First post, by nztdm
So, I've collected a bunch of old parts, and am in the process of finishing up a DOS+Win98 gaming PC build.
A-bit BE6-II rev1.0 (BIOS 70)
PCI Fast Ethernet
PCI USB2.0 (ALi)
MSI MS-6905 Master 2.3 Slotket
With a P-III Coppermine (667/133/256), all is well.
With a P-III-S Tualatin (1400/133/512), all is well only if In-order Queue Depth is set to 1.
The only other IOQD setting is 8, and that causes minor instability. I have found a way to reproduce the freezes. The "vertex shader" benchmark in 3DMark 2001 SE will always freeze within seconds at IOQD=8, but will run forever with no issues at IOQD=1.
Things that make no difference:
Changing down to 100MHz FSB
3-3-3 vs 2-2-2
Cache latency maximum
3x128MB Infineon PC133 CL3 vs 2x256MB Kingmax PC150
The modification I performed for Tualatin support is removing AN3, AJ3, AK4, and connecting AK4 to AK26 with a 1K resistor.
Perhaps one of the other two places to connect AK4 will make a difference: http://www.oocities.org/_lunchbox/tualeron_slot1_bx_mod.html
Update: seems to be stable with 1x512MB Kingston PC133 CL3 (only sees 256MB, as usual for BX). I have two more of these sticks I could try (even if 3 work, I think 2 is plenty).
Would still be nice to get to the bottom of this. Maybe I just have some bad RAM, maybe the system is more stable with a single RAM module, maybe my slots are crusty, and need some contact cleaner.
Two PCI slots were damaged when I got the board. Took me ages to find (like really, I re-capped the whole board [these boards are notorious], and then noticed bent pins inside two PCI slots, shorting out). These pins get bent when someone pulls out a universal PCI card on an angle; the edges of the card catch on the pins.