The OS/2 Museum recently acquired this mystery 386 board (click on the image to see a high-resolution photo):
This is in theory a killer 386 board: onboard Am386DX-40, a socket for a replacement 386 or 486DLC processor, a FPU socket, 256KB cache, 8 SIMM slots for up to 32MB RAM, six 16-bit ISA slots, and best of all, a clock chip that can be set via jumpers to 16/20/25/33/40/50 MHz. Anyone familiar with typical 386 systems knows that such boards with user-selectable clock frequency are rather uncommon. Changing the clock speed normally involves replacing the crystal, which is hardly something one would want to do regularly.
So where’s the catch? So far I have been unable to get a Cyrix or Texas Instruments 486DLC processor working in the board. Both processors work in a different board (older, without proper 486DLC support) but when inserted in this one, the system does not boot up at all and shows no signs of life.
When another Am386DX is plugged into the CPU socket, the system works and both processors get warm. It’s unclear which one is actually used (or possibly both?).
Near the soldered-on PQFP CPU, jumper J1 location is marked, but the jumper is not installed. Without a manual, it is unclear whether the board really supports a socketed CPU, and if so, whether any jumpers need to be set. The function of J1 is not obvious.
My attempts to identify the board via TH99 have not been successful. The board is Made in Taiwan and there may be a manufacturer’s logo (three triangles). There is no obvious model designation but on the bottom side, the string “386IS-B-6-2” is etched. On the component side, a large “REV : 6.2” marking is visible. The PCB was manufactured in week 8 of 1993.
The chipset is an SiS 85C460; this chipset was mostly used in 486 boards but no details (let alone datasheets) are available. The FPU socket has space for a soldered-on 87DLC math chip, but none is present. However, plugging in a 387 FPU does work.
The board has a socket where a Dallas RTC probably could be installed. Instead, a Winbond W85C168 RTC is plugged in, in conjunction with a dreaded rechargeable NiCd barrel battery near the keyboard connector.
If anyone has information about what this board is, or even better, how to get a Cyrix or TI 486DLC processor going in it, please leave a comment.
Update: I forgot to add the AMI BIOS ID string: 40-0000-001594-00101111-111192-SIS3486-F. Unfortunately the string is quite unhelpful. The manufacturer ID is Mitac, but the only similar BIOS string I found is for a Flexus board. In either case, there’s no additional information about the board.
Near the SiS (lame) chip, I see 3 triangles arranged Sierpinski (or Triforce, for the gamers) style, and a ‘WETK3167’ marking – possibly Weitek?
Yes. The FPU socket should support both a regular 387DX (tested, works) and a Weitek 3167 FPU (not tested, never even seen one of those).
I’ve seen boards like this back in the day. The idea being that AMD was going to release faster and faster 386 processors (except they didn’t). I recall another variation like this but with a 486 socket.
For some reason I want to say they were cheap, because of the surface mount CPU, as back then we didn’t like the idea of processors being so married to the board. For the CAD people though a 486 was the only way to go, as a 386+387 was too slow… and the 386 wanted a synchronous 387, I don’t even recall if AMD even made a 40Mhz 387?? I do remember from the launch of the first 386 boards (before the Compaq Deskpro) they supported 80287’s as the 387 wasn’t ready… Or from what I vaguely remember while looking up DOS Extenders a long while back.
But yes, I remember these, and that a 486SX-16 was slower than these, but every 16 I ever saw could easily clock to 20 (and beyond, I ran mine a 33Mhz).
I’m surprised that battery hasn’t leaked yet, it looks to be in pretty good shape still.
Is there a bios ID string on the bottom of the screen when it POSTs? That usually indicates some information including manufacturer.
No idea about the board, but to check which of the CPUs (or both) is used, couldn’t you pull some signal of one of them low (or high) while running and see how it behaves? Check that the signals are not shorted between the CPUs first. An address line would be most fun because you could test it without crashing the system, but the bus is probably shared between them.
Jason, did AMD even make 387’s? I’ve never seen an AMD-branded coprocessor. Cyrix, IIT, ULSI, & CHIPS were the choices for 40 MHz. If memory serves, the Cyrix FasMath was the champ. I also seem to recall it being the most common. I rarely saw the other ones. I remember those 287+386 combos. There was also the 287XL which came out when the 387 was finally released that was based on the 387 and not the 287 (and was significantly faster).
I’d be curious if you could get that board operating at 50 MHz stable with cache enabled. That’d be a killer system for a 386. Only way one could really improve it is with EISA or VLB, which is very obscure in 386 land.
Best I can find on the 460/461 chipset is this: http://fuynfactory.dyndns.tv/wiki/index.php/Chipsets_SiS_para_80486#SiS_461_.2885C461.29
Which tells you basically nothing other than it was a 486-targeted SiS chipset that supported 386’s. Judging by motherboards that used it, it apparently also supported VL Bus.
Actually there’s an old newsgroup post from 1994 out there of someone trying to find a datasheet for this chipset then. Apparently they didn’t succeed.
Actually, the battery is a bit weird. I’m almost certain the battery had been replaced at some point, because a) it looks newer than the rest of the board, and b) there are minor signs of leakage on the board but not on the battery itself. If you look at the photo carefully, there are hints of green copper oxide near the keyboard connector.
As for the BIOS ID string, I’ve updated the post.
I don’t think AMD ever made 387s. They did make 287s, but only briefly (Intel was against). The 386+287 was just desperation because Intel didn’t have the 387 ready in time. The performance was very unimpressive. I would imagine that even if AMD considered making Intel-compatible 387s, they had to consider the existing Cyrix etc. alternatives. They’d probably have a hard time selling a 387 clone.
