No need for a Glide app that only uses the FBI. There's an environment variable for this.
Add the following line to C:\autoexec.bat and reboot (can also be added to the registry):

1SET SSTV2_TEXMAP_DISABLE=1

Now, based on my experience, that behavior (freezing + screen patterns/artifacts that are not related to textures) is usually indicative of an FBI chip issue (or its memory ICs).

In that particular case, there's another environment variable that can work, which disables half of the FBI memory (leaving you with only 2 MB):

1SET SSTV2_FBI_MEMSIZE=2

If you're lucky, the damaged component was part of the disabled 2 MB of memory. 😀
I actually have a card that I've temporarily "fixed" by physically removing those 2 MB of memory ICs 😁 When I get a chance, I will replace them with new ones.

I have implemented both of those commands in autoexec.bat and rebooted. Booted Win95 and ran GLQuake.

SET SSTV2_TEXMAP_DISABLE=1
You will get an error when you try to view the Voodoo2's advanced tab options.
Still blank screen in GLQuake.

If I then add, SET SSTV2_FBI_MEMSIZE=2, but now I get:

Could not initialize GL (wglCreateContext failed).
Make sure you are in 65536 color mode, and try running -window.

The environment variables aren't really meant to be used at the same time.

My advice is to use the first one (which disables the TMUs completely). If you still get a freeze (most likely)... then you should look at the FBI and its memory.

In that case, you can then disable 2MB out of the 4 available to the FBI. If you still get a freeze like this, then the issue is with the first two memory pairs (or, in the worst case scenario, the FBI connections which can be external or... internal).

Once you disable 2 MB of FBI memory, you are only able to run the games at 640 x 480 x 16. For some it will work only with double buffering, for others triple buffering will also work.
So, make sure that GLQuake is set to run at 640 x 480 x 16.

Since this might help someone, here's how a FULLY WORKING Voodoo 2 card with 2 MB of FBI memory looks like.
The behavior when removing those memory IC pairs is identical to setting the "SSTV2_FBI_MEMSIZE=2" environment variable. So if your card works with this environment variable and doesn't work without it (i.e.: you get freezes or "2D" artifacts that aren't part of the texturing), you need to focus on the four memory ICs which are missing from my pictures (a good starting point is checking the continuity from the FBI pins back to the memory IC pins).

If, on the other hand, you still get freezes with 2 MB of FBI memory, then you need to focus on the four FBI memory ICs which are still present in my pictures.

Keep in mind that, in some rare cases, the culprit is the actual FBI chip, in which case (if a reflow doesn't work), your only hope is replacing it with an FBI from a donor board (don't think these chips still exist NOS anywhere).

bloodem - thank you for this information. With SET SSTV2_FBI_MEMSIZE=2 as the only 3dfx related setting in autoexec.bat, I am able to run GLQuake in 640x480x16. Therefore, one or more of the vacant memory modules shown in your image must be bad on my card. I will go through them one-by-one to replace them. I hope I have some in a bin as I was hoping to fix the Diamond card next and don't want to pillage its memory.

feipoa wrote on 2022-11-10, 09:12:

bloodem - thank you for this information. With SET SSTV2_FBI_MEMSIZE=2 as the only 3dfx related setting in autoexec.bat, I am able to run GLQuake in 640x480x16. Therefore, one or more of the vacant memory modules shown in your image must be bad on my card. I will go through them one-by-one to replace them. I hope I have some in a bin as I was hoping to fix the Diamond card next and don't want to pillage its memory.

You're welcome! Great to hear that you're on the right path! 😀

It looks as if the layout of our two cards are different.

On your card, you have identified U18/U20 on the top and U20/U22 on the bottom for the chips to eliminate in 2 MB frame buffer mode. On my card, the silkscreen shows,

U4/U5/U6/U7 on front and U19/U20/U21/U22 on back. Are the corresponding positions for my card, U5/U7 top and U20/U22 bottom?

I was also wondering if these cards need 25 ns memory at their stock frequency? My card uses EliteMT M11B416256A-25J, which is 25ns/100mhz, EDO 256kx16. I don't have any extra of these IC's, but I have 5 pieces of ISSI IS41C16256-35K, which is only 35 ns. For 25 ns, I only have one piece of NOS V53C16258HK25. Both are EDO 256kx16.

feipoa wrote on 2022-11-10, 10:36:

It looks as if the layout of our two cards are different.

On your card, you have identified U18/U20 on the top and U20/U22 on the bottom for the chips to eliminate in 2 MB frame buffer mode. On my card, the silkscreen shows,

U4/U5/U6/U7 on front and U19/U20/U21/U22 on back. Are the corresponding positions for my card, U5/U7 top and U20/U22 bottom?

