A few examples of boards from the grand old days. As you will soon discover for yourself, motherboards varied a lot more from one model to another in those pioneering times, and some of them were simply weird.

Unidentified XT board

A very typical XT clone board, indistinguishable from a thousand others. This one is from the golden age of the XT clones, after they had finally pushed aside the CP/M and Apple DOS machines, and while the 286 was still more a high-end promise than a mass-market reality.

If you are used to modern mainboards—something from the Pentium era, let's say — there is quite a lot that is different.

First, notice the expansion bus: eight 8-bit XT slots, which (in XT-class boards) always ran at the same speed as the system and the CPU: in this case, 8MHz. In the over-engineered and very conservative IBM PC, it was 4.77MHz. (Clone systems were often twice as fast as Big Blue's.) Most of the later clone boards ran at either 8 or 10MHz. A very few even ran at 12MHz, but without the decoupled expansion bus that the 286 boards introduced, this was very difficult to make reliable. Remember, the entire system ran to the beat of a single clock. (Modern systems typically have four or five main clocks: board, CPU multiplier, RAM, PCI, and AGP all run at different speeds.)

Next, look at the horizontal row of six sockets just below centre: the blue-labelled chip on the right is the Phoenix BIOS, the five empty sockets are there for IBM ROM BASIC. Why just about every XT clone board ever made had these extra sockets, no-one knows: you couldn't buy the ROMs from IBM—not for a non-IBM main board—so perhaps there were cloned third-party replacement products on the market. Doubtless there were: you could buy almost anything in XT days; this was long before the dead hand of monopoly crushed the innovation and the creativity out of the PC industry. If you can find an old copy of Byte or APC or even Popular Computing, from 1984 say, or even 1989, spend a half-hour reading through the small ads in the back—you will not believe how much choice there used to be.

By the way, have you ever been working on an old machine, a 486 say, and not being able to boot from hard drive or floppy, run into this error message: 'NO ROM BASIC'? The sockets for those missing ROM chips disappeared with the arrival of the 286, but the BIOS error message lingered on for another decade.

Now, look at the switch block (above and right of the BIOS chip). XT class machines all had the exact same switches: everyone used to know them off by heart. The first pair set POST test mode and co-pro present or absent, the second pair memory size, the last pair set the number of floppy drives, but it was the third pair that mattered the most, as they set the video card type: monochrome, CGA, or no video at all. This was the setting that used to confuse people, as for the more advanced display cards (EGA and VGA) you had to set the mainboard switches off. In fact, it was more logical than it seemed at first: the XT BIOS handled mono and CGA screens itself, but EGA and VGA cards had their own BIOS.

  • CPU support: Intel 8088-8 fitted, 8087 socket provided.
  • Speed: 8 MHz.
  • Slots: 8 8-bit
  • I/O: None.
  • RAM: 36 16-pin DIPP sockets, 640k fitted, 640k maximum.
  • Cache: None.
  • Chipset: None, Phoenix BIOS.
  • Date: November 1988.

VLSI 286-10

An old-style 286 board, no doubt quite expensive when it was new. The manufacturer is unidentified but there are several interesting things to notice.

Start with the power supply connector at upper right: it is PC compatible, not the more familiar XT connector that was to become standard for the next ten years or more. This was not a bad idea, as the one piece PC power socket could take a two piece AT power supply plug with a little fiddling and trimming of the plug to fit, but the reverse did not apply; so building it this way left it possible to use the board to upgrade an older system.

Next, look at the chipset: no less than five large square chips, all labelled VLSI, and socketed to allow removal and replacement of them—in those days, main boards were so expensive and labour so cheap that it was economic to attempt to repair them, rather than simply throw them away and plug in a new one as we do today. Warranties were much shorter back then too: now if a board is out of warranty then it is obsolete anyway; this was not true in 1989.

Notice that, although the chipset is a long way away from the single chip and two chip chipsets that were to become common two or three years later, it still manages to integrate most of the system functions. Earlier boards had a great many more minor support chips.

