At any given time we usually had one or two boards that stood out from the crowd and went into most of our new systems. At the time these boards were current (1996 and early 1997) the superb FIC PA-2005 was the one to have, though the Gigabyte 586s was also popular.

It's slightly misleading to call these ‘early’ Socket 7 boards as there were a couple of generations of Pentium-capable boards prior: first the dreadful horrors that the Pentium-66 used, then some only marginally less attractive things that ushered in the era of the ‘real’ Pentiums (3.3 volt parts like the 90, the 75 and the evergreen 100). But those very early models sold in tiny handfuls; the boards on this page are the ones we used in volume once the Pentium era really started. Most were best-sellers in their time and we still used to see a few of these out there in older Red Hill systems past the end of the century.

Chaintech 486

Chaintech 486SPM

These were our best-selling board for most of 1996. Fast and reliable, they were delightfully simple to set up once you learned the odd-ball jumper arrangements. (These days, when we see one, we need to drag the manual out and puzzle over it for a while — maybe we just used to think they were simple because we saw so many!)

The Chaintechs had an unusual and very refreshing set of rules for RAM compatibility: any RAM, any slot. With most 486-era boards, you had to pay attention to the order you used the RAM slots in, the particular type of RAM, and the mix. Not these!

The 486SPM supported all the 486 and 5x86 CPUs; we mainly used it with the Cyrix 5x86-100 and, perhaps making an over-fine distinction, preferred the contemporary FIC PIO-3 (below) with the AMD 5x86-133 .

Chaintech 486
  • CPU support: Any 486 or 5x86, 25 to 133MHz.
  • Speed: 25, 33, or 40MHz.
  • Slots: 3 PCI, 4 ISA
  • I/O: 2 32-bit HDC, FDC, 2 serial, one parallel, LBA.
  • RAM: 4 72-pin Fast Page up to 64MB.
  • Cache: Socketed, 256k asynch standard, 512k optional.
  • Chipset: SiS, Award BIOS.
  • Best with: Cyrix 5x86-100
  • Date: 2nd May 1996.

FIC PIO-3

The last of a long line of first-class 486/586 boards from FIC, and our number one choice for the hugely popular AMD 5x86-133 CPU.

These were the most reliable motherboard we had seen since 386DX-40 days. If you want things dependable, there is nothing quite like a mature, proven technology. Though the 5x86 was hardly a speed-king even when these were new, you'd be surprised how well these went if you give them 16MB of RAM and a decent hard drive. Our main shop demo computer still used one of these boards right up until 1998 and it was plenty fast enough, even for most multi-media applications.

The PIO-3 had a long market life and found its way into a lot of office workstations well after its leading-edge days were over. Alas, though we stockpiled several dozen when they went end-of-life, they are long gone now. It was the end of the line for new 486 boards, and the end of an era.

  • CPU support: Any 486 or 5x86, 25 to 133MHz.
  • Speed: 25, 33, or 40MHz.
  • Slots: 3 PCI, 4 ISA
  • I/O: 2 32-bit HDC, FDC, 2 serial, one parallel, LBA.
  • RAM: 2 72-pin Fast Page, up to 64MB.
  • Cache: Socketed, 256k asynch standard, 512k optional.
  • Chipset: VIA Pluto, Award BIOS.
  • Best with: 5x86-133.
  • Status: Legacy.

Chaintech 5IEM

The FX was Intel's first mainstream Pentium chipset, and the first really usable one. Before the FX arrived, Pentium boards were less than wonderful. Much less. The FX though, brought Pentium systems out of the realm of the exotic and into the mainstream of everyday computing.

Chaintech's 5IEM was as typical an FX board as any, though made with a little more care and attention to detail than some.

The cache RAM on these is interesting: notice the row of eight socketed cache chips at lower right: this was one of the last generation of boards to use the old 486-style asynch cache RAM. As you will see if you read on, the industry moved to the faster pipeline burst cache soon after this.

In the board illustrated you can see an empty brown COAST slot at lower centre, just right of the CPU socket. Before much longer these same boards shipped with the cache sockets left empty and a pipeline burst COAST module in the slot instead. This improved performance noticeably but made the 5IEM very difficult to fit into desktop cases, because the COAST socket was a long way over toward the right-hand side of the board and the cache module would usually foul the hard drive bay. Sometimes a pair of tin snips was the only answer.