Yeah this is a bummer, but not the end of the world.
I've actually identified them with an infrared thermometer. With 2 MB of FBI memory disabled, I left the card for ~ one hour to loop the Unreal flyby intro, and then checked the temps of each memory IC in 3 specific regions (top / middle / bottom). The average temps for the unused memory chips were noticeably lower than the active ones (~ 3 - 5 degrees C, if I remember correctly - I did these tests about 2 years ago).

feipoa wrote on 2022-11-10, 11:21:

I was also wondering if these cards need 25 ns memory at their stock frequency? My card uses EliteMT M11B416256A-25J, which is 25ns/100mhz, EDO 256kx16. I don't have any extra of these IC's, but I have 5 pieces of ISSI IS41C16256-35K, which is only 35 ns. For 25 ns, I only have one piece of NOS V53C16258HK25. Both are EDO 256kx16.

I would say that 35ns could be an issue. IIRC, this is equivalent to ~ 83 MHz for EDO memory, so I would imagine that it's a bit risky to try and run it 24/7 at 90 MHz (which is the default Voodoo 2 frequency).
Some memory ICs might be capable of this, others... not so much.

Thanks for the tip. I didn't get quite the delta that you got, but enough of a delta to make a determination. On average, the inner two chips were 2 C hotter than the outer two chips. Thus with Frame Buffer limited to 2 MB, I suspect the chips circled in red are the ones in use.

The chips on the bottom have more oxidation than those on the top, so I will likely start with the bottom, lower right chip. I will use my remaining 25 ns chip to swap in. Once I remove the old memory module, I will test it on a Trio64V+ PCI card with sockets.

Turns out that the temperature readings didn't correspond to the active memory. To limit to a 2 MB frame buffer on my card, I must remove all 4 memory IC's on the back as shown:

After removing the memory, I measured two faulty memory chips. Unfortunately, I don't have two 25 ns chips to replace them with. I have put in an order for matching chips and will report back after I solder them on.

As for my Diamond card, I am stumped. Seems like if the TMU's were working that mojo.exe would detect them. I measured all pins on the two TMU's and they are all making good contact. I visually inspected all the SOJ memory using 10x magnification and all the joints look good. If the memory was bad, mojo.exe would still show both TMU's, right? I'm thinking, at minimum, two TMU's on the Diamond card are bad.

Doesn't the FBI also serve as the go-between for the TMUs and the PCI bus? I'd think the FBI being bad would be more likely than both TMUs failing

maxtherabbit wrote on 2022-11-11, 15:16:

Doesn't the FBI also serve as the go-between for the TMUs and the PCI bus? I'd think the FBI being bad would be more likely than both TMUs failing

It could very well be, but at least mojo.exe detects the FBI, but not both TMU's, nor the TMU memory.

I received my replacement memory for one of my bad Voodoo2 cards. Although only 2 memory chips were bad for the 2nd megabyte of FBI memory, I replaced all 4 in that region. The card appears to be working fine now. Does anyone know if these memory chips are counterfeit, or if EliteMT changed their font slightly from 1998 to 2008?

Still need to fix the Diamond card, but need FBI/TMU.

mario990 wrote on 2020-04-16, 10:37:

I will replace FBI 500-0009-01 systems with Voodoo Diamond 3D II and ICUVGA-GW803 cards for the test. Diamond 3D II has already […]
Show full quote

I will replace FBI 500-0009-01 systems with Voodoo Diamond 3D II and ICUVGA-GW803 cards for the test.
Diamond 3D II has already soldered the new FBI 500-0009-01 I wrote about it in my first post, but I will exchange it and see what will happen ...

There is one more thing that puzzles me, TMU0 and TMU1 500-0010-01 systems were still dead when they were mounted on Diamond 3D II,
which may indicate a lack of power supply or connection with the FBI, you can see it here.
I did the test if the FBI Diamond 3DII correctly recognizes RAM with a full 4MB bank and 2MB desoldering, as you can see it recognizes correctly.DIAMOND 3D II log not ok 4MB FBI.jpgDIAMOND 3D II log not ok 2MB FBI.jpg

The log below ICUVGA-GW803 is good, despite the fact that TMU0 and TMU1 500-0010-01 had one leg broken off.
This leg connected with the memory, so the log is correct and the game Need For Speed III crashed due to a problem with texturing.ICUVGA-GW003 log ok.jpg

@mario990 - did you ever find the issue with that Diamond card? I have exactly the same "0xdead" and "57005" errors on my card, so I am keen to know what you found if you did fix it?