(VLSI, by the way, in amongst their many other products, made chipsets for quite some time. Their last chipset would be one of the very few 75MHz capable Socket 7 designs. Like all the independent chipset makers, their market share would suffer greatly because of Intel's push to extend their CPU near-monopoly into chipsets as well and, unable to compete with Intel on even terms, they stopped trying. VLSI merged with the giant Phillips group in 1999.)

The RAM arrangements are typical for a higher-end board of this period: lots of 16-pin and 18-pin sockets, cleverly interspersed so that you could use either sort. This board has had 18 16-pin 256kb chips (512k plus parity) and 18 64kb chips at the front of the board to make 640K in total, though it would have been possible to upgrade with bigger chips.

  • CPU support: Intel 286-10 fitted, 287 socket provided.
  • Speed: 10 MHz.
  • Slots: 5 ISA, 3 8-bit
  • I/O: None.
  • RAM: 16 and 18-pin DIPP sockets, 640k fitted, up to 4MB.
  • Cache: None.
  • Chipset: VLSI, Award BIOS.
  • Date: 8th September 1989.

DTK PEM-2500 Cache 386

A seriously expensive main board. It's difficult to appreciate just how big this board is until you see it in the flesh—look at the positioning of the keyboard socket! If you have ever wondered why large, full-length boards are called "Baby AT", now you know: this is Full AT. The component count is extraordinary: well over 100 IC chips and somewhere around 300 parts in total.

There are two reasons for the complexity. First, this board has what was then the ultimate in high-performance: 64k of cache RAM. You can see the 10 8KB static RAM chips forming a long horizontal row at centre right. Just to their left is a connector for a proprietary RAM expansion board (which looks a little like an ISA slot). But there are also four long narrow SIPP sockets (rows of round holes, front centre, just to the right of the black expansion slot), and 36 sockets for DIPP RAM (at front left).

The second reason for the complexity is that, unlike all modern boards and even the not-much-younger 286 board above, it was not built around a dedicated, special purpose chipset, but made up out of a host of individual general-purpose parts. Later on there would be integrated chipsets for the 386; and by about 1994 it would be routine to do everything that this board needed 100 chips to do with just one chip, but this was a mainboard in the old style: made with the parts that were available at the time.

Scroll up and look at the VLSI 286 board again: it seems complex by modern standards - especially when you remember that at this time all the peripherals like drives and serial ports were handled by a separate card - but the five big square VLSI chips between them do nearly all the work. Set aside the CPU, the RAM, the three BIOS chips, and the ISA slots, and all that's left is the five-chip chipset and about a dozen little "glue" chips to tie it all together.

But where the 286 board above used a ready made VLSI chipset, the monster at right has just one dedicated chip—the same VL82C100 that is one of the five parts making up the 286 chipset. Everything else is done the hard way. (What does that VLSI chip do? At a guess, it's an I/O controller for the ISA slots.)

It is often difficult to date old boards. Sometimes there is a warranty label still attached, which is usually the best guide. Just the same, these can be difficult to decipher, as every firm has its own way of encoding the vital information. Here, for example, we use a seven-digit label: a single letter supplier code so we know who we bought it from; two digits to show the length of warranty in months; two digits for the day it was delivered to us; one digit for the month (January to September is obvious, zero is October, a decimal point means November and a dollar sign indicates December); and the last digit shows the year. (If we can't guess the correct decade, then it's safe to say it's probably out of warranty!) Other retailers and wholesalers and OEMs have their own systems. Some date codes are obvious, others are obscure and you need to look at several of the vendor's products to puzzle out the system, and some remain forever mysterious. Some places stamp goods with the date the warranty expires instead of the date it starts.

When there is no label, you are reduced to less certain types of detective work. Sometimes you can remember selling Product X at the same time as a Product Y for which you already have a date; sometimes the supplier code does not give the date but at least shows where it came from and if they only started trading in 1992, or closed their doors in 1997 you have a hint. Sometimes there are other components in the system that appear to have been fitted at the same time: if you can't date the motherboard you might be able to date the CPU or the RAM. And sometimes you just have to look at the product, compare it in your mind to other products that were broadly similar, and take a guess.