On the other hand, the 5IEM was moderately priced, as reliable as any, and generous with ISA slots — having no less than four, which was very handy for people with lots of add-on cards. In those days, remember, nearly everything used ISA; PCI was just for video cards and perhaps a high-end SCSI host adaptor.

  • CPU support: 6x86 120, 150 & 166, P54C to 166.
  • Speed: 50, 60, or 66MHz.
  • Slots: 4 PCI, 4 ISA
  • RAM: 4 72-pin FPM or EDO, up to 128MB.
  • Cache: Socketed, 256k asynch standard, PLB optional.
  • Chipset: Intel Triton 430FX, Award BIOS.
  • Best With: Pentium.
  • Status: Legacy.
FIC PA-2002 F12/20/5/96

FIC PA-2002

A popular board for a while, forerunner of the excellent PA-2005, and one of the first half-decent non-Intel chipset mainboards for Socket 7, but not a board to love by any means: these were tricky to set up right and fussy about RAM and CPU matching, although perfectly practical once you got the hang of them. The PA-2002 was at its best with a Pentium and could be problematic with Cyrix or AMD CPUs in those early days. We preferred the then-current Chaintech 5IEM or QDI Explorer II for the 6x86.

One unexpected bonus of the Apollo chipset, almost unique amongst Socket 7 boards, is that the memory management unit could address a single 72-pin SIMM module if required. This imposed a slight performance penalty, so it was usual to fit RAM in pairs.

The humble 2002, essentially an entry-level board with what was then a little-known chipset, seems a long way away from the best-selling VA-503+ of 1999-2001, but this board was the start of the independant motherboard makers long, slow fightback against the market domination of a single company, and is significant for that reason.

  • CPU support: 6x86 to 166, K5, P54C to 166.
  • Speed: 50, 60, or 66MHz.
  • Slots: 4 PCI, 3 ISA
  • RAM: 4 72-pin FPM, EDO or BEDO, up to 128MB.
  • Cache: COAST socket, 256k pipeline burst standard, asynch or PLB up to 1MB optional.
  • Chipset: VIA Apollo 570M, Award BIOS.
  • Best With: Pentium.
  • Status: Legacy.

The PA-2002 provides a good illustration of the type of difficulty motherboard manufacturers always faced with new technology: at the time this board was designed, the K5, the 686 and the Pentium MMX were themselves still under development. While CPU and motherboard manufacturers work closely together on future products, and exchange hand-built engineering samples with each other, there is no substitute for actual testing in the field.

Notice how a ground-breaking product is almost always troublesome for the first six months or so it: the VESA and PCI local buses, PCMCIA, Windows 95, high-speed serial ports, and Plug and Play were all fraught with difficulties in the early days, but settled down into stable, reliable performers as time went by and the bugs were sorted out, one by one. (OK, Plug and Play wasn't anywhere near sorted out until early in the new century — for years we used to call it "plug and pray", though "plug and swear" might have been more appropriate — but most things settle down faster than that.)

The conclusion is obvious: try to sit six months or so behind the 'bleeding edge' of new technology. At the time the PA-2002 was new, the behaviour of Pentiums was well understood, and the combination of the two generally produced a stable system. In contrast, the K5 and 6x86 were very new and less well understood, and the combination was correspondingly less likely to be stable.

BEK 5719

So long as the 486 reigned OPTi was the world number one chipset maker. OPTi 386 products were excellent and their 495 and 895 486 chipsets were generally regarded as the best ones on the market.

Then Intel, long the dominant CPU maker but a very ordinary chipset maker in those days, decided to get serious about their long-neglected chipset division. Intel used its dominance of the CPU market to bring on a number of new technologies which were designed to be (a) universal standards, and (b) standards that Intel controlled and could manipulate to suit itself. The PCI bus and the Pentium architecture were mighty weapons in the hands of a firm with the power to dictate the ground rules. Even more powerful, though, was Intel's willingness to lean on PC OEMs and motherboard manufacturers who bought non-Intel chipset products. As a chipset maker, Intel went from nowhere to world number one in just a couple of years. Before long Intel had more than 50% of the chipset market with the next biggest player having just 14%. Like all the independent chipset makers, OPTi found the going difficult after this. OPTi did not manage to produce a PCI 486 chipset (at least not one that we ever saw). Nor did they transition to the Pentium promptly.

