The Pentium III(*) was released by Intel in 1999 to directly compete with AMD's K6-2 and K6-3 processors. As the successor to the Pentium II, it was meant to dominate the high end. By this time Intel had actual competition from AMD's K6-2 chips, though this light competition was nothing compared to the beating the Pentium III recieved at the hands of the Athlon Thunderbird in late 2000. But I'm getting ahead of myself, and chronological order is the best kind of order(**), so. The story of the Pentium III begins with the...
The first Pentium IIIs, featuring the 0.25-micron Katmai(***) core, were released on February 28th, 1999. The Katmais had 512KB of L2 cache; however, the cache was not actually on the CPU die, meaning that it ran only at half speed. The Katmais were available in Slot 1 only, as Intel was still in their slot-loving phase at the time of the Pentium III's release. If this sounds an awful lot like the Deschutes Pentium II to you, you're right; in fact, the Katmai's only real advantage over the Pentium II was the addition of the SSE (also called MMX2 in the Goode Olde Dayse) instruction set. Of course, that's not counting a certain controversial "feature" of the Katmai: the PSN.
What is the PSN, and why did I lead up to it as if it were evil? Well, Intel decided to burn a unique serial number (called, creatively, the "processor serial number", or PSN) onto each Katmai. The number could be read over the internet, and it was intended to be used to help verify the user's identity. That's right-- your identity was burned onto a little chip in your computer that, in theory, anyone could access. It could be turned off with software, but it was quite possible for someone to remotely turn it on again without you knowing. The execs at Intel, apparently, saw nothing wrong with this. Amazing, no? Intel caved, because the bad PR was doing a number on their stock, and let users deactivate the "feature" in BIOS.
The Katmai went all the way up to 600MHz, and Intel pulled out all the stops to stop the AMD K6 and Athlon (the original Pluto core K7, the first "Athlon Classic," was also released in 1999) from succeeding. (Unfortunately, with the Katmai, "all the stops" meant "raising the clock speed and praying.") The 600MHz Katmai was in both 0.25 and 0.18 micron versions, the 0.18 having a smaller die size, a lower voltage, and a higher FSB. The AMD Athlon would kill the Katmai at the same clock speed, and Intel, seeing that they had a dud of a CPU on their hands, dropped the Katmai late in 1999, and began producing the...
The Coppermine was the real deal, a Pentium III worthy of the extra I. A .18-micron chip with 256KB of on-die L2 cache that was SECC2 (Single Edge Contact Cartridge 2) instead of SECC (Single Edge Contact Cartridge, also known as just SEC), the Coppermine's performance was far above that of the Katmai at the same clock speeds, and the Coppermine was able to scale higher. Even though the Coppermine only had 256KB of L2 cache, the cache was on the die itself and ran at full speed, which ended up being faster than having twice as much cache off-die. While initially only available in Slot 1, Intel soon started producing Socket 370 Coppermines, as socket CPUs are much cheaper to make than slot ones. The next-generation Celeron (sometimes called the Celeron II) used the Coppermine 128 core, which was just the Coppermine with less L2 cache. The new Celeron wasn't out for long before the Duron was released and beat the pins off the thing. Didn't stop people from buying it, though.
Obviously, there were improvements to the Pentium III other than an 0.18-micron process, speedier cache, and raised FSB (well, sometimes raised). There aren't many, but the main one is Advanced System Buffering. There are many buffers on the processor that buffer information going between the processor core and the system bus. The number of all the types of buffers ("fill," "writeback," and "bus queue entries") was increased on the Coppermine. This wouldn't have been very necessary on the Katmai, but remember that many of the Coppermines have a faster bus speed, and more of these buffers ensure it will run at full speed. And, of course, the 0.18 micron die means that it produced a lot less heat. The old Katmais had a core voltage of around 2.05v, while the Coppermines, with their smaller transistors, were able to move down to 1.65v, although a couple of the speedier ones had to have 1.7v.
