A History of Overclocking Part One

In the beginning: 8088
(1983-1984, 3µm process, 4.77-8MHz)

Wind back your temporal distortion machines to 1983, a time when Intel was just behind the scenes and supplied the processor to the indomitable International Business Machines, or IBM. PCs were sold whole rather than in components and IBM was the dictator of the market, releasing its 16-bit 8088-based XT (eXtended Technology) to great success. The 8088 was slightly crippled in comparison to its predecessor the 8086, in that the 8088 could only communicate externally through an 8-bit data bus while the 8086 had a 16-bit one. The wider bus made the 8086 expensive, and so IBM decided to cut costs. The FSB was the only bus – everything ran at the same speed.

Intel’s 8088 was kept at 4.77MHz by IBM, which wanted stable corporate machines, however Intel went to 8MHz, and clones managed a mighty 10MHz – notably from AMD, which had a licence to Intel’s x86 family. Overclocking could be done by desoldering the clock crystal and adding in an after-market one, but this often resulted in instabilities due to everything running on the same bus.

The Math Co-processor was released, an extra unit that did floating point calculations – as CPUs had no FPU at this stage to cut down on production cost – and required an extra socket. This didn’t benefit many people at this stage purely because of the penetration of the personal PC in the market, and the small amount of software actually available needed to be programmed to talk to the FPU directly. As games didn’t take advantage of it at this stage, only CAD/spreadsheet/3D users were assisted, and it wouldn’t be until the 386 that the idea became beneficial for the mainstream and would really take off. Until that point CPUs did integer calculations only – they could still figure out numbers with decimal points, they just had to do it with a bunch of extra integer calculations, meaning they were a whole lot slower at doing it than a dedicated FPU.

RAM was rated at about 210ns – there was no need for any other spec, it simply ran at the speed the bus did. It wasn’t dictated by the bus speed though, RAM was fully asynchronous. In the XT it came as a Dual Inline Package (DIP) module, the default being 64KB in size.

80286
(1984-1986, 1.5µm process, 6-12MHz)



Push forward to 1984 and IBM is set to follow up the success of the XT with the AT (Advanced Technology), based on Intel’s new 80286 processor, and with a full external 16-bit data path.

The AT was initially available at a mammoth 6MHz or 8MHz, and while there were stories and myths about being able to get the older XT to go faster, the new system presented the consumer with viable overclocking for the first time.
By offering two models, people began to mix and match components.

Getting the extra 2MHz out of the 6MHz system required swapping in the faster 8MHz clock crystal, and for later revisions a BIOS chip as well to bypass overclocking protections once IBM cottoned on. This inevitably required soldering.

Fortunately at this stage the ISA bus had already been decoupled from the FSB, and so add-in boards usually didn’t suffer from overclocks – everything else however was susceptible.

Since everything but ISA ran off the same clock, overclocking meant everything would run faster, inevitably resulting in hardware that couldn’t keep up and software that errored, as many programs were tied into the system speed rather than their own timers. This meant apps, especially games, ran faster than intended and in some instances became unplayable or simply crashed with divide by zero errors. Applications became unstable or simply didn’t work.

This gave rise to the ‘turbo’ button – a button that appeared on PC cases during the 286-to-486 era. While most saw it as something that made your system go faster, its actual purpose was to slow down fast PCs for compatibility with these older programs, either by clocking the CPU down directly or disabling the cache, forcing it to always retrieve from the slower main memory. Most mainstream consumers simply left the turbo on, assuming the programs that didn’t work were broken. Still, definitely a form of overclocking.

An Intel licensee called Harris Semiconductor produced 286s capable of 16, 20 and 25MHz, far outstripping Intel’s offerings – even though this was seven years later than Intel, and by this stage the 486 was already on sale. AMD was still an Intel licensee at this stage, but this situation wouldn’t last long.

The 286 featured two operating modes – Real Mode, that essentially made it a very fast 8086 that could access 1MB of memory, or Protected Mode that while offering a number of advantages, its biggest one – other than the ability to terminate misbehaving programs without bringing down the whole system – was being able to offer up to 16MB of memory to applications. The biggest problem was that you had to reset the PC to switch between modes, ignoring the fact that most PCs didn’t have more than 1MB of RAM in those days anyway. As a consequence, Protected Mode was mostly ignored until the 386.

At this stage memory was Fast Page mode RAM (FPRAM), in 30-pin Single Inline Pin Package (SIPP) format. Rather than specifying a row and a column every time to isolate a location and read or write a bit, FPRAM kept a row highlighted and strobed the columns if there were successive entries in the row that needed attention, saving on RAS precharge time.