So when the long promised OPTi Viper chipset finally arrived, remembering their excellent 386 and 486 chipsets, we expected that it would be a humdinger. (After all, how hard could it be to improve on the unlovely FX and VX chipsets anyway?) We were to be sorely disappointed. It seems that the Viper had only one thing going for it, a low price, and with very few sales to keep them going OPTi had only one place left to go: dissolution of the company. The remains were up for sale for several years, but with no viable products, no assets to speak of, and no prospects, there were no buyers.

The independent motherboard makers of the world would realise before too long that by caving in to Intel's chipset division they would only make themselves more vulnerable to Intel's fast-growing motherboard division, but this came too late for OPTi, and too late for most of the others. By the end of the decade there would be only four chipset makers of any significance left: VIA, Intel, SiS and ALI.

Above. Not a wonderful motherboard. We had tried a couple of earlier BEK mainboards for 486 chips and been less than happy with them. This one though would probably have been better had it not been let down by the OPTi Viper chipset. We sold just one or two of these and soon realised that despite the magic of the OPTi name they were below par.

One of them managed to stagger on for long enough to be traded in a few years later and find its true role: a curiosity for our collection. (Why does the board have both asynch and PLB cache fitted? Presumably we were swapping the cache around, trying to find a way to make it work more reliably. Not with any success, obviously, otherwise we would have sold it.)

  • CPU support: 6x86 to 166, K5, P54C, P55C.
  • Speed: 50, 60, or 66MHz.
  • Slots: 4 PCI, 4 ISA
  • RAM: 4 72-pin FP or EDO, up to 128MB.
  • Cache: 256k 15ns or COAST module.
  • Chipset: OPTi Viper, Award BIOS.
  • Date: February 1996?

QDI P51437 Chariot

Another Intel FX chipset board, one of four or five very similar ones made by QDI, and noteworthy as one of the first boards to go beyond the then-new but always unlovely COAST slot cache mounting method.

Traditionally, you mounted RAM in sockets. Going back to XT days and earlier, the idea of socket-mounting RAM was that a bad RAM chip could be replaced easily, and that you could start with a modest amount and then upgrade later. Cache RAM, when it came along in 386 days, followed the same rules. Only very cheap and nasty products soldered their RAM in.

Then came the switch to pipeline burst cache. A typical pair of PLB cache chips had 100 pins each and it wasn't practical to mount them in DIPP sockets. Instead, the industry switched to COAST slots with the two PLB chips surface mounted onto a small daughter board. (See the picture of the PA-2002 above.) At least in theory, you could still add extra cache RAM by buying a larger COAST module, or replace a faulty unit with a good one.

But as cache RAM became faster, it became more and more difficult to establish an electrically clean connection to it. Sockets rely on mechanical friction for a good contact, surface-mount is soldered. So although surface-mount cache could not be upgraded (where socketed cache or COAST slot cache could), in these fast, modern boards it had become a necessity. After PLB arrived, and until surface-mounting came in as the standard practice, about half of all the motherboard problems we'd see in the workshop were cache problems.

With its surface mounted cache, the QDI P51437 was immune to the horrors of the COAST slot, and this alone made it an attractive proposition. We were doubtful of them at first—how could a good mainboard not have proper sockets?—but soon realised that these thin, cheaply made boards were actually more reliable than many of their more expensive competitors.

  • CPU support: 6x86 and K5 to 166, Pentium Classic.
  • Speed: 50, 60, or 66MHz.
  • Slots: 4 PCI, 3 ISA
  • RAM: 4 72-pin FP or EDO, up to 128MB.
  • Cache: 256k asynch.
  • Chipset: Intel Triton 430FX, Award BIOS.
  • Best With: Pentium.
  • Status: Legacy.

PC Chips M919

Of course, there is more than one way to solve a cache problem. Why not fit a COAST socket for appearances, leave it empty (because cache RAM costs money), and surface mount a pair of black plastic rectangles with "write back cache" labels? It would be lots cheaper to manufacture, wouldn't it? Of course, it would not be honest or even legal, but so what?

