The little chip that could

For most people who work on a PC, the Pentium processor is just the speedy chip inside. For Intel, the Pentium chip family, which celebrates its 10th anniversary this year, has been a blockbuster success, and an ongoing saga of learning and improvement.

John Crawford, an Intel fellow who co-managed the design of the original Pentium processor after serving as chief architect on the Intel 386 chip, remembers the daunting challenge faced by the original Pentium design team. Earlier teams designing the Intel 486 processor had figured out how to increase throughput via high-performance pipelining — breaking instructions down at every step and balancing them at every instruction cycle. “So the issue with Pentium was what to do next,” Crawford recalls.

At first, the design team tried to “double the pipeline,” or run two pairs of instructions through at every clock cycle, but this wasn’t effective. Then they tried applying “branch prediction,” or “predicting breaks where you might lose performance, resteering the front end of the pipeline speculatively,” Crawford explains, and that worked better.

“It was the fastest microprocessor development project I’ve ever been associated with,” remembers Crawford, who says the Pentium design effort, involving a team of more than 100 engineers, took almost three years from its beginning to the product’s March 1993 launch.

But the Pentium was not yet home free. In 1994, Intel was hit by “the floating point divide fiasco,” as Crawford calls it, a scandal that broke out when Intel tried to downplay the importance of small, occasional errors that customers had discovered in the Pentium chip’s floating point divide operation. “Rather than having full 53-bit accuracy, you might have lost up to 12 bits off of that,” Crawford says. “We’re talking about parts per trillion of inaccuracy.”

Publicity about the problem was so intense Intel was forced to offer replacement chips to any customer who wanted one, at a cost of half a billion dollars. “Once we found out the problem it was fairly straightforward to fix,” Crawford says.

According to Crawford, Intel is the only company that publicly posts all flaws, or errata, it discovers in its chips, no matter how insignificant. “It was a learning experience for us — a fundamental change came out in terms of how we deal with errata in our chips,” he says.