AMD K6-2, AMD K6-2+, AMD K6-III, and AMD K6-3+

This family of processors is what beat the Pentium II at its own game. AMD sought to extend the life of the Socket 7 platform with the Super7 initiative which specified external bus frequencies up to 100 MHz. Intel never bothered with more than 66 MHz, even with the Celeron, so the AMD K6-2 processor joined the value+performance race with an exceedingly viable alternative.

I didn't get into the Super7 scene until shortly after I busted the CPU socket on a reshopped MediaGX computer I bought from the returns rack at Micro Center. There was more to learn about this stuff so I decided to build a computer from parts obtained at the cheapest prices at various online and storefront vendors. For my first project I chose the 350 MHz AMD K6-2 processor because I was trying to learn more about why Intel said the Socket 7 platform was obsolete past 233 MHz, but AMD was making chips that ran in excess of 300 MHz on the same platform. At the same time the AMD K6-2 was beating the then-new Pentium II and the CeLeron at the same clock speeds.

The AMD K6-2 was almost entirely based on the NexGen processor architecture that AMD acquired about a year before the introduction of the K6. These chips did not resemble the earlier K5 chips at all, and supposedly the NexGen technology was what became the K6 and K6-2 processors. The curious use of a microcode translator to convert x86 instructions to something called RISC86 was interesting, too, becuase AMD supposedly had to license that part of the chip's workings from Intel. Cyrix at the time was using a reverse-engineered x86 instructions without RISC conversion and I wondered what it all meant.

The K6-2 has 64KB of L1 cache on the chip. You might think this isn't a big deal but at that time the Pentium II at the time had just 16K of L1 cache on its chip. Both chips rely on L2 cache to get more speed out of their RISC86 pipeline (as small as it was then). The K6-2 used the cache on the motherboard and the Pentium II used cache mounted on the processor slot cartridge. What set the K6-2 apart was that it could communicate with up to 2MB of motherboard L2 cache at 100 MHz while the Pentium II used 512KB of off-chip cache at a pathetic 66 MHz. This was a big reason that the K6-2 beat the Pentium II at the same clock speed and for much less money. At this time the CeLeron came out with half the cache (128KB) but all of that cache was full-speed on-die cache and performed closer to the Pentium II for much less money but still much more than the K6-2. The Pentium III answered this question finally by putting 256KB of the L2 cache on the chip itself like the CeLeron already had. Still the K6-2 gave more bang for the buck by compensating with 1MB and 2MB motherboard caches and curious new chips called the K6-III, K6-3+, and K6-2+ that started coming out as the Super7's days were winding down.

And what about those K6-III, K6-3+, and K6-2+ chips? They don't qualify for bottomfeeders because their cost never came down low enough for me to care. The K6-III with 256KB of full-speed, on-chip L2 cache consumes much more power and generates excessive heat, requiring a subset of more expensive motherboards and cooling solutions. The fascinating low-voltage follow-on AMD K6-2+ (128KB L2 cache) and K6-3+ (256KB L2 cache) chips were never made available to the general public as discreet components, though aftermarket shops like carry them at fire-sale prices now. A few Sony and Compaq laptop models were produced with K6-2+ but I hadn't seen any K6-3+ models. I do have a 450 MHz K6-2+ chip but it only works on two of my Super7 motherboards, so it unfortunately can't qualify as a bottomfeeder. It is very fast, though.