We met just three of these particular boards as they were on the market well after the majority of buyers had moved on to Pentium and 6x86, but none of the three worked reliably. (Which is why people brought them in for us to fix, of course.) Looking at the thin as tissue paper board, it is no wonder they were buggy. It's a shame really, as they had a nice layout and if PC-Chips had made a decent honest product a dual VESA PCI 5x86 board would have sold very well in the upgrade market at that time.

Before a kind reader sent us some more information, this board was unidentified. At the time we wrote: "Notice that this board is devoid of anything that might identify the manufacturer, but it somehow looks familiar. Six to one it was those champions of the crooked, PC Chips. Again."

Then Jay wrote to confirm our suspicions:

It's a PC-Chips M919, and that cache is quite fake. (The traces go nowhere!) This board was released in several revisions taking various configurations of 30 and 72-pin RAM and they are all equal in quality. They claim to support 50MHz bus speeds but are really hopeless above 33MHz (mine won't run for more than ~15 seconds at 40MHz, nor at all at 50). The cache slot is not a COAST at all - it really takes a special asynchronous module (something like DIPPs on a stick) rather than the pipeline burst cache the Pentium boards took. I am currently using the board with a 5x86/100 to play old DOS games and watch demos. Even under pure DOS, the M919 is prone to EMS errors and random lockups. PC Chips strikes again.

  • CPU support: 486 and 5x86.
  • Speed: 25 to 50MHz claimed.
  • Slots: 1 VESA, 3 PCI, 3 ISA
  • RAM: 4 72-pin.
  • Cache: Fake.
  • Chipset: UMC 888, AMI windows BIOS.
  • Best With: A photographer and a lawyer.
  • Date: July 1996.

QDI Explorer I

To replace the FX chipset, Intel introduced two new ones: the VX was a direct FX replacement for the general market, while the HX was a more specialised high-performance design, and quite expensive.

It is fashionable these days to complain about the old VX chipset, to decry its 'absurd' 128MB RAM limit or the still more limiting design restriction that made the VX unable to cache beyond the first 64MB. Much as we love to sink the slipper into Intel given half an opportunity, this modern view is rubbish — by the standards of the day the VX was fast, reliable, reasonably inexpensive, and capable of doing everything that was asked of it. It was a very large and powerful system indeed that had more than 64MB of RAM in those days — most shipped with 16 or 32MB. (Note that the TX which came out a few years later to replace the VX and which maintained the 128MB limit was a different matter: that was a truly cynical and crippled product which, strangely enough, got far more praise than it deserved.)

The QDI Explorer was one of our first VX-based boards. They were a typical good Pentium board, rather thinner than the FIC or Chaintech equivalents (thick boards are usually more reliable and thin boards always make us nervous), but they proved very stable in practice — largely because, like the QDI FX boards, they were almost immune to cache problems. The Explorer I was better with a 6x86 than most first-generation boards, but still most stable with a Pentium. Notice that this transitional design still has a COAST socket for expansion (it's the long brown connector above the CPU socket) but that there are two cache chips mounted on the board itself (upper right corner).

  • CPU support: 6x86 120, 150 & 166, P54C to 166.
  • Speed: 50, 60, or 66MHz.
  • Slots: 4 PCI, 3 ISA
  • RAM: 4 72-pin FPM or EDO, up to 128MB
  • Cache: Surface mount, 256k pipeline burst.
  • Chipset: Intel Triton 430VX, Award BIOS.
  • Status: Legacy.

QDI Explorer II (jumpered version)

We used these extensively. They came out a little later than the PA-2002 or Chaintech 5IEM and were our favourite for the then-new and tricky Cyrix 686 until the much improved FIC PA-2005 and Chaintech 5IGM arrived.

The build quality of Explorers, initially excellent, declined over time: the early ones were good but the later revisions had smaller heatsinks for the voltage regulators (no doubt to save money) and tended to run rather hot with 3.3 volt CPUs. In consequence, the later revisions were not as reliable, and we eventually stopped buying them.

(Like most boards, the standard Explorer II had the traditional jumpers for CPU selection. The Explorer SpeedEasy, described below, was the jumperless one).

