bad mainboards

PC Chips fake cache 486

Let's begin with the most famous of them all: the fake cache 486 boards that PC Chips produced in the mid-Nineties. Take a careful look at it. You can date the board quite easily: it has a Socket 3 connector for a 486 CPU; and VESA slots, so it can't be a very early 486 board (they had no VESA), nor can it be a late model one, because the last of the 486 boards had PCI slots. Middle period 486 then.

Now look at the RAM slots: 4 30-pin and 2 72-pin. That tells you that it is from the transitional phase when 30-pin RAM was on the way out and 72-pin RAM on the way in.

That puts it in the 1994 to 1995 period. Memory can only go so far, but if it is to be trusted these were around in mid-1995. And they were everywhere! They were probably the single most common 486 board you could buy for a little while.

What made them so successful? One thing, and one thing only: the price.

These things were going for AU$80 ex tax in trade quantities. All the other 486 boards were around the $140 mark. And most of the shops that were selling these were not paying sales tax either. (We could go into a long explanation of the now defunct Australian sales tax regime here, and explain how about one third of the entire computer industry used a series of artful dodges to evade paying the 22 percent wholesale sales tax, and damn near ripped the guts out of the entire industry, but perhaps we can leave that for another day.)

Net result was that you could walk into a computer shop and buy a good 486 mainboard for about $180 retail. Or you could walk into the fly-by-night shonkster's shop three doors down the road and buy one of the boards in the illustration for $90. OK, it wasn't legal, because the shonk was dodging sales tax, but he was the one who would go to jail if he ever got caught, not you, so why not? Even if he paid tax it was still only $110. A 486 board is a 486 board, isn't it?

Let's look a little closer.

We can start by looking at the board itself. (Picture at right.)

On the top you can see the same no-name 486 board. The board underneath is a Gigabyte 486VF, another board from the same general era. Notice how much thicker the Gigabyte PCB is? That's the first sign. Some perfectly good main boards are quite thin, but in general thick boards are good boards, thin boards are thin because it's cheaper to make them that way.

Now hold the board in both hands and flex it: it bends like cardboard. Do the same with a good main board (use an old one, because this is not good for them) and it will bend less than half as far with the same force. Nothing conclusive yet, but there is a second black mark.

Now let's look at the chipset. This is where it starts to get interesting.

← Easy enough to see that it's a UMC chipset.

Nothing wrong with that is there?

UMC stopped making chipsets not long after this, or at least stopped making them under the UMC name — UMC is are a major foundry and manufacture all sorts of semiconductors for outside companies under contract — but UMC 386 and 486 chipsets were perfectly OK.

But is it a UMC chipset? It seems an odd way to label it. Compare to the other 486 chipsets at right and below.

Well, perhaps UMC just labeled their chipsets differently, with a peel-on sticky label rather than marking the actual chip itself?

Nope: the chipset on the lower left is another UMC and, just like all the other UMC chipsets we have ever seen, the chips are labelled in the ordinary way.

Still not convinced?

Well, let's peel the label off and see what we get.

There it is, naked and anonymous.

And here at last we have the first of the several unique things about these boards: they had a rather special 'universal' chipset. Now a small shop buying ten pieces at a time didn't get this choice, but if you were a bigger buyer you got the option.

"Want to buy some 486 boards?"

"What price?"

"If you take 100 pieces, $76."

"OK. What chipset are they?"

"What chipset would you like?"

No kidding. That is exactly what went on — we never bought any, but we know people who did, and you could have any chipset you wanted. Well, you could have any chipset label you wanted.

We only remember seeing these boards with UMC or PC Chips labels, but apparently you could ask for anything at all: UMC, SiS, Intel, you name it. After all, from the importer's point of view, if it helped close a sale for 500 motherboards, how much did it cost to get some sticky labels printed up?

Now a couple of small things: look at the heatsink in the picture at right. A lot of boards used cheap and nasty heatsinks for the CPU voltage regulators, including quite a few rather good ones, but it is one more of those small signs that add up to an overall total. As with everything about these boards, the manufacturer has taken the cheapest possible option.

There are no sockets for the cache RAM chips or the BIOS chip either. (See picture below and right.) This meant that you could not replace a faulty cache chip or re-flash a faulty BIOS, but so what? In manufacturing quantities, those ten missing sockets per board probably cost a total of one dollar, and a dollar saved is a dollar earned.

By saving a couple of cents here and maybe a whole dollar there, they had managed to reduce the manufacturing cost of the mainboard by maybe five or ten US dollars.

That made it one of the cheapest boards on the market, but it still should have been costing them somewhere around AU$120 a piece to land boards in Australia. They were already the poorest quality motherboard you could buy, but how did they manage to make a profit selling them for AU$80?