The 287XL showed up circa 1990, quite a bit later than the 387. But yes, it was based on the 387 and a lot faster. I have a 287XL and it’s at least 50-60% faster than a 287 would be. So effectively it was possible to have: 286+287, 286+387 (aka 287XL), 386+287, or 386+387. Intel always likes to simplify software developers’ lives 🙂
I’ll see how well the board works at 50MHz; so far I just saw that it gets past POST while figuring out what the clock jumpers do. The chipset was presumably capable of driving a 486DX-50, although how well the 386 board is designed for 50MHz operation is another question.
If you know which signal and how to do it safely without the risk of accidentally shorting something, I’m interested 🙂 I’ll have to sit down with an ohmmeter and see if I can figure out how the things are hooked up. At first glance it seems that the signals from the onboard CPU and the socket are simply connected together.
says the manufacturer code:
(001594) is “Trigon” which jives with the triangular logo on the motherboard.
Okay… so it’s not one totally obscure manufacturer but a different one. I’m afraid that doesn’t really help me much 🙁
Yes, the manufacturer ID and the logo suggest the board was made by “Trigon”, but that’s as unhelpful as “Flexus” who probably made more or less the same board. The ESCOM connection is very interesting though, and it is plausible that this board had been used in a PC sold by ESCOM.
Good job figuring out the ID number though 🙂
Sometimes boards came with option that no one used. I had a 486SX that had an OPTi slot per the motherboard manual, which looked like an EISA slot (or VLB), but I can’t recall for sure. I just remember trying to contact the company who had address in L.A., and they said to not worry about it as there were no cards for the OPTi slot.
I had a motherboard similar to this but with ALi chipset. It had a 386 PGA socket and a DX-40 soldered in. The behaviour was the same. With a 486DLC inserted the system would not power on. However, one day I decided to carefully slice off the surface mount 386 and the PGA socket was then usable. I’ve often seen boards with solder pads for both PQFP and PGA but usually only one is actually in place. Perhaps they just got lazy and forgot to leave the PGA socket off.
I’d much prefer not to make irreversible changes to the board… but on balance, I’d rather have the PGA socket than just another boring Am386DX soldered on.
I have the same motherboard but without integrated AM386DX-40 CPU.
My board does not have a J1 Jumper either.
I tested it with different jumper settings.
I tested a Cyrix and TI 486DLC CPU at 33 MHz and an AM386DX-40 and additionally a 387 Coprocessor at 25 and 33 MHz. Is it also possible to run the 386 and 387 asynchronously?
What I can say so far:
J16 closed: very slow! Did not boot, Error with the keyboard
J14+J15+J16 closed: 16 MHz
J14+J15 closed: 20 MHz
J14+J16 closed: 25 MHz
J15+J16 closed: 33 MHz
J14 closed: 40 MHz
J15 closed: 50 MHz
Unfortunately I was not able to find the manual of the board either.
Yes, those jumper settings sound right, I remember a 16-50 MHz range with some combinations not working.
I honestly don’t remember if the 387 can run asynchronously, only that the 287 can.
Just to give a tiny bit of background information :
ESCOM was in the 90s an German PC manufacturer which had some kind of a a “battle” with VOBIS – both tried to be the number 1 at that time. Both were low-price manufacturers, and it was VOBIS who even bundled PCs with OS/2, not with Microsoft Windows 95 – for some time, which tells a bit about the market power the company once had in the German market.
There do exist Wikipedia articles about both – with the German-language ones being a bit longer.
ESCOM had been bought by COMTECH, but both are out of business now.
The COMTECH brand name is still used, however.
Actually my first PC was from Vobis, and I’m quite familiar with ESCOM machines as well. And I know that OS/2 was quite popular in Germany, and especially Vobis offered PCs with OS/2 preloaded. Vobis also preloaded DR-DOS on some of their systems at one time.
Microsoft was quite unhappy about that and put a lot of pressure on Vobis to force them to offer only MS-DOS/Windows.
The Vobis/ESCOM hardware was generally cheap but decent. I believe both Vobis and ESCOM used a lot of “no-name” boards, probably manufactured for them somewhere in Taiwan but based on proven designs.
Pingback: The Nearly Ultimate 386 Board | OS/2 Museum
I have just come about this board myself and i’m wondering regarding the CPU issue that if maybe the QFP is removed, the PGA might then work with all compatible CPU’s.
Did you happen to give this a try? I would be keen to hear your results if you have.
Thanks, and regards.
The QFP does not need to be physically removed. What you need is to activate the QFP processor’s FLT# pin which should be pin 28. That tri-states all of the onboard CPU’s outputs and it is “not there”. There may be an unpopulated header for a jumper on your board which is connected to the FLT# pin.
The short answer is that yes, I got the PGA socket working.
I’ve had a similar board, but without the PGA socket.
That board had some interesting bugs related to RAM parity check.
In Bios you could turn parity check off, but you couldn’t enter setup without having memory with parity… But that silly requirement only applied to the first bank. In practice you could have four simms with parity in the first bank and four simms without parity in the second bank. If you had that configuration you wouldn’t see anything strange running DOS, Windows 3.x e.t.c. but Linux would on boot say that it recieved one NMI and say “Dazed and confused, but trying to continue” and everything worked fine.
Maybe that configuration didn’t work with Windows 95 and/or OS/2 Warp, i don’t remember and I don’t have the card in working order any more.
The board is not exactly bug-free. For example it is unable to boot when the CPU frequency is set low, and the frequency setting itself doesn’t work as it should. It works reasonably well with default settings at 25/33/40 MHz.