K6-III+ and K6-II+

Posted on July 21st, 2011 in Prehistory

The processor industry was like this: Intel made processors and other companies either made licensed copies or clones of the Intel chips. However, after the 486 era, the “others” made their own processors.

The ancestor of the K6-II+ and K6-III+ was the K5. AMD worked on it for a while, going for their own design. They released it very late, as both Intel and Cyrix had processors running at 200 MHz when AMD released the 100 MHz version. Subsequent K5 versions were always late, however cheap and thus they had limited success. They only made sense as entry-level chips because they were so late.

AMD made the K6 processor after acquiring or merging with NexGen, who were a separate and very young processor maker who had good and fresh new ideas. They were essentially the driving force behind the K6, a chip that was very good. Because it was very good, processor supplies were frequently finished and even the prices started to rise. AMD was known as the ultimate bargain, decent speed for a low price (even more so because of the K5). But because the prices rose as AMD’s factories couldn’t keep up with demand, the K6 became more like money well spent. Soon after that, other companies released other chips competing in both price and performance. Back then there was Intel, Cyrix, AMD (a while before there was NexGen, too) and even IDT, a company making other kinds of parts, who were trying their hand at x86 chips.

Apart from the competition and the ever-changing market, AMD was also the target of re-marking. Their processors at the time had a metal cover. The labels meant to inform any human eyes that the processor is that model and runs at that speed, were made with some sort of ink. They could be erased and modified to manipulate buyers into thinking the processor is better than it actually was. For example, Red Hill Technology, long time sellers of computer systems who appear to have accumulated vast knowledge about this “prehistory” of computing as I like to call it, explain here that the usually remarked processors were 486SX-33 but in reality it was a 486SX-25 (the numbers after the dash represent the speed in MHz), 486DX/2-66 actually being a DX/2-50, 486DX/4-100 supposedly made by Intel, in reality an Intel 486DX/4-75 or an AMD DX/4-100 and Pentiums 166 that were actually Pentium 133 or 150. Red Hill say the practice is unethical and illegal, and really, who could argue with that.
Also, Red Hill advises people to buy boxed, retail processors (these always cost more) because it is much harder to remark them. They also list a few reasons why remarking isn’t done much anymore. But their article is quite old, as nowadays the practice of remarking is, I believe, 100% gone. The markings on processors are today very hard to erase or fake (and they have been like that for a lot of time) and each processor has a serial number or part number that is unique among the other part numbers generated by that specific manufacturer.

Despite remarking and competition and an ever-changing market, the K6 sold well. AMD appears to have always had them in insufficient supply, except for the last two K6 models, one of which was the 266 MHz version. This is because another AMD plant developed a new, smaller chip technology – 0.25 microns, or in more scientific terms, micrometers or μm. A little off-topic: pasted from Wikipedia, one micrometer measures 1.000×10⁻⁶ meters, or 0.000001 meters. And, as Fail Blog quotes it here as a win, “ATTN. imperial unit barbarians,” a micrometer is 3.281×10⁻⁶ feet, or 0.000003281 feet. Or, 39.37×10⁻⁶ inches or 0,00003937 inches. I’d advise said “barbarians” to not confuse micrometer, a word that should actually be spelled micrometre, with the instrument called micrometer.

Anyway, AMD then released the K6-II. As it was apparently also made in 0.25 microns, supplies were as plentiful as the previous 0.25 micron K6 models. What was the difference? The 3DNow! instructions. These were specifically made to handle floating point operations, heavily used in 3D graphics. Like with all newly released processor features, if the software doesn’t make use of it, it goes unnoticed. Or, it sometimes even works slower than inferior versions, on the whole. This was the case with the older Pentium MMX – the MMX instructions, as far as I remember standing for “MultiMedia eXtensions,” when some non-MMX Pentiums could outperform the MMX ones in software that wasn’t making use of the new instruction set. Speaking of MMX, this was another feature the K6-II had: a much faster MMX unit. By now, software was making good use of that.

