Looking at what could be wrong with the keyboard interface and started taking a closer look around the port.

There's what appears to be a typical keyboard / PS2 fuse near the AT keyboard port (FS1), which seems to be directly connected to the solder point I have highlighted (at least when measured with resistance I get a measly 0.1 ohm between those two points. Interestingly, you can also see there are solder points for a standard PS/2 keyboard header below the AT keyboard port....

However if I measure resistance across the fuse, my multimeter (admittedly an ancient Fluke 73 of about 30 years old) instead of giving me either a steady/fixed resistance reading or an open-loop result, starts measuring at 10M Ohms and increasing until it goes off-scale at 30M Ohms. I do wonder if this fuse has failed in some way that it has not gone open-circuit, but is otherwise limiting current to the keyboard port, leading to the weird behaviour I have observed (the key repeat issues and the Dell keyboard not even powering on).

Simplest test will be to bypass the fuse with a wire link, I guess.

I wouldn't have thought a fuse could do this, but.....

The Dell PS/2 keyboard now powers up!

Even better...

https://youtube.com/shorts/9GHAhATI0YU

No more dropped keys when typing fast! And key repeat is working as you would expect it (hold a key and it repeats indefinitely, not the one-finger-one-keypress-one-character mode it was in).

What type of failure is that from a fuse? I've never seen anything like it - my experience is that fuses are either failed... or they're not. This one was clearly in some failure state where it was passing electrical current (remember I measure +4.94 on the AT keyboard pin), but had incredibly high resistance... that's just bizarre.

I've just cut the fuse and replaced it with a wire jumper for now, but I'll aim to get a replacement fuse fitted asap.

I think this means that short of fitting the quieter PSU cooling fan (it's a noisy as hell 40x40x10mm part!), this little thing is ready to be put through its paces.

megatron-uk wrote on 2023-01-07, 10:21:

starts measuring at 10M Ohms and increasing until it goes off-scale at 30M Ohms.

Capacitative charging? Is that fuse on the Ground side? Might explain why there was still 5V on the pin.

I think it is the keyboard ground line, yes.

So, the keyboard has a good +5 supply, but no good Ground? And I think clock and data lines both have pull-ups on them. So the only time the keyboard would see a ground on any pin would be when the motherboard tried to send a status byte to the keyboard and pulled clock and data low. Doesn't sound like that'd work well enough to send any bytes. Maybe someone's used a resistor//capacitor to separate shield ground and keyboard/digital ground at low frequencies (people do strange things with shields), and the keyboard is using the shield as its digital ground?

Completed a full suite of benchmarks with this thing:

Looking at Phils VGA card benchmark spreadsheet, the scores are about par for the course for the WInchip C6 200; around the same level of performance as a non-MMX Pentium 133-166. So performance-wise these are not a great chip. The chipset/motherboard doesn't appear to be making things any worse than the C6 alone would perform though.

One disappointment is the disk performance. 1.8MBytes/sec is a fairly poor showing and the level of performance I would expect if the XT-IDE bios wasn't being shadowed... but it is. I don't know much about the ALI chipset and whether the IDE controller is connected to the PCI bus, or on ISA. If the latter, then it may benefit from the stock ISA clock being increased (left at the bios default of 7.14MHz now, since I only have a sound card in an ISA slot).

It will be interesting to see if things get any more sprightly with a P200 and then a P200 MMX.... if the cooling and power supply of this thing hold up.

That does seem a tad slow. I get 1800 kB/s on my 486 with VLB I/O card.

drosse1meyer wrote on 2023-01-09, 19:20:

That does seem a tad slow. I get 1800 kB/s on my 486 with VLB I/O card.

Tell me about it; on my FIC-386SC (with 486 DLC-40) I can hit 3Mbytes/sec with the same type of card (Sandisk 16GB Ultra), CF reader and a bog standard ISA multi-io card (that's with the XT-IDE rom shadowed). On a 286 with a basic 8GB Transcend CF I can just about get 1.5Mbytes/sec.

As a workaround for now I've installed xhdd.sys and given it a 5MB disk cache which has perked things up a bit.