The answer is very, very simple: fraud.

A typical 486 mainboard had 256k of cache RAM. You can see some in the picture below.

Now cache RAM needs to be faster than main RAM or it is useless. In 1995 most standard fast page RAM was rated at 60 or 70 nanoseconds. Cache RAM was usually rated at either 15 or 20 nanoseconds. It wasn't possible to get 20 nanosecond performance out of cheap dynamic RAM in those days, so cache RAM was always SRAM: static RAM, which uses a lot more transistors per bit and costs quite a lot more.

In fact, 256k of static cache RAM in 1995 cost between $40 and $50. It was the single most expensive part of a motherboard. In the early '90s some of the cheapest and scungiest distributors started selling low-end 486 boards with the cache sockets unpopulated. (See picture at right.) This made the boards cheaper but you can imagine what it did to the performance. We saw quite a few 486SX-33s with no mainboard cache RAM and they were very slow; nowhere near as fast as a properly cached one.

→ Illustration: three 486 boards. Clockwise from top left: 20ns cache chips in an unidentified UMC chipset board; unpopulated cache sockets in a Soyo board, about 1994; soldered-in chips in the PC Chips board.

Before too long people got wise to this, and refused to buy boards that had no cache RAM in their sockets. After all, those rows of empty sockets stood out like dogs balls: even the most non-technical buyer would look at the board and ask what the missing stuff was.

Besides, for the retailer, leaving the cache out meant extra service calls. The board was not just a lot slower, if it ever dropped its CMOS setup it wouldn't boot. The factory BIOS default on 486 boards, quite sensibly, was external cache enabled. With cache enabled but not actually present, the system can't boot. You had to go into setup, switch of the cache in BIOS, and save the changes before it could start up. Sure, it was a problem you could solve over the telephone, but then you had to explain why it had happened, and not many retailers like saying 'because, Sir, I sold you a motherboard with some parts missing from it because it was cheaper that way'.

The fashion for leaving cache RAM out came and went quite quickly. By 1995 it was all over.

Here is our friend the anonymous 486 board again. See those black plastic things with metal legs on in the picture below? They look rather like cache RAM chips, don't they? In fact, if you took the trouble to search out the part number in a semiconductor handbook, you'd find that it indicates a 20 nanosecond cache RAM chip. Only trouble is, those black plastic things with metal legs ain't cache RAM chips, they are ... well ... black plastic things with metal legs on.

The board below it is a variation on the theme: the same model of board, the same manufacturer, the same fraud, just different part numbers. (In fact, this second one was faster: the top board took 20ns before it failed to function as advertised; the bottom one, having fake chips marked as 15ns parts, under-performed 5 nanoseconds sooner.)

But surely it must have been a simple matter to detect this fraud? All you had to do is go into the BIOS, switch the external cache on, and see if the system hangs on boot or not? Nope: these boards worked just fine with the BIOS cache setting either way. Well, they used to hang quite a lot, but that was just normal random hangs and crashes, something that cheap boards do all the time, not a complete failure to boot like you used to get with unpopulated cache socket boards.

How could this be so? How could the board run with cache enabled in BIOS but without any cache? Simple: the manufacturer had reprogrammed the BIOS to permanently disable external cache no matter which way you set it, and to report "256k cache enabled" during the boot sequence no matter what.

This too should have been reasonably easy to detect: all you had to do was swap the BIOS chip over with another board, and the deception would be unmasked. And this is why the BIOS chip was soldered directly into the board, instead of being socketed: not just to save using a 10c part, but so that you couldn't easily remove it and detect the fraud.

The really silly thing about the fake cache boards was that they actually worked reasonably well some of the time. Not very fast and usually not for very long, but they went OK with a DX-66 or DX-100 CPU. They tended to have a lot of niggle problems—occasional GPFs, traps and hangs, inability to run certain programs, and they were absolutely hopeless as soon as you pushed the bus above 33MHz—so a DX-80, for example, was out of the question. But of course, by removing the cache RAM and crippling the performance, they had also removed one of the most common problem areas in a mainboard. With no cache issues to worry about, a board has to be really bad to give you problems. These, in other words, were really bad.

→ More fake cache, probably from the same manufacturer, but on a more recent PCI main board. Details here.

Why did people buy them? Because they went to cut-rate fly-by-night dealers and bought on price alone. And because the dealers also bought on price alone. And it made a significant difference to the price: a decent quality board with 256k cache RAM (say an OPTi 895 or an FIC PVT) in mid-95 when these were about was about $140 in trade quantities before tax or freight, those PC Chips things were selling for $80.