Another very important enhancement was the introduction of 100 MHz FSB, with the K6-II. AMD did make models with 66 MHz FSB, but they were. of course, more interested in selling the 100 MHz ones. We are told by Red Hill, however, that while the K6-300, apparently the absolute last K6 model released, officially only supported 66 MHz FSB, two out of three ran just fine at 100 MHz and 3x multiplier. The K6-II now had the markings engraved in the metal cover, making it unbelievably hard, if not impossible, to remark the processors. AMD released 266, 300, 333 (this one was weird because it needed a 95 MHz bus), 350, 366, 380, 400, 450 and even 475 MHz versions (and a further few others). The weirdly clocked ones weren’t even available for sale in retail stores, so they were most likely made for the companies who assembled computers in large numbers, or in other words the OEM market. The models with weird clock speeds were usually running on a 66 MHz bus and the user would still get to see the magic psychological number as his processor speed: over 300, close to 400 or over 400, until the new processor models, with real speeds around 400 or over 400 would be released and the user would be invited to pay for the all important new model, again. Of course, when the new was released, advertising would go a little wild (depending on marketing budgets), reviewers would say how great this new model is, while others would test and compare the newly released models from competing manufacturers, to each other.

About this time, AMD released the K6-III. These were among the last members of the K6 family, along with some final K6-II models. AMD made K6-II 500 parts and even a sad 550 MHz one that wasn’t working right, even when set strictly to factory specifications. The same Red Hill report that they were stuck with selling trays of these and they managed to get rid of them by downclocking them to 500 MHz and selling them at a loss. For some reason (maybe AMD factory overclocking? This was unfortunately a known tactic from CPU manufacturers, during those times) the parts ran absolutely stable at 500.

Anyway, the new K6-III had a slightly more subtle improvement: processor cache. All the processors since a long time ago have had L1 (which stands for level 1) cache. L2 cache was usually little or absent, or sometimes on the motherboard. I remember some motherboards with slots for L2 cache, called SRAM (nowadays only some high end, specialized server systems have upgradeable SRAM) and the most I have heard of was 2 megabytes. To put things in a little historical perspective, the really ancient 286 processor that you might have heard of, had absolutely no cache integrated in the processor. Instead, the cache memory was on the motherboard and, on some 286 motherboards, the RAM chips looked very similar to the cache chips. I have to say, cache memory is different and much more expensive than RAM, because it is “associative” memory. For each memory block, you have to have an allocated address. There’s a separate chip doing the allocations, as far as I know, so that’s why half the cache size is usually memory for data and the other half, memory for instructions.
Well, the K6-III added a lot of cache. This wasn’t at all a bad idea. Any K6-II seems to have had 64 KB of L1 cache, 32+32. The K6-III had the same amount of L1 cache. But L2 cache, meaning “level 2,” which is some sort of “second-stage” memory storing things so that the processor can easily access them when it’s ready to do so, was not to be found on the K6-II at all. The K6-II simply used the motherboard’s cache memory as L2 cache. Some motherboards had 512 KB, others had 1 MB. The K6-III had 256 KB, but apart from this L2 cache (the more of it, the better, for any processor) the K6-III could also use the motherboard’s cache memory as L3 or tertiary cache, the same that the K6-II would use as L2.

People bought K6-III processors and they used them like any other desktop processor. But what AMD did not do nicely was that they released the K6-III+ and barely told anyone about it. These became famous later, when people uncovered them from underneath the mist of “nothing to see here, move along” pulled on by marketing departments.
Okay, these were meant to be mobile processors. So AMD had no apparent reason to market these to end users. End users must be lobotomized halfwits who absolutely must have no idea what goes on inside their laptops. They must only be fooled into paying for whatever crap companies want to feed them. Thus, why all the informing users that we released a kick-ass processor? They must be buying the desktop versions and shut up. The only people who could find out about them (keep in mind that all the processor industry was making a lot of noise about their products, of course confusing users) were the importers, the retailers and so on. The computer shop on the corner was probably able to get these and sell them, but first they had to know about these models.

