A New Hope
May 1998, Atlanta, Georgia, the E3 Expo was the introduction arena for AMD's "sure fire win" against Intel, the K6-2, formerly known as the K6-3D. The elegant fix to AMD's FPU problem was not throwing more raw performance at the applications that made the original K6 choke rather giving it the ability to more efficiently use the power it did have. The technology? 3DNow!
The 21 instructions that made up 3DNow! were not revolutionary, but they were a step in the right direction. For once AMD was the market leader in a step that the rest of the industry would follow through with, the step was towards a technology known as SIMD. Intel actually made the first move towards Single Instruction Multiple Data execution with their 57 MMX instructions that did so very little with the Pentium MMX and Pentium II. The whole idea behind SIMD execution is that if you're performing one operation on multiple types of data, why not perform that one operation on all the data simultaneously instead of waiting for the operation to be completed on one set of data before repeating the process. Therefore we get the basic definition of SIMD, or Single Instruction Multiple Data (in this case SIMD-FP as it applies to FPU instructions, whereas MMX offered SIMD-Int for Integer instructions) a hardware algorithm that allows a single command (or instruction) to be applied to multiple sets of data simultaneously.
The problem with Intel's MMX instructions was that the only type of SIMD being used was SIMD-Int, for integer based calculations, an area that honestly wouldn't benefit considerably from SIMD execution capabilities. Using Intel's branding, the term MMX became a household name while the majority of hardware enthusiasts were complaining about the uselessness of the technology. What AMD did was apply the SIMD execution algorithm to Floating-Point instructions, thus resulting in SIMD-FP, and also resulting in a significant increase in FP performance. As it seemed, AMD had finally won and Intel's "tyrannical" reign was over, the K6-2, with 3DNow! support was able to beat the Intel Pentium II in games that took advantage of the 21 new instructions. Then came the Celeron…
Intel's low cost competitor to AMD's platform, the Celeron, was what forced AMD into yet another price war. AMD would drop prices, then Intel would drop them even lower on the Celeron line. Intel had the ability to drop the prices of a line that was relatively inexpensive to manufacturer, and could compete on a clock for clock basis with AMD's latest and greatest. AMD was struggling once again, this time motherboard manufacturers didn't seem to support AMD's Super7 platform as well as the K6-2 demanded. Graphics card incompatibility problems and poorly manufactured motherboards plagued the platform.
In November of 1998, AnandTech was able to take a look at AMD's next attempt at winning back the performance market, the K6-3. Later renamed to the K6-III (it's amazing what three I's can do), the processor took the Celeron route and integrated a full 256KB of L2 cache onto the die of the processor therefore operating at the processor's clock speed and giving the fastest Pentium IIs a run for their money. By this time, 3DNow! was being implemented in more and more games, however the K6-III still lagged behind in terms of non-optimized FP code not to mention the continuing motherboard issues and chipset incompatibilities. Not more than three days after we previewed the K6-III, the first person to ask us the million-dollar question appeared in our mailboxes, "When will you be previewing the K7?"