A look under the hood of Pentium 4

Intel's next-generation chip is humming with an all-new architecture. But what everybody really wants to know is how fast will it go?

Intel opened the hood of its next-generation microprocessor this week, revealing details about the Pentium 4's new microarchitecture to developers.

The company said its NetBurst microarchitecture, which makes its debut in the Pentium 4, would carry the company four to five years into the future. Over time, NetBurst will be used in other Intel chips, including mobile processors and processors for workstations and servers, the company said. But analysts attending the Intel Developer Forum questioned whether NetBurst's hyper-pipelined microarchitecture would cost the Pentium 4 in speed.

A pipeline is a structure, used in processors, that breaks a task into a number of small pieces in order to process it more efficiently. The pipeline is made up of a number of stages, each of which does a small amount of work. The length of a chip's pipeline is what governs its clock speed performance.

Pentium 4 has a 20-stage pipeline that will allow it to achieve high clock speeds. However, the higher clock speed will not translate directly into the same kinds of performance increases that would occur by increasing a Pentium III's clock speed, analysts said.

The Pentium 4's longer pipeline comes into play here. It takes time to fill registers, which store information, in the pipeline. This translates into lost performance per clock cycle.

Longer pipelines are also more prone to errors, said Steven Leibson, editor-in-chief of the Microprocessor Report. The pipeline may need to be refilled, taking up clock cycles where the CPU sits idle waiting for information to process.

The "long-pipeline tax" is what Leibson calls it. "What that means is, you get big bubbles if you don't predict branches correctly," he said.

Intel's Pentium 4 architecture includes features designed to predict pipeline branches more accurately in order to increase performance, Leibson said. But the features have not been tested by third parties, such as the Microprocessor Report.

Intel says Advanced Dynamic Execution (ADE) combats long-pipeline performance issues in NetBurst. Two elements of ADE work to keep the pipeline stages filled with a continuous flow of instructions and to improve branch prediction capabilities, the company has said. Together, they keep the pipeline fully tasked to avoid Leibson's pipeline bubbles.

Other NetBurst features include Intel's Rapid Execution Architecture, which aims to raise performance by using two Arithmetic Logic Units to process certain instructions, such as additions or subtractions, every half-clock cycle. This, Intel said, leads to increased throughput, as certain instructions are executed at twice the rate of processors' core frequency.

NetBurst beefs up the chip's processor caches, as well. The Level 1 cache, for instructions, is now called Execution Trace cache. The cache stores decoded instructions, eliminating the need to wait for them to be decoded first, a performance improvement. Intel also increased the throughput of its Level 2 Advanced Transfer cache between it and the CPU by a factor of three to 48GB per second. The cache will scale with processor clock speed increases.

A longer pipeline makes for higher clock speed. But does it sacrifice performance? The answer is yes and no.

Asked if a Pentium III would outperform a Pentium 4 at the same clock speed, Intel's Albert Yu said, "It's technically correct, but it's artificial."

Intel will not release Pentium III and Pentium 4 chips on the same manufacturing process at the same clock speed at the same time, Yu said. Pentium 4 chips will be several hundred megahertz higher than Pentium III chips, even with both of the chips moving to 0.13-micron technology, which will give the ability to further raise the clock speed of the Pentium III.

Forthcoming Pentium III chips based on 0.13-micron process -- which are expected to reach speeds as high as 1.4GHz -- would outperform the Pentium 4 chips due out this fall on 0.18-micron technology. But Yu said that was an apples-to-oranges comparison.

When both chips are being manufactured on 0.13 micron, the Pentium III will range in speed from 1.2GHz to about 1.4GHz, however, Pentium 4 will be at 2GHz and higher, according to Intel roadmap information posted on the Web recently.

See Chips Central for daily hardware news, including an interactive timeline of AMD and Intel's upcoming product launches.

What do you think? Tell the Mailroom. And read what others have said.