My current work on this thing is involving ripping out the tiny internal custom PSU and its horrible 40mm fan and replacing it all with a silent 90w PicoPSU.

megatron-uk wrote on 2023-01-09, 20:39:

Tell me about it; on my FIC-386SC (with 486 DLC-40) I can hit 3Mbytes/sec with the same type of card (Sandisk 16GB Ultra), CF re […]
Show full quote

drosse1meyer wrote on 2023-01-09, 19:20:

That does seem a tad slow. I get 1800 kB/s on my 486 with VLB I/O card.

Tell me about it; on my FIC-386SC (with 486 DLC-40) I can hit 3Mbytes/sec with the same type of card (Sandisk 16GB Ultra), CF reader and a bog standard ISA multi-io card (that's with the XT-IDE rom shadowed). On a 286 with a basic 8GB Transcend CF I can just about get 1.5Mbytes/sec.

As a workaround for now I've installed xhdd.sys and given it a 5MB disk cache which has perked things up a bit.

My current work on this thing is involving ripping out the tiny internal custom PSU and its horrible 40mm fan and replacing it all with a silent 90w PicoPSU.

damn now you're starting to make me think my 486 drive is slow. which is possible, as i could never the the DOS drivers to properly work.

It's all about ISA clock (I can damn near run it at 14MHz stably on that system) and ensuring that you copy the XT-IDE rom to ram - if you're using an IDE rom replacement of course (otherwise any bios based disk access has to go over the ISA bus to access the routines in the ROM chip)!
No such problems if you have a drive overlay like Ontrack, since it's all in ram anyway.

PicoPSU is fitted and working. I re-used the original AT power switch wired to the PS_ON/+5VSB headers on the ATX breakout board I fitted the PicoPSU to (easier than cutting up an ATX power cable/crimping a load of wires).

It's gotten rid of that nasty little 40mm PSU fan, quietened the overall noise substantially (only the new fan on the heatpipe/heatsink, which I may try to run with a voltage reducer - it's constant +12v by default), as well as reduced the weight of the case (it's going on a bookshelf above my desk, so light == good!). I don't have any mechanical drives, so the 90w psu is more than adequate.

megatron-uk wrote on 2023-01-10, 08:16:

It's all about ISA clock (I can damn near run it at 14MHz stably on that system) and ensuring that you copy the XT-IDE rom to ram - if you're using an IDE rom replacement of course (otherwise any bios based disk access has to go over the ISA bus to access the routines in the ROM chip)!
No such problems if you have a drive overlay like Ontrack, since it's all in ram anyway.

Surely the instruction fetch times on accessing routines from ROM are totally dwarfed by the 'mechanical' delays caused by the actual reading/writing of the device? Or if not that, the transfer of the data?

Think about ROM access speeds (=>100ns), across the ISA bus (~8MHz, 8/16bit), compared to main memory (<100ns, 16/32bit, whatever transfer speed your memory is at).

I've done quite a bit of testing, including the benchmarks here for my FIC 386-SC:

https://www.target-earth.net/wiki/doku.php?id … 86_shootout_fic

Those results are for 10MHz and 13MHz ISA bus speeds, respectively. Enabling shadowing on the XT-IDE rom brings substantial speed ups.

megatron-uk wrote on 2023-01-09, 20:39:

Tell me about it; on my FIC-386SC (with 486 DLC-40) I can hit 3Mbytes/sec with the same type of card (Sandisk 16GB Ultra), CF re […]
Show full quote

drosse1meyer wrote on 2023-01-09, 19:20:

That does seem a tad slow. I get 1800 kB/s on my 486 with VLB I/O card.

Tell me about it; on my FIC-386SC (with 486 DLC-40) I can hit 3Mbytes/sec with the same type of card (Sandisk 16GB Ultra), CF reader and a bog standard ISA multi-io card (that's with the XT-IDE rom shadowed). On a 286 with a basic 8GB Transcend CF I can just about get 1.5Mbytes/sec.

As a workaround for now I've installed xhdd.sys and given it a 5MB disk cache which has perked things up a bit.