The moral of the story? It is very easy to sit back and criticise the dealers who sold this fraudulent crap. But why bother? Most of them are broke by now: you can't pull this sort of stunt forever; sooner or later the public wake up to it. And it's easier still to criticise the wholesalers who imported it — several of the company directors responsible had nice long government sponsored holidays, though unfortunately not for this scam, it was unrelated crimes: tax evasion on a truly massive scale.

And we should rightly blame the Australian trade practice authorities did absolutely nothing about it because they, like their equivalents in other countries, were too ill-educated in matters technical to understand it. Most of all, it is easy to blame the manufacturer who started the whole thing, and we most certainly do. They are now one of the largest mainboard makers in the world, astonishingly enough. We hope that the people responsible for the fake cache scam are no longer employed there, though we have no way of knowing.

But there is a deeper moral: in a way, the real culprits were the hordes of greedy retail buyers who made it all possible. By relentlessly shopping on price, price and price, a great many Australian PC buyers left themselves wide open to this fraud. In the end, the one essential thing that makes con tricks possible and practical is the greed of the victim.

The VX Pro

More recently — up until about the end of 1997 — there were some cheap motherboards with a "VX Pro" chipset floating around at swap meets and with the bottom-dwelling dealers. We didn't see that many because we never sold them, but we got a pretty good idea what they were like by seeing them come into our workshop for repair. (See the Amptron PM8600 entry. Let's quote an American retailer who knew them well:

"They have an imitation "VX Pro" chipset, which gives considerably slower performance than a real Intel 430 VX chipset. How slow? ZD Winbench 97 shows at least 10% slower CPU Marks, 16% slower Disk Win Marks (using latest bus mastering drivers for both boards), and 23% slower CD-ROM and video playback performance, compared to VX chipset boards."

"Second, they are made by PC Chips, renowned as the industry's cheapest, lowest-quality manufacturer. Their defective rates are legendary, between 10-15% in an industry where 5% or less is considered good. PC Chips was also the originator of putting "fake" plastic cache chips on motherboards. With these VX Pro boards you really do get what you pay for. ... The price is unbeatable, but price has always been PC Chip's sole selling point. ... Why would you buy a motherboard made by the industry's absolute lowest bidder?

The full text of this article is now lost, but for the record it used to live here.

Illustration: A BX Pro board from the same manufacturer, which has sat out in the shed for some years gathering, as you can see, a lot of dust. We know nothing about these; it came out of a dead system and is the only one we have ever seen. Not having thought to save a dead VX Pro for our collection, this one will have to do.

But has the worm turned? PC Chips made some half-decent mainboards in the following year or two, using the ALI Aladdin 4 chipset under the 'TX Pro' badge. To our great surprise, the three or four we saw in the workshop ran quite well, even at 75MHz. We could not imagine that the time would ever come when we would start selling this manufacturer's products — not given their appalling history — but it looked as though they might have seen the error of their ways. (A little later, we would indeed sell quite a quantity of them, though without knowing where they had come from. See the details here.)

Meanwhile, remember that poor quality boards come and go. It's a cyclical thing. The industry expands during boom times and marginal new players enter the market. Then as growth slows the weakest dealers look for a way to maintain profitability by cutting costs, and the poorest manufacturers look to grow their market share by providing very cheap bottom-end parts. Short-term survival is the name of the game. Quality and reliability are sacrificed to price, price, price.

The unwary consumer suffers at first, but word gets around eventually and the shoddy-parts dealer's reputation suffers while his warranty costs go through the roof. (Of course, this type of dealer often contrives to avoid providing any warranty service, in which case his reputation gets worse even faster.) Because of his poor quality goods, he looses his original hard-won core of regular customers. Just to stay in business, he has to keep on attracting a steady stream of new customers. The only way he can do this is by cutting his prices even further. Now he can't afford skilled staff or decent components and his quality drops still further. In the end, his reputation is dreadful, his take-home profit is nil, and he goes out of business — leaving his customers without any warranty service, of course, and probably owing a lot of money to the trade.

This is a common pattern. We have seen it again and again over the years. It's a constant part of the PC industry (and probably of most other industries as well). But there are times when it becomes particularly noticeable. Towards the end of 286 days there was a rash of poor quality motherboards, fuelled by the struggling bottom-end dealers trying to get their costs down. It happened again around 1994 and '95 with 486 parts, then again in '97 with the VX Pro. Right now the industry seems stable. It has endured the post-2000 downturn and is starting to recover from it—but still we haven't seen the expected rash of cruddy motherboards. Is it just around the corner, or have people finally learned that quality is always cheaper in the long run?