I may have misled you into thinking the Coppermine was a big success. It wasn't. You see, the processor isn't everything, and at the time of the Coppermine's release Intel was having tremendous platform troubles. Either users could stick with their sturdy old BX boards, which were unable to handle frontside bus speeds above 100MHz, or they could switch to the RDRAM-using Intel 820/Intel 840-based motherboards. Now, RDRAM has a bad reputation these days as being overpriced and slower than dual channel DDR-SDRAM (a reputation it deserves, if you're wondering) but you have to understand that this is nothing compared to the astronomical price users had to pay for RDRAM back at the start of 2000: nearly a $1000 US for just 128MB of the stuff. Obviously nobody would want to pay more for just 128MB of RAM than they did for the processor and motherboard combined. VIA's Apollo Pro 133A chipset turned out to be the best choice for Pentium 3 owners who wanted a 133MHz FSB, but the good old BX motherboards generally did better than the 133As even with a 133MHz FSB CPU in the 133A and a 100MHZ FSB CPU in the BX. And you thought the Athlon had growing pains!
Naturally, with all these platform troubles, there was a whole mess of Pentium IIIs available. A Pentium III with the Coppermine core would have an "E" after its name, and a Pentium III with a 133MHz frontside bus would have a "B" after its name. There was a whole lot of Coppermines due to the differing FSBs; after all, you can't have a 667MHz processor with a bus speed of 100MHz, because the multiplier can only be a multiple of one half. In other words, the 500, 500E, 533EB, 550, 550E, 600, 600E, 600EB, 650E, 667EB, 700E, 733EB, 750E, 800E, 800EB, 850E, 866EB, 933EB, 1000E, 1000EB and more were all available, most available in both Slot 1 and Socket 370, and with varying voltages. Confusing, no? But the Pentium III's history doesn't end here, for there is one more core, the...
Originally Intel planned to announce the Tualatin long before it actually did, but it decided to concentrate on the still-0.18-micron Pentium 4 instead. The Tualatin had next to no fanfare when it came out, which is maybe a little unfair, as it was easily the best Pentium III processor yet and absolutely KILLED P4s at the same speed. Intel could have fit a bigger 512KB L2 cache onto the chip (and they did so, with their mobile and server versions), since it was 0.13-micron, but they didn't, for fear of making it a competitor with their flagship desktop CPU, the Pentium 4.
The Tualatin also has a neat little thing called Data Prefetch Logic. Basically it looks for patterns and guesses at what's going to be used next, putting that data into the L2 cache. If it's wrong, there's no performance penalty (it's simply overwritten), and if it's right, the time needed to fetch that data from the computer's main memory is reduced to a perfect zero. This does wonders for processor efficiency. They also have an Integrated Heat Spreader (IHS), which helps a lot with processor cooling and also helps to prevent the processor core from being damaged by improperly set heatsinks. The Pentium 4 also has this handy device, and future processors will probably have it as well.
The Pentium III saw many incarnations of itself in its time, and it is perhaps unfair that the Tualatin, the fastest and newest Pentium III with the best platform support, is the least common today. The Tualatin could take on Athlons at the same clock speed and do fine; the Pentium III-S, that's the server version with 512KB, even did better than the Athlon at the same clock speed. But not only did Intel not care about it, the Athlon XP popped out around the time the Tualatin hit 1.4GHz. It may be some comfort to the poor Tualatin that it seems Intel will regret their decision to go for low IPC and high clock speeds; in the distant future, they'll be moving a pumped-up version of today's Pentium-M onto the desktop. And does anyone know what the Pentium-M is? Why, it's a P6-based CPU-- an improved Tualatin. As we can see, good will always win in the end.
* Some people insist that you call it the Pentium !!!. These people are stupid and you should ignore them.
** This is a lie. Still, there's a certain ring to Katmai, so it goes first
*** Katmai is a wonderful name for a CPU, as it has nothing to do with explosions or massive amounts of heat.