This was the last year of the wonderful old 386DX-40. They are hopelessly off the performance pace now, of course, and unable to run any present-day software, but they were very competitive in their time and to this day remain the simplest, most fuss-free boards we ever worked with.

Forex 486DLC

The 486DLC used the same pin-out as the 386DX, so in theory you could use any 386DX main board to run a DLC.

In reality this was not so: most DLCs came in PGA form and were sold as individual chips, where the vast majority of 386 parts were surface mounted onto the motherboard at the factory and sold as a unit. In any case, you needed an appropriate BIOS update, and in those days before flash RAM, that was a factory job too (unless you owned an EPROM burner).

So although the main boards used by DLCs were simply modified 386 boards, it is usual to think of them as a separate class of product. In any case, there were only a relatively small number of manufacturers that took the trouble to make a DLC-ready version of their boards.

The board illustrated, like nearly all these early ones, is unidentified. It's not one of the three or four that we sold in some volume, but is broadly similar. The wholesaler's warranty sticker dates it: 20th December 1993.

  • CPU support: 386DX or 486DLC, 387 socket provided.
  • Speed: up to 40MHz.
  • Slots: 1 VESA, 6 ISA, 1 8-bit
  • I/O: None.
  • RAM: 8 30-pin SIMM sockets, up to 32MB.
  • Cache: 128k 20ns, expandable to 256k.
  • Chipset: Forex, AMI Color BIOS.
  • Date: 20th December 1993.

PX486P3

When you pick up an old mainboard and it looks instantly familiar, it's a fair bet that you sold quite a few of them at one time. The warranty sticker shows that we sold this one in February 1994, though with a 486SX, not the DX-33 in the picture. (In fact, it was an SX-25 clocked at 33 MHz: 486SX-25s were nearly always capable of 33MHz without the slightest fuss.)

The PC industry moves in stages: a new invention comes along (the 486 CPU, in this example) and the surrounding parts evolve quite rapidly towards perfection — almost never to perfection, but in that direction. For any given set of requirements and at any given technological level, there is often only one best way to fill those requirements, with just minor variations possible.

In 1994 a competitive 486 system needed eight SIMM sockets, three or four ISA slots, at least one VESA slot for the video card and perhaps a second for the I/O card, some cache RAM, and the ability to accept any of the current 486 CPUs. There was no need for anything much else and the PX486P3, like all of its competitors, was focussed on providing those essentials at the minimum cost.

Although on first sight there seems nothing out of the ordinary about this board to make it interesting, that is exactly the point! The logic of competition had seen to it that nearly all the 486 boards on the market at that time were functionally identical: this one is just big enough to fit the RAM and VESA slots (because making the board larger costs money and achieves nothing), it is jumper settable for a variety of clockspeeds (because it costs more to have several different boards where one would do), it has sockets for 256K of cache, and it does all this with the smallest number of components possible.

In nature, this is called 'convergent evoluton' and it explains why an English wren and an Australian wren are almost identical, right down to details of plumage and behaviour, despite having no genetic relationship at all to speak of. Because both birds have the same 'job description', and there is only one best way to comb the surface of the earth for small insects, survival of the fittest sees to it that English and Australian wrens have the same shape and habits. Similarly, any manufacturer that made a mainstream 486 board in '94 that was not more-or-less the same as the PX486P3 could be sure of loosing money on it.

When the technological ground rules are stable for a time, the only competitive advantage that can be found is price, and it is quite common to see price competition in stable markets force two sorts of change: on the one hand you get innovation to cut manufacturing costs by using different components (which is exactly what the 486 boards of '94 had already done — one and two-chip chipsets, smallest practical board size, and so on) and on the other hand, you get vendors cutting corners on quality and performance (for examples of this, see the contemporary 486 boards with missing or even fake cache).

But sooner or later, another set of ground rule changes comes along. This is exactly what was to happen with the 486 market. This board and all the others like it ruled the market for a time, then there was suddenly a need to support the new breed of non-Intel CPUs, and a step-down voltage regulator for the 3 Volt chips, and LBA addressing for bigger hard drives, and then 72-pin RAM. Like dinosaurs faced with global climate change, the board manuacturers had to adapt or dissapear.

  • CPU support: Any 5 Volt 486.
  • Speed: 16 to 50 MHz.
  • Slots: 3 VESA, 3 ISA, 1 8-bit
  • I/O: None.
  • RAM: 8 30-pin SIMM sockets, up to 32MB.
  • Cache: 256k 20ns.
  • Chipset: OPTi 495SLC, AMI Color BIOS.
  • Date: 4th February 1994.

Biostar 386DX-40

The 386DX-40 had a phenomenally long lifetime. There were other families of CPU that lasted longer, but no single speed grade would ever stay the number one seller for so long again.

There were several reasons for this. First, the 386 replacement parts were no great improvement to begin with: even with VESA video a 486SX-33 was not a must-have upgrade; it wasn't until the 486 parts hit 66MHz that they became compelling. Second, with most people running DOS or Windows 3.1, a good 386 was still perfectly usable; Windows 95 would not become common for two or three years yet. Third, AMD had the most powerful of incentives to keep on making 386DX-40s: Intel had the AMD 486 tied up in the US courtrooms, so AMD had to keep on making 386 parts or close up shop, and Intel, with the 486 market to themselves for just a little longer, could sit back and enjoy small volumes but nice high prices. Finally there was the cost factor: 386 boards became ever smaller and simpler and cheaper to manufacture.

