those areas. Three-D is an example of that."

The idea of putting logic and memory on a single chip is another idea being discussed in several quarters. For example, at the recent Hot Chips conference at Stanford University (EN, Aug. 26), David Patterson gave a presentation on what he called IRAM (intelligent RAM), the concept of putting logic on DRAM; and other companies say it's time for this to happen. Among those companies is Intel. This integration will create problems of its own, however, according to Dr. Wirt.

"This is attracting a different problem. Today, microprocessors have gotten a lot faster than memory. If you grade the (speed) curve of microprocessor gains versus memory gains, the memory curve is almost flat. Every time I go out to get memory, I lose performance.

"Cache has hidden that for a while. The fundamental problem there is that the two processes--although they may have had common roots--have wandered apart. The fundamental process that logic is built on and DRAM is built on are different. DRAM people have looked at that for the back end of memory chips. We, the logic people, have looked at that and will continue to evaluate whether it's the right approach for now."