With this particular motherboard, that was what we did. We looked at the massive form, the high chip-count, the RAM type, the fact that the 386DX-25 only started selling in mid-1988, guessed DTK probably took a few months longer to get their DX-25 product out than ALR and Compaq did, and plumped for early 1989. It was only when we plugged the old girl in to take a screen shot of the weirdo DTK BIOS — yes, she still works just fine — that we noticed that the BIOS is copyright 1990!

Now how the hell did DTK come to be selling dinosaurs like this as late as 1990? And, perhaps an even more interesting question, who would have bought it?

Obviously someone did, and they traded it in to us in about '92 or '93. We sold it, traded it back a few years later, sold it again, and when it came back for an upgrade the next time, retired it to the dust-gathering department.

There is another close-up view of this board in the entry on the 386DX-25 CPU.

  • CPU support: Intel 386DX-25 fitted, 387 socket provided.
  • Speed: 25 MHz.
  • Slots: 6 ISA, 2 8-bit, proprietary RAM expansion.
  • I/O: None.
  • RAM: 16 and 18-pin DIPP sockets, 2MB fitted, 4 SIPP sockets, probably up to 64MB.
  • Cache: 64k 25ns, probably expandable to 256k.
  • Chipset: None, DTK BIOS.
  • Date: 1990.

Unidentified 286-16

This unidentified but very typical late-model 286 board was much the same as dozens, perhaps hundreds of its competitors in the final days of the 286 as an in-demand new product. Many of the modern features that we would expect the boards of the mid '90s to have were already present. The board is small and neat with a low chip-count, a simple, clear layout, and well-organised minor connectors — something that most boards of the early 00's comprehensively fail to achieve, by the way.

The CPU (lower right) is surface-mounted. This was to be by far the most common method right through the 386 era and up until the fashion swung over to the LIF sockets that the early 486 boards used.

There is a later revision of the same main board below; it's worth comparing the two.

  • CPU support: AMD 286-16 fitted, 287 socket provided.
  • Speed: 16 MHz.
  • Slots: 5 ISA, 2 8-bit
  • I/O: None.
  • RAM: 20-pin DIPP sockets, or 4 30-pin SIPP sockets up to 16MB.
  • Cache: None.
  • Chipset: Headland H12A, AMI BIOS.
  • Date: June 1991.

CH-386-33A

A couple of years can make a great deal of difference. This board is barely half the size of the monster above, would have cost perhaps a quarter as much, and is unquestionably superior in all respects. The chip count is down to about 25 with just three main ones: the OPTi 391 and 392 integrated chipset, and a Chips and Technologies I/O controller. The vast array of DIPP RAM sockets has gone, replaced by a simple and much more practical set of eight 30-pin SIMM slots. Presumably, the designer was not sure that SIMM RAM would take off in the market, as he has also made provision for a proprietary RAM card slot. (You can see the pads where it could have been factory-fitted on the extreme left of the board.)

This board retains the 256k cache of the older board above, but with the benefit of an integrated chipset can do this with vastly less complexity. The eight socketed 256kb static RAM chips are at lower right, together with three 8KB tag RAM chips. (Why three? Three 16kb x 4 chips seems an odd arrangement.)

Perhaps most interesting of all though is the CPU: a socketed AMD 486DX-40. At the time this would have dropped the price of the board and CPU package by no more than $30 or $50, and upped the performance in comparison with an Intel 386DX-33 by a small but noticeable amount, but it was the first sign of the price-performance revolution that was to come.

  • CPU support: 386DX: AMD 386DX-40 fitted, 387 socket provided.
  • Speed: 33 or 40 MHz.
  • Slots: 6 ISA, 2 8-bit
  • I/O: None.
  • RAM: 8 30-pin SIMM sockets up to 32MB, proprietary RAM expansion slot not fitted.
  • Cache: 256k 20ns.
  • Chipset: OPTi 391/392, AMI BIOS.
  • Date: 5th May 1992.