The K6-III+ was released in three famous versions. The easiest to find is the 450 MHz. There was a 500 MHz version and a very rare 550 MHz one. I remember Tom’s Hardware was sent one 550 MHz for review, they were very pleased with it, but they had to return it since AMD recalled them before even releasing them to market. AMD did this, as far as Tom’s Hardware assumed at the time, so that they can focus on the upcoming Athlons.
Probably because this was a mobile processor, it was made easy for the laptop makers to modify the speed of it. Another revealing aspect of how much greed there is in the industry: the laptop makers could modify the laptop’s speed easily, add or remove features and modify the price at will. And the processor costs the same. The user, of course, has to be an idiot and just buy whatever they want to give him for the amount of money he has, without any questions. And if you’re tech savvy and you actually know what’s inside these computers, you either have to conform yourself to the majority, find a rare and hyper-expensive laptop model that doesn’t insult your intellect, or skip getting a laptop altogether. Shoo, this is for unconscious, sleep-walking people only.
This easy modification is unbelievable: the processor can have the multiplier speed modified on-the-fly, in other words, while the computer is on and after it has already loaded an operating system. There are utilities to change the multiplier on these “+” K6-III and II models, for DOS, Linux and Windows. But the best thing you could do was to set the multiplier to 2x and the FSB to the maximum that the processor / motherboard / cooler combination could take. The multiplier set to 2x was actually interpreted as 6x by these processors (I think the regular K6-II did that as well, but it was really unstable and sometimes the computer didn’t even boot up) and they ran stable when set to that “2x as 6x” multiplier and the max FSB you could throw at them. I have a 600 MHz K6-II+ computer, of course set to 6x 100 MHz and a 500 MHz K6-III+ computer, set to 6x 83 MHz because the motherboard doesn’t support 100 MHz FSB. These 83 and 75 MHz FSB speeds were originally made available for compatibility with the Cyrix processors.

The K6-II+ was actually a K6-III+, but with just 128 KB of L2 cache instead of the original 256 KB. Again, branding and marketing coming out as more important than reality and, ultimately, than the truth.

The only actually notable difference in specifications was that these ran at a lower voltage. The normal voltage for some of their desktop counterparts was the absolute maximum voltage for these mobile processors. But quite a few motherboards of that time were compatible – the “+” models usually went great at 2.0 volts instead of the usual 2.2 needed by the desktop versions. I think 2.2 volts was the maximum they could take, virtually making them compatible with anything.

So, there you have it: K5-too late, but cheap, K6-good performance, little supply, K6-II – good performance, K6-III – even better performance. And then you get these wonderful “+” parts, but nobody ever mentions them. These could be heavily overclocked (there were reports of running them at 800 MHz) and ran cool, as they were meant for the mobile market. But why tell the users about them? No, users have to be ignorant and uninformed. How are we going to manipulate them if we inform them of everything they should know?!

One last piece of info: you can test any of these processors if you have a motherboard supporting the “+” versions. The K5 was made for the Socket 5, but any Socket 5 processor can be fitted on a Socket 7 motherboard. The K6, K6-II and K6-III were made for Socket 7. Intel’s last processor for the Socket 7 was the Pentium MMX, about the time when AMD was making the K5 and a little into the K6 timeframe. AMD soldiered on with the same socket and at some point, revised the specifications to make the Super Socket 7, with support for 100 MHz FSB and support for AGP being two notable additions. Any Socket 7 processor worked in the Super 7. Socket 5 processors, however, did not always work. The motherboards didn’t always supply the needed voltages. On the other side, Super 7 processors didn’t run at full speed in plain Socket 7 motherboards.

There you have it. History should remember the K6-III+.