  • CPU support: 6x86 to 166, K5, P54C, P55C.
  • Speed: 50, 60, or 66MHz.
  • Slots: 4 PCI, 3 ISA
  • RAM: 4 72-pin FPM or EDO, up to 128MB, 1 168-pin Synchronous or EDO up to 64Mb
  • Cache: Surface mount, 256k pipeline burst.
  • Chipset: Intel Triton 430VX, Award BIOS.
  • Status: Legacy.

Chaintech 5VGM1

Old Dependable. This VX-based board was our favourite standby for quite some time. It wasn't as neat a design as the PA-2005, and we generally preferred the VIA Apollo chipset to the Intel VX as the Apollo was always better suited to the by then very popular Cyrix and IBM CPUs, but these were really solidly built. They just went: first time, every time. When we were troubleshooting and suspected the motherboard, plugging a 5VGM was one of our first steps.

Notice the two large brushed aluminium regulator heatsinks at lower right and compare them to the flimsy black-painted things on the QDI Explorer II above. You can usually tell the quality mainboard manufacturers from the also-rans, because the best makers are prepared to spend a few cents more on the minor components.

  • CPU support: 6x86 to 166, P54C, P55C.
  • Speed: 50, 55, 60, or 66MHz.
  • Slots: 4 PCI, 3 ISA
  • RAM: 4 72-pin FPM or EDO, up to 128MB.
  • Cache: Surface mount, 256k pipeline burst standard, 512k optional.
  • Chipset: Intel Triton 430VX, Award BIOS.
  • Best With: Pentium.
  • Status: Legacy.
FIC PA-2005 illustration

FIC PA2005 (original version)

This early version of the great PA-2005 (also listed on the following page used the Apollo 580 chipset instead of the 585, and was usually shipped with 256k cache, rather than the 512k that the later versions were blessed with. The two PA-2005 varieties were almost identical except that this original had a maximum bus speed of 66MHz (which was the standard at the time). The later version had a 75MHz bus to suit the Cyrix 6x86-200.

The 2005 proved to be one of the all-time classic main boards: simple, reliable, and incredibly long-lived. We sold them in large quantities, and still keep one in the Red Hill Workshop for RAM testing because, like the 2002, it can run with a single SIMM. If we had to name just one all-time favourite board, this would probably be it. (By the way, the PA-2005 was the board AMD originally selected for benchmark testing their then-new K5 CPU, though we sold the vast majority of ours with IBM 6x86s.)

  • CPU support: 6x86, 6x86MX, K5, K6, P54C, P55C.
  • Speed: 50, 55, 60, and 66MHz.
  • Slots: 4 PCI, 3 ISA
  • RAM: 4 72-pin FPM, EDO or BEDO, up to 256MB. Like the PA-2002, it can operate with just a single 72-pin SIMM if required.
  • Cache: Surface mount, 256k pipeline burst standard, up to 1MB optional.
  • Chipset: VIA Apollo 580VP, Award BIOS.
  • Best With: 6x86-166.
  • Status: Legacy.

FIC PT-2200

Intel's premium HX chipset was traditionally more expensive than all others, and HX boards were always priced to reflect this. The justification for the expense was measurably higher performance. With a Pentium or Pentium MMX, an HX chipset board was simply the best thing you could buy.

However, this only applied to the Intel CPUs. The HX didn't do anything particularly special with an AMD K5 or K6, and was nowhere near capable of getting the best out of the Cyrix and IBM CPUs, turning in benchmark figures that were well below what could be achieved with much more humble but better-suited boards. Given the very high proportion of Cyrix CPUs we sold in those days, HXs were something of a rarity at Red Hill.

Once the similarly expensive but faster and more versatile PA-2007 arrived, the HX started slowly fading away — though with better than TX speed and rock-solid stability we still regarded HX boards as an excellent choice for Intel CPUs long after they were officially obsolete. The 2200 was our most common HX choice simply because we were already buying lots of PA-2005s from the same source, and there didn't seem to be any reason not to get our HX boards there too. If there was ever a great deal of difference between different HX boards, we never noticed it.

  • CPU support: 6x86 to 166, 6x86MX to 200, K5, K6, P54C, P55C to 200MHz.
  • Speed: 50, 55, 60, or 66MHz.
  • Slots: 4 PCI, 4 ISA
  • RAM: 4 72-pin FPM or EDO, up to 512MB
  • Cache: Surface mount, 512k pipeline burst.
  • Chipset: Intel Triton 430HX, Award BIOS.
  • Best With: Pentium, Pentium MMX
  • Status: Legacy.