Jaton JAM-3302 386SX-20

Not one of our own boards and not a common one either, but interesting because of several quite unusual features.

First look at the connectors at top right, next to and below the power socket: there are serial and parallel ports, together with floppy and IDE controllers. This was long before having an I/O section integrated became common, and it achieved nothing - as it was done by simply taking the same chips that would be used to make an orthodox IDE I/O card and putting them on the main board instead. Integrated I/O only became sensible in '95 when the new breed of PCI chipsets added control logic to the chipset itself and thus saved the cost of a separate chips.

While you are looking at that part of the board, notice the heavy green corrosion stains: this board spent several years lying at the bottom of a pile of junk outdoors and it is astonishing that it can still POST. The chipset is unusual too: you almost never used to see Intel chipsets in those days. They were generally regarded as slow, stodgy and expensive, and the products of the major chipset makers of the day - companies like Chips and Technologies and Headland and OPTi - were much more popular.

Neatly surface mounted below the ISA slots are the RAM chips - another unusual and unpopular design decision, as you could not replace a faulty chip without replacing the entire board. (Mind you, given the history of this particular board and the fact that it still works, perhaps there was method in their madness!)

Finally, the Quadtel BIOS. This was not unusual, quite a lot of 286 and 386SX boards had Quadtel BIOS, and although it had its quirks, it was generally just as good as any of the other common ones - AMI, Award, Phoenix - and clearly better than the always buggy DTK product.

  • CPU support: 386SX: Intel 386SX-20 fitted, 387SX socket provided.
  • Speed: 20 MHz.
  • Slots: 5 ISA.
  • I/O: Integrated: Goldstar I/O controller on board.
  • RAM: 1MB surface mounted4 30-pin SIMM sockets, up to 16MB.
  • Cache: None.
  • Chipset: Intel S82343/344, Quadtel BIOS.
  • Date: 1992?

Early BIOS makers

The development of the generic PC was made possible by the efforts of the earliest BIOS manufacturers. Although the task of making a functional BIOS was a relatively straightforward one, the market dominance of the IBM PC and the appalling coding habits of many programmers meant that it was not sufficient to have a BIOS that worked, it had to be functionally identical to the IBM PC/XT BIOS or many programs would not run. Companies like Chips and Technologies and Phoenix pioneered PC-compatible BIOS in XT days, and went to great lengths to make sure that their products were both legitimate and compatible. By the early Nineties though, a decently functional BIOS could be almost taken for granted, and the main interest of the working technician became ease of use. In the workshop, time is money, and in the field a good BIOS interface meant fewer service calls.

→ On the right, the most common BIOS of them all until the early Nineties: AMI.

← On the left, AMI's chief competitor, Award. Both were simple enough to use and both tended to be trouble-free. The usual keystroke to enter AMI BIOS was DEL or sometimes ESCAPE, for Award it was normally CTRL-ALT-ESCAPE.

AMI introduced a new and much more comprehensive BIOS around the time that the 386 became popular, AMI Color BIOS, which was clearly the best of them all and was to become all but universal. In general, only the proprietary box makers like Commodore and Compaq would use less a friendly BIOS, typically the compatible but poorly organised Phoenix one. (Why is it that the box makers to this day nearly always choose such dreadful BIOS interfaces? Is it that they have less respect for their users than the people who make real PCs? One assumes so.)

Now two of the less common ones.

← On the left, Quadtel. Quadtel BIOS was quite common in 386SX machines and once you got used to its oddities it was rather good. You could do interesting things like allocate EMS and XMS memory in hardware, which was quite a useful trick in the days before QEMM revolutionised the art of memory management under DOS. You might remember that the Quadtel video BIOS powered the ubiquitous Trident 9000 512k ISA graphics cards too.

