How CPU multipliers came to be locked

It was 1996. I was a senior in college, and I went to the computer store in the student commons to get a cable or something. I ran into an old classmate working in the store, who went on to work as an engineer for Boeing. We talked for a few minutes, and he told me about a web site that I just had to visit. I still remember the URL for some reason. He grabbed a piece of paper and scrawled “http://sysdoc.pair.com” on it.

It was my introduction to the world of PC hardware enthusiast sites. That mysterious URL was the early address of Tom’s Hardware Guide. The front page mostly consisted of links to articles telling you how to overclock Pentium CPUs using undocumented jumper settings on Asus motherboards, and the ads were largely mail-order houses offering specials on Asus motherboards and low-end Pentium CPUs.

Dr. Thomas Pabst was a German M.D. living in England. He enjoyed playing video games on computers, and this web site was his hobby. He had articles explaining bus speeds and multipliers, and personal experience with overclocking–running CPUs at higher rates than the speed stamped on the chip.

I remember reading an article where he discussed his frustration with one of his systems. He had a mid-range Pentium CPU–I don’t remember anymore if it was a P120 or a P133–and he wanted to overclock it. He couldn’t get the chip to run at 200 MHz. It did run at 166 MHz, and at 180 MHz. And he talked about the dilemma he faced, deciding whether to run at the higher clock rate (180 MHz, 60 MHz x 3) or the higher bus speed (166 MHz, 66 MHz x 2.5). He grudgingly went with 180 MHz, but it meant he was sacrificing 10% of his memory and PCI bus speed to get 10% more CPU speed.

But the cutting edge stuff on his site regarded a couple of jumper settings on the Asus P55T2P4 motherboard. Using this particular board, you could get not only the Intel-sanctioned 50, 60, and 66 MHz bus speeds, but you could also get a 75 MHz bus speed (used by certain Cyrix CPUs) and even an 83 MHz bus speed.

Dr. Pabst giddily showed how a Pentium running on a 75 MHz bus with a 2.0 multiplier outperformed a Pentium 166 running at the supported settings of 66 MHz x 2.5. And a Pentium running at 83 MHz x 2 outperformed a Pentium 200. The holy grail was a chip capable of running at 83 MHz x 2.5, yielding a clock rate of 208 MHz.

Of course he didn’t advocate just buying a 200 MHz Pentium and running it at 208 MHz. That was too easy. And overly expensive, in his view.

A side effect of these nonstandard bus speeds was that the PCI bus also ended up running at nonstandard speed. So he also discussed what video cards could handle those conditions.

And in yet another article, he explained the reason why all of this was possible.

Intel initially released its Socket 5 P54C Pentiums in late 1994, running at 90 and 100 MHz. But early on, they found a significant number of their parts couldn’t run reliably at 90 or 100 MHz. So they did some more testing on the defective chips, found that a significant percentage of them ran reliably at 75 MHz, and decided to create a 75 MHz product so they could sell those chips. They priced them at a significant discount, since otherwise the chips would be discarded.

As manufacturing processes mature, more and more of the chips are able to run at full speed. So eventually, Intel didn’t have enough defective chips to meet market demand for the 75 MHz part. So some unknown number of faster chips just got marked for 75 MHz and sold that way.

And that was what Tom’s Hardware was trying to take advantage of. Buy the low-end part and see how fast it would run without crashing. And since these were still the days of Windows 95 and booting into DOS to run some software, it was sometimes possible to get things running pretty far over their rated speed, since those older operating systems didn’t place as much demand on the CPU as Windows NT does.

A few years later, a back-looking article by Frank Völkel claimed that all of that early coverage of Pentium CPUs and Asus motherboards changed the whole industry. That sounds pretty self-congratulatory, but I think it was true.

Within a year or two, some other enthusiasts launched their own competing hardware sites. It gave them a chance to share their own findings, and in at least one case, the goal was to post new content more frequently than Tom’s Hardware did.

Overclocking wasn’t a new phenomenon at all. I first read about it in the late 1980s, and people had been doing it for several years even before then. That particular Asus motherboard was in the right place at the right time. It made overclocking easier than it had been, and the Internet was in position to tell large numbers of people about it. It also happened to be a good board for people like me. I used that board in a lot of systems, on the theory that if it could handle the stresses of overclocking, it would run really well when you ran it the way the chips were designed to run. And it did.

Regardless of how many people actually followed the instructions, there was enough interest in it that these web sites started making money. Some motherboard manufacturers’ attitudes changed too. Other motherboard manufacturers started making boards that tried to be as overclocker friendly as the Asus P55T2P4 was, and they printed the settings for nonstandard bus speeds like 83 MHz and nonstandard voltages right there in the manual. You didn’t have to go to an enthusiast web site to find them.

Intel couldn’t do much to stop it without breaking backward compatibility, at least not in the short term. But once they released a new CPU architecture, they didn’t have to worry about that anymore. So in 1998, they started locking the multipliers on Pentium II CPUs. And that really put a damper on overclocking. Now you couldn’t buy the entry-level 233 MHz CPU and change the multiplier to run it at 266 or 300 MHz. All that was left was changing the bus speed to get modest improvements. And with the new architecture, Intel didn’t have to worry about AMD and Cyrix introducing odd bus speeds and making them mainstream either.

That was the main reason Tom’s Hardware was about the only place that got excited about the original Intel Celeron. The first 266 and 300 MHz Celerons didn’t have any L2 cache, so they were terribly slow. But the L2 cache was a limiting factor in overclocking early P2s, so Tom’s Hardware tried running those Celerons on a 100 MHz bus. They found the chips ran at a 33% higher clock rate with little or no difficulty. And 3D gaming benefited more from a fast math coprocessor than it did from L2 cache, so those Celerons running at 400 and 450 MHz ran Quake better than anyone expected. A year or two later they got excited about Celerons again, when they figured out how to re-enable Celerons’ SMP capability so you could run two of them on a motherboard, suddenly dropping the cost of building a dual-processor setup by about 75%. Today it’s hard to imagine hardware enthusiast sites getting excited about Celerons, but things were different then. Today, vendors send them review hardware. Back then, it was still mostly guys writing about what they’d bought with their own money.

For a few years in the late 1990s, Intel and the enthusiast sites were constantly playing cat and mouse. If you’re wondering why overclocking is hard to do, that’s why.

%d bloggers like this: