Yeah, I'm really unhappy about the soldering, I should've run more tests but it was already day 2 of doing just that and I was getting nowhere. It's one thing when it's necessary to solder stuff, and another when it does nothing and I make a so-so job of it... Oh well. It works - as well as it did, anyway.
So I didn't touch RAMs yet, although I have to say the entire bank is NOS chips and I did test them before I populated the mobo (but that was months ago). Today I went after the HIMEM issues and I found that even enabling A20 gate with my own code I get these problems. So here's some more observations:
- A20 gate is controlled by KBC and the only chip that seems to receive that signal is the ST62BC002 / HADR, it controls how CPU address lines get passed to memory (depending on switch configuration) and to ISA slots.
- When A20 gate is disabled, everything is OK. Enabling it, just flipping that one KBC output port bit, starts the problems with 16-bit VGAs.
- Because the mobo reboots with A20 enabled (BIOS disables it when booting OS) I get the screen corruption even during memory check.
- It really helps if the VGA card is installed in the slot next to KBC and '245 chip, I get almost perfect operation, but there are still some randomly corrupted spots, just way more rarely than in other slots.
- MR BIOS solved my CF card problems, so I guess that was unrelated and down to this particular card. It's a great BIOS too, nice look, fast booting.
- For some reason I'm unable to run NSSI while using MR BIOS, it hangs on bus detection even with /SAFE switch, this is not happening with original AMI BIOS. Same with Snooper.
- Also, 1024x768 modes on Trident 8900C VGA hang with MR BIOS, but work fine with AMI. Other modes, including SVGA, seem to work fine on both.
Frankly I could just roll with MR BIOS and use Trident in the lowest slot, that would almost eliminate all the video and CF card problems, if it wasn't for the fact that NSSI and Snooper hang. I could just not use them but I don't like that. I do not get why A20 gate is affecting ISA in that way - is my '002 chip busted or is it yet another thing with this mobo and video cards. It does not boot at all with Trident 8900D, I suspect the ALE signal is toggling during memory refresh, it's a known issue with some mobos.
Also, somewhat related, XTIDE ROM doesn't really solve the CF card issues I had, but MR BIOS did. Maybe if I used the bigger XTIDE version (over 8KiB)...
EDIT: To clarify, this issue might not be a recent development, it's possible the mobo was always like that but until now I did mostly run benchmarks with no HIMEM (so A20 gate disabled) and/or with the Trident 8900C in the lowest slot so I just didn't notice any problems. I was also using a different CF card, that I move around for testing, now I've swtiched to a card I was hoping to use with this system.