→ On the right, DTK, which was thankfully not common. It had a very practical interface, complete with zooming animated boxes that are not something we can capture in a screenshot but would not have looked out of place in a kindergarten, but DTK BIOS was always questionable on the compatibility side. We have long since forgotten which software packages it broke - there were not that many but enough for it to be a problem - and it usually appeared on cheap and slightly odd little motherboards. The board we used to get the screenshot, though, was none other than the monster 386DX-25 pictured near the top of this page.

Unidentified 286-16

A later revision of the same main board pictured above, intended to eke out a few more sales before the 286 disappeared forever.

Notice the single-chip Headland chipset at centre right and the yellow retailer's warranty sticker on the EEPROM just above it. The date: 30th November 1992.

RAM arrangements were state of the art: sockets for 1MB of DIPP RAM at front left and centre, and instead of the SIPP sockets of the older board, four of the new-style 30-pin SIMM slots at left.

At this time, DIPP RAM and SIPP RAM was just giving way to SIMMs on the general market; this particular board was shipped with eight Sanyo 80ns 4256 DIPPs (4 bits wide by 256kb) but as many others had four 256k SIMMs instead, or even an extravagant pair of 1MB SIMMs to make 2MB in total. Most 286 boards with dual RAM connections allowed you to fit a mixture of the two types; something that most 30-pin/72-pin 486 boards and 72-pin 168-pin Pentium boards did not.

As all the better boards were soon to do, it allowed you to choose whether to fit parity RAM or not. This board has the four parity RAM sockets left unpopulated, so you would have had to take care to disable parity in the BIOS.

  • CPU support: AMD 286-16 fitted, 287 socket provided.
  • Speed: 16 MHz.
  • Slots: 5 ISA, 2 8-bit
  • I/O: None.
  • RAM: 20-pin DIPP sockets, or 4 30-pin SIMM sockets up to 16MB.
  • Cache: None.
  • Chipset: Headland H12A, AMI BIOS.
  • Date: November 1992.

Unidentified 486 SX-20

An excellent illustration of the perils of being an early adopter: this board, for all that it would have cost a small fortune when it was new, had all the instant snap of a slug on Valium.

We first met it in early 1994 when we pulled it out of an upgrade, and not knowing what else to do with it, kept it for a while until we sold it second-hand to a lady who had not much to spend and no knowledge of computers, but was absolutely certain that anything less than a 486 just would not do. Despite our firm advice to take a 386DX-40, she insisted on choosing this slow and ugly thing. In reality, any 386SX-33 was faster, and even a good 386SX-25 would have run it close.

Someone added RAM to it later on (notice the mismatched eighth SIMM) but she kept it for an incredibly long time, eventually trading it back in on a two-year-old K5 or 6x86-200 or something of that order. We remember being surprised that she didn't want a Pentium: "Oh no", she said, "I ignored your advice the first time and regretted it ever since. That thing is really slow."

So far as we know, her K5 is still running Windows 95 just fine, and no doubt we will see her for an Athlon XP in 2005 or so. (We don't normally recall such detail from so long ago, of course, but this remains the only 486SX-20 we have ever seen and is memorable for that reason. If it is a fair guide to the performance of SX-20s in general, it is no wonder it was the only one!)

(A late note in 2004: our 486SX-20 lady is now the very happy owner of an Athlon XP.)

All that, however, is more the fault of the CPU than of the board; it probably would have gone perfectly well given a 486-33 and a different clock crystal. Notice that the chip count is gradually decreasing (as compared with the older boards above), and that although it is an early generation 486 board — Intel only announced the SX-20 in April 1991 and these boards probably hit the market not long after that — it is not much more complex than the CH-386-33A 386DX-40 board above — just twice the price and a whole lot slower.

  • CPU support: 486: Intel SX-20 fitted, Weitek socket provided.
  • Speed: 20 MHz.
  • Slots: 6 ISA, 2 8-bit
  • I/O: None.
  • RAM: 8 30-pin SIMM sockets up to 32MB.
  • Cache: None.
  • Chipset: OPTi 493, AMI BIOS.
  • Date: November 1992.