The Pentium III: In Brief

The "next-generation" in Intel's lineup of performance PC processors, the P3 is actually based on the same Pentium II core, with a few additions that, if nothing but from a marketing point, should make it a "must-have" upgrade. There have been rumors before release that the Katmai will ship with a 64KB L1 cache, but at least for the release of the 450Mhz and 500Mhz versions, that isn't true. The Pentium III will initially ship with 32Kb cache, the same as the Pentium II.

How will this affect its performance? Consider the fact that the PentiumMMX 200 was roughly 10% faster than the Pentium Classic 200, due to the doubled cache size (32KB on the MMX unit vs. 16KB on the Classic). Though a 10% performance gain per doubling in cache isn't necessarily the case, the Pentium III could very well have taken advantage of the additional 32K, if nothing for prefetch memory streaming in SSE. More on that later.

Intel has long seen the need to cater to gaming and entertainment titles, (a process that Apple Computer is only now accepting in their new corporate strategy). The reason is simple - entertainment not only drives technology, but also brings in new users and creates a market for upgrades and future processes. This drive spurred the addition of "MMX" to the Pentium and Pentium II lines, the Multimedia Extensions (the naming of which Intel vehemently denies) which were supposed to substantially accelerate multimedia operations.

The premise behind MMX was that certain applications, such as video, 3d, and multimedia, can be accelerated greatly if specific arrays of data common to these applications are executed quickly, and in the case of MMX and KNI, in parallel. This kind of processing is known as "Single Instruction Multiple Data," or SIMD.

In reality, MMX had little "real-world" effect on applications, and became known more for its marketing than for any real performance gain. AMD however, has seen the potential of MMX, and had it implemented for floating point calculations as the "3DNow!" instruction set in its K6-2 CPU to optimize 3D geometry and physics operations. 3DNow! has been partially responsible for negating the lackluster performance of AMD's non-pipelined FPU, at least for Quake II, where the heavily 3DNow! optimized port boosts performance to slightly exceed that of an equivalently clocked Pentium II.

Katmai continues to support MMX, but what sets it apart from the P2 are 71 new SIMD instructions designed to greatly enhance the performance of floating point operations, just like 3DNow! Tentatively called "Streaming SIMD Extensions," or "SSE" for short, these new instructions are designed around the same idea as 3DNow, that specific arrays of floating point numbers (used widely for 3D calculations) can be executed in parallel, saving time in execution. But almost all of us know about MMX, which failed to live up to the hype. What makes the Pentium III different?