Summary

Optimizations of the new Athlon4 or Palomino include the addition of a thermal diode as protection against burn-out as well as implementation of the outstanding 51 SSE instructions missing from the original Athlon and Thunderbird core. We have taken both an Athlon4 "Palomino" and an Athlon "Thunderbird and run them through a series of more or less SSE optimized application benchmarks. The spectrum of the benchmarks encompassed gaming applications, voice recognition and content creation applications as well as professional CAD programs. On a clock by clock comparison, the Athlon4 "Palomino" showed up to 89.6 % performance increase over the standard Athlon "Thunderbird" with the smallest delta in gaming applications and the highest gains in CAD applications. The successful implementation of the SSE instruction set in the Athlon4 makes the latest AMD CPU a superior candidate for all professional design and CAD applications.

Athlon4 "Palomino" (left) and Thunderbird (right) top view, note the different shape of the core in the Athlon4 with an increased die size of now about 37.5 Million transistors compared to 37 Million transistors in the classic Thunderbird. Photographs courtesy of AMD.

AMD's Athlon, particularly the Socket A version, nicked Thunderbird has established itself as one of the hottest CPUs on the market, depending on who you talk to, of course. Heat, on the other hand is not necessarily the most desirable feature, particularly if there is no protection from overheating in form of a thermal diode. To correct this problem and provide thermal protection in form of automatic shut-down, AMD has finally integrated the thermal diode into the core of their latest spawn, the Athlon 4, based on the Palomino core.

Other improvements of the die are the addition of the outstanding 52 instructions for a fully fledged SSE or SIMD set of 71 instructions. The other 19 instructions were already part of the original Athlon design. As a quick rundown on SSE or SIMD, the latter abbreviation stands for Single Instruction, Multiple Data and essentially means that one instruction can be repeatedly used to process an entire string of data without having to reload the instruction from the instruction cache.

Since the introduction of SSE with Intel's PentiumIII quite a few software vendors have optimized their programs for SSE and, while there has always been the argument that these companies would favor Intel, there can be no doubt that any optimization is better than no optimization. Consequently, SSE should be considered as something good and not merely an Intel marketing tool. Especially in computationally very intensive applications SSE, can make a huge difference, that is, some CAD applications see as much as 80-90% performance increase just by using SSE. Coincidentally, the mid and high end CAD and graphics workstation market is one domain where AMD has had more difficulties finding a foothold.

There have been many speculations about which applications are SSE optimized and which ones aren't, amongst those mentioned are Quake3 Arena and Adobe Photoshop. This is, however, not where the list ends or should end because, as we will show in the following, there is a lot to be gained from this kind of optimization.