Well, this was not really necessary, but I felt the itch and had to try it.

I was testing some more cards today and came across this super generic and cheap looking "UNION" S3 Virge DX card. When I ran WinTune97 it was clear that this card was abysmally slow for a VirgeDX. It scored 15MP, compared to 35MP for a Diamond Stealth 3D 2000 Pro which runs at 75Mhz and is basically the fastest configuration of a Virge DX.

I had my suspicion that this was caused by the card being configured for 2-cycle EDO mode, which utterly destroys performance vs 1-cycle but increases compatibility with slower memory chips.

I ran Astra32 to get some information about the card and it reported not only that the clock speed was 50Mhz (33% slower than the Diamond), but that it was running 2MB of "FPM"?? I don't know exactly why it reported that, but it reports 1-cycle EDO on other cards, so I figured at the very least this card was not configured for 1-cycle.

After the experience gained from this thread, I figured it was fixable. The chips on this card are Alliance 50ns EDO, which based on Trio64V+ testing should be okay for 1-cycle mode, at least at 50Mhz. It seems the VirgeDX uses the same basic pinout as the Trio64V+, so I was able to find the same pins that toggle 1-cycle or 2-cycle.

Low and behold, they hid them under the BIOS chip, and the resistors were indeed missing! They must have designed this PCB to be stable with far worse memory than this particular card ended up with, so they had to leave it gimped in 2-cycle mode.

So, I did what any "normal" person would do... I found a couple of spare 10k resistors, used my hot air gun and a bit of solder paste and turned this into a 1-cycle card.

Now, it scores around 27MP in the same test, up from 15! Astra32 now reports that it is using EDO in 1-cycle mode too, so it was indeed successful.

... it's still a junky model of a lame card, but, hey it's significantly faster now. I tried clocking it a tad higher, just to 60Mhz and it started artifacting, so maybe this memory is right on the edge of being good enough for 1-cycle. Either way, I'm happy to have made this thing less awful. 🤣

I'm impressed this kind of mod even exist 😀

Pushing the Tualatin even further. And it wouldn't have been a coincidence that I had to come across @VooDooMan's thread with the RAM binning.

Great, now I have to add another variable. As if 186FSB wasn't hard enough 🤣

GigAHerZ wrote on 2026-01-20, 10:35:

IIRC, if you connect certain pins on LPT parallel port on your Toshiba laptop, it will erase the bios password. 😉

I already did a while ago, and I just set up a drive in it with DOS and Windows for Workgroups 3.11

I have bought a bi stable switch for an adapter: AT 2 ATX for my Asus P55T2P4.
Hope it will work.

I added a feature to my disassembler to import symbols from PDB files, based on the spec from Undocumented Windows 2000 Secrets. It works on some old Win 2000 PDB files from ~2005. Then I tried to find more PDB files online, but they are a bit hard to come by. On one page with 99% broken links I managed to get some PDB files for XP SP2. There is also this half-baked program PDBdownloader which needs .bloat framework and likes to fail silently without downloading anything. In any case, it turns out that these PDB files are a different, incompatible format, although it looks similar to the old one so I can probably reverse it.

AlphaWing wrote on 2014-07-12, 17:00:

I have bad luck with win9x and P4's. Never had a single P4 board that would run 9x stable or without some weird issue. That will […]
Show full quote

I have bad luck with win9x and P4's.
Never had a single P4 board that would run 9x stable or without some weird issue.
That willy I had to put 2000 on because it would just randomly hardlock no matter what I tried in 98,98se and ME.
Runs 2000 rock solid tho 😐
An I850 Rdram based board I have is exactly the same way.

There was an unofficial CPU Microcode patch released for windows Millennium and probably windows 98SE.

I have heard CPU drivers from Windows XP 32bit improve stability in windows 9x .

A memory patch should be applied to any 9X system with more than 512MB RAM.

On Friday (two days ago), I purchased all four of brand new Socket 3 Anchor Electronics (Santa Clara, CA; located a block from NVIDIA HQ) had at the store.
In my prior trip about 2 weeks ago, I bought three Socket 3 and six VL-Bus connectors they had.
They did replenish the VL-Bus connectors on the rotating display stand.
There were two Socket 3 back on the rotating display stand.
I immediately took two of them and asked the store attendant if they had more.
There were two more in the back, so I took them as well.
Now I have a grand total of seven Socket 3 on hand.
The Socket 3 itself cost $4.95.
Not a bad price considering the scarcity of Socket 3.
In case someone wants them, they have several inventory of blue colored Socket 1 (it says "Over Drive Ready") for $7.95.
They also have inventory of non-ZIF PGA sockets like 132, 168 (or 169), and 273 as well.

I have off day from work today so I decided to do some cleaning and maintenance for this Lian Li aluminium case. I think it is PC7A II plus, but the basic design of these remained the same from early 2000s with slight modifications, I think PC60 was the original case. I've had this case for a while now, but I haven't done anything with it so far, but I already have a Socket A build being planned, or possibly P4 Northwood and Intel 850 combo.

Damn these are fine looking cases, perhaps the finest of all time. The downside of course is that these get scratches if you look them wrong, but gladly this unit has been stored pretty well and there is nothing major.

I also managed to find the CD/DVD aluminium cover plate separately these cases, so I don't even need to use silver drive. I have now beige samsung IDE DVD drive behind that plate.