My current work on this thing is involving ripping out the tiny internal custom PSU and its horrible 40mm fan and replacing it all with a silent 90w PicoPSU.

Circling back to this - I took another look at the Holtek VLB IO card's rather paltry doc PDF, and noticed there's a set of jumpers for "Speed Grade Initialization" - apparently the default is 0, so just to see what happens, i set it to 3. Now I'm gettin about 2900 kb/s, so a bit faster, but probably not as fast as I should be getting on VLB 😀

Trying to now get a P200 MMX working in this system. It doesn't specifically list MMX support, but it has onboard voltage regulators as well as listed support for the Cyrix 6x86L which has the same split voltage specifications (2.8v / 3.3v) as the P55C.

The motherboard manual lists this:

I've got the CPU type jumpers set to the Intel processor values, have enabled dual power mode and set core voltage to be 2.8... and I get no life out of the machine at all; no error codes, no video, nothing.

Presuming that a motherboard has the correct socket type for an MMX, and the correct voltage specifications... is there anything else that would be needed to configure it differently to a standard non-MMX processor of the same speed?

Hmm... think I may have killed it!

Took the processor back out and started measuring voltages on the socket. With 2.8v / dual voltage settings jumpered, I can indeed get 2.8v on VCC2 pins.... but wherever I measure VCC3, it only shows 0.9v.

Jumpered back for the Winchip (single voltage, 3.52v) both VCC2 and VCC3 are back to both measuring 3.53something volts. But, the system still won't power up with the Winchip back in its socket (which is not a ZIF or a LIF, it's some weird slide-the-processor-in-slightly sideways design, and it's the two screws of the heatsink/heatpipe that physically hold the processor in place).

I am wondering if jumpering for dual voltage and powering it like that has pushed some ~25 year old component over the edge.

Okay, it is working; both in single and split-voltage modes - the issue is the processor socket itself.

What I've realised is that because of the space constraints of this motherboard and chassis, what they've done is use the base part of a ZIF socket, and then altered it so that the upper section that normal moves to lock the processor in via the lacking arm is no longer part of the socket - just the bare 'U shaped' receiving pins that are normally hidden from view. You instead have to drop the processor into the holes (which are pretty much free-floating - there's almost no contact between cpu pins and the receptacles in the socket) and then insert a small flat-bladed screwdriver into a special wedge-shaped cut-out at the side of the socket, where the heatsink retention clip would normally be.

That pushes the processor sideways into the receiving pins, just as the usual locking arm would. It feels a bit disconcerting to do it by hand, but it works.

Anyway.... the original Winchip 200 works as normal... and the P200 MMX....

Running through benchmarks and it does seem quite a bit quicker (almost an extra 20fps in Doom high detail for ~78fps, and just under 20fps more in Quake 320x200 for ~48fps)...

But.... I'm also getting a couple of random resets/power-offs which didn't happen previously, as well as the CF card not getting detected in between reboots. I think the split-voltage mode and the extra draw of the P200MMX may be just on the limit of what the tiny PicoPSU can deal with, as if I watch the ATX breakout board closely, the power indicator on it flickers just before I get a reboot.

I've got 2x 32MB EDO modules, a 10/100 PCI NIC, 4MB PCI Virge and an ISA ESS1868F in this system now - so it might be worth pulling the cards out to see if it settles down, if so than a bigger PSU would probably resolve it.

what is the pico psu rated at? i can't image a P1 based PC taking more than like 60-70 W total, depending on the number of components

I believe it's either an 80w or 90w rating, of course that's likely to be a peak rating, and who knows whether the 12v power brick feeding it is still up to the task.

It's easy enough test, as I'll just temporarily plug a full ATX PSU into the header and see if the reboots/brownouts go away.

The limiting factor is that PicoPSU is meant for later hardware that puts most of the CPU power source to 12 volts. With 5 volts it could hit the limit rather quickly and especially with recently made ones or the cheap chinese clones, since 5v isn't a priority with modern hardware.

BTW, the CPU socket you've described isn't unique to this but it's mostly found on Pentium laptops where the CPU could be replaced.