QDI Explorer II SpeedEasy

A modified Explorer II (see above) and the first jumperless motherboard. The SpeedEasy BIOS auto-detected your CPU and then allowed setup of speed and voltage in software. This was a revolutionary idea — but not one we liked much. It just allowed too much scope for user error. With a traditional jumpered board, the CPU matching was performed by a technician, and once it was set right, it couldn't be changed by accident or error. But with these your children, or an ill-informed friend, could change the speed and voltage settings from the keyboard. This could result in loss of performance or, worse, permanent damage to the CPU.

We trialled these boards for a while, and were impressed with the cleverness of the SpeedEasy technology, but on balance preferred the more accident-proof jumpered boards. Note that at a pinch these could be used with a 686-200 but the Triton chipset wasn't really up to it. Clock doubled at 75MHz they hung and GPFed. Clock tripled at 50MHz they ran reliably, but performance was crippled by the slow bus speed. For the P-200+, you needed a proper 75MHz-certified chipset like the Apollo 585 or SiS 5571. (VLSI and OPTi also produced fast 75MHz chipsets, apparently, but we never saw them in the flesh.)

↑ Illustration: detail from an early SpeedEasy board. Notice the printed instructions for jumper setting the front side bus speed at left and, half obscured behind the capacitor at right, the missing, blanked off jumpers. QDI had the SpeedEasy BIOS working but had not yet got around to changing the screenprint.

  • CPU support: 6x86 to 166, K5, P54C, P55C.
  • Speed: 50, 60, or 66MHz.
  • Slots: 4 PCI, 4 ISA
  • RAM: 4 72-pin FP or EDO, up to 128MB.
  • Cache: 256k 15ns or COAST module.
  • Chipset: Intel Triton 430VX, Award BIOS.
  • Best With: Pentium.
  • Status: Legacy

Gigabyte 586s

An excellent board based on the unusual SiS 5571 single-chip chipset. These were marketed as an entry-level board, but our tests showed that they were comfortably faster than any of the contemporary Intel and VIA based boards—including the much more expensive TX and HX models.

Like the Apollo boards these were certified at 75MHz, but are in practice were not as solid at the higher speed as a PA-2005. As usual, Gigabyte's documentation was only fair, but the boards themselves were almost up to FIC and ASUS reliability standards, and very fast. We had never been big Gigabyte fans here at Red Hill, but we certainly liked the 586S, and sold quite a few of them.

  • CPU support: 6x86, K5, K6, P54C, P55C.
  • Speed: 50 to 75MHz.
  • Slots: 5 PCI, 3 ISA
  • RAM: 4 72-pin FPM or EDO, up to 128MB.
  • Cache: Surface mount, 512k pipeline burst.
  • Chipset: SiS 5571, Award BIOS.
  • Best With: Anything with a 66MHz bus.
  • Status: Legacy.

ASUS VX-97

An excellent board from an outstanding maker, but rather too close to the end of the very successful VX chipset's life. We had originally intended to use these to replace the Chaintech 5VGM as our standard VX board, but even this fine implementation of the aging VX chipset didn't really match the SiS 5571-based Gigabyte board above.

As always with ASUS boards in those days, they were well made and beautifully documented. Performance was good for a VX but 6x86 support was only fair—it was not certified for several of the 166MHz versions and (like all Intel chipset boards) couldn't do 75MHz for the 200 Classic. Notice how neatly the voltage regulator (below the CPU socket) is integrated into the structure of the board itself—typical ASUS attention to detail.

  • CPU support: 6x86 to 166, K5, P54C, P55C.
  • Speed: 50, 55, 60, or 66MHz.
  • Slots: 4 PCI, 4 ISA
  • RAM: 4 72-pin FPM or EDO, up to 256MB, 2 168-pin Synchronous or EDO up to 128Mb
  • Cache: Surface mount, 512k pipeline burst.
  • Chipset: Intel Triton 430VX, Award BIOS.
  • Best With: Pentium, Pentium MMX.
  • Status: Legacy.