→ We sold a lot of these Biostar DX-40 boards, and with perfect satisfaction. They, and a few others similar, were the last incarnation of the DX-40 as a mainstream product: very simple, utterly reliable and shrunk to the absolute practical minimum. There is not a square inch of wasted space. Notice the co-pro at bottom right: quite unusual in a 386. It was probably fitted as an aftermarket part: 387 chips were quite expensive and by that time it would have made more sense to buy a 486DX, or perhaps a 486DLC with co-pro, unless you already had the 386.

The only place left to cut the manufacturing cost further was by going back to only four SIMM sockets, and that's exactly what the next and final batch of DX-40 boards did.

This was a mistake. By then the DX-40 was very much an entry-level product, selling in small numbers and mostly being used for upgrades, perhaps to replace a 286-16 or a 386SX-20. Because the 386DX had 32-bit RAM addressing and 30-pin SIMMs were 8 bits wide, you had to fit four SIMMs at a time (not two, which you could do with a 386SX or 286). So where a four-slot 386SX board could have 1, 2, 4, 8, or 16MB, a 4-slot 386DX could only have 1, 4, or 16MB, using 256k, 1MB or 4MB SIMMs. No-one who could afford 16MB of RAM was buying 386 boards by then, and no-one would shell out the price of a new main board just to put 1MB in it, and that left the 4MB option only, cutting out both the very cheap 8 x 256MB option and the 8MB option — both of which made quite a deal of sense, one for a fast DOS machine, the other for a machine that cost about the same as a 486SX-33 with 4MB but would run Windows 3.1 noticably faster.

(It would be nice to round out our tour of the 386DX-40 with an illustration of one of those 4-slot final versions. Unfortunately, we don't have one and are unlikely to see one now: we didn't sell very many of them in the first place and with their RAM limitation they were the first ones to be discarded.)

  • CPU support: 386: AMD 386DX-40 surface mounted, IIT 387 fitted to this example.
  • Speed: 40 MHz.
  • Slots: 5 ISA, 1 8-bit
  • I/O: None.
  • RAM: 8 30-pin SIMM sockets, up to 32MB.
  • Cache: 128k 15ns.
  • Chipset: Biotech 3491, AMI Color BIOS.
  • Date: 20th March 1994.

OPTi 495SLC

We have no idea who the manufacturer was after all these years, but we sold a great many of these excellent boards.

They were based on OPTi's extraordinary 495SLC universal chipset and could take any CPU from 386DX-25 through to 486DX/4-100. Yes, both 386 and 486, on the exact same board. All you had to do was set the jumpers and swap the chip.

Mostly we used them for the 486DLC, with and without co-pro, in 33 and 40MHz flavours, but we put quite a few AMD 486DX/2-66 chips into these boards too, and maybe one or two DX/4s.

Illustration: OPTi 495SLC main board with TI 486DLC-40 CPU. There are four main squares front and centre. Clockwise from bottom left: a large empty socket; the 486DLC (you could put a 386DX there instead if desired); the 496SLC chip itself; and two sets of silver pads for factory mounting of a PQFP (surface mounted) CPU, which could be either 386DX (the small inner set of pads) or 486SX. (486DX too, for that matter, but no-one would have done that with a DX-33, they were far too expensive to ever consider permanently mounting one on a motherboard — imagine if the board failed!)

Aside from having a very tricky jumper setup — understandable given the ultra-flexible nature of the board — these had only one other quirk: if you fitted a DLC CPU, an obscure and counter-intuitive BIOS setting was needed for stability. This was perhaps the only time that we have ever discovered anything useful in the back half of a main board manual. Come to think of it, we still have manuals for these, they might tell us who the manufacturer was. But probably not — main board manuals in those days were informative about jumper settings (bar the obligatory mistake or two) but almost never mentioned the manufacturer by name. This is so that the motherboard maker could chase big OEM sales without having to do a special edition of the manual — OEMs used not to like having any brand name in their machines except their own.

Great care required! A close up of the same board with a DLC maths co-pro correctly fitted. Notice that while a 486SX or DX filled the entire socket and was oriented as marked on the board (the white pin 1 marking is easily visible at the lower right of the CPU socket in the picture above) the 387 or 487DLC maths co-pro did not go in the same way. None of the parts were keyed. Essentially, you had a simple choice: get it right the first time, or let the smoke out. With DLC co-pros selling for around $200 a pop at the time, a little RTFM was recommended.

  • CPU support: 386DX or 486DLC, 387 socket provided.
  • Speed: up to 40MHz.
  • Slots: 2 VESA, 7 ISA
  • I/O: None.
  • RAM: 8 30-pin SIMM sockets, up to 32MB.
  • Cache: 256k 20ns, or 128k if desired.
  • Chipset: OPTi 495SLC, AMI Color BIOS.
  • Date: Probably mid-1994.