VOGONS


First post, by grantek

User metadata
Rank Newbie
Rank
Newbie

I recently came into possession of this card that looks like a Guillemot Maxi Gamer 3D, and have had problems getting it going:

guillemot-maxi-gamer3d-voodoo-1.jpg
Filename
guillemot-maxi-gamer3d-voodoo-1.jpg
File size
547.66 KiB
Views
558 views
File license
Public domain

It looks clean (probably cleaned up by the previous owner if there was any visible damage), I've seen a few photos of these with different branding on the lower right of the PCB, this one has the Maxi Gamer 3D logo silkscreened on it, then silkscreened over with a white box.

After a lot of troubleshooting and experimentation with the driver variables, I've found:

  • Tomb Raider in pure DOS works perfectly
  • The card locks up after a few seconds in Windows 98 or 95 - in GLQuake it sometimes gets partway into the opening demo.
  • Currently using Glide v2.46 from the Guillemot driver CD, also tried 2.48 from the reference and Iceman driver packs
  • Downclocking to 25MHz (SST_GRXCLK=25), GLQuake works with no problems except slow performance.
  • At 27MHz I start getting card lockups again, sometimes as early as the 3DFX splash screen.

The most interesting thing I've found is downclocking to 20MHz, I get no lockups but see vertical bands of corrupt dots on the screen:

quake-dots-1.jpg
Filename
quake-dots-1.jpg
File size
464.19 KiB
Views
558 views
File license
Public domain
quake-dots-2.jpg
Filename
quake-dots-2.jpg
File size
490.85 KiB
Views
558 views
File license
Public domain
quake-dots-3.jpg
Filename
quake-dots-3.jpg
File size
488.83 KiB
Views
558 views
File license
Public domain

I'm hoping this means it's an issue with the FBI DRAM, because they're easier to replace than a failing FBI chip.

Can anyone with a working Voodoo1 (preferably with similar 50ns Alliance RAM) try SST_GRXCLK=20 and let me know if you can see similar frame buffer corruption? It's easier to see with textures disabled (SST_TEXMAP_DISABLE=1).

Also, does anyone know how the RAM clock is controlled in relation to SST_GRXCLK?

Reply 1 of 10, by Spark

User metadata
Rank Newbie
Rank
Newbie

I have that one, and had no end of problems with it locking up after a few minutes use, and then just not working at all when starting a game. This was in windows 98. Leaving the PC off for a few minutes would get it briefly working again.
I noticed the two 3dfx chips were very hot. I added a heatsink on each chip using regular double sided tape, and a single standard 40mm fan placed over the two heatsinks using sticky fixers, which solved all the problems. Now it has 100% stability and even overclocks to 57mhz without issue. I know this card didn't need cooling when it was new (I know the original owner). Hope this works for you.

Reply 2 of 10, by grantek

User metadata
Rank Newbie
Rank
Newbie

Thanks for the tip! I added some spare heatsinks with double-sided thermal tape, and pointed a case fan at them. It definitely helped, but didn't immediately get the card going above 30MHz

What did get it going was playing around with the MEMFIFO settings, I narrowed it down to SST_MEMFIFO_TEX=0 to get the card working stable at 50MHz with the cooling (this was against my suspicion of the texture memory being fine due to the lack of corrupt textures).

With this setup, when I took the fan away, the card would eventually lock up, and the heatsinks were pretty warm - so it's definitely helping. I do see some rare texture glitches (that I don't see at 25MHz), I'm not sure if those suggest something in the FBI/TMU chips, the RAM, or if they're timing issues caused by disabling whatever the FIFOs are supposed to do.

But it's definitely odd that the card's stable at 25 and 50MHz, but not in between (e.g. 30MHz, even with SST_MEMFIFO_TEX=0).

I'm hesitant to try replacing the DRAM chips if it won't obviously help (I do have a set of 25ns chips I could replace them with), so I'll try some other non-Quake benchmarks before resorting to that.

Reply 3 of 10, by matze79

User metadata
Rank l33t
Rank
l33t
3dfx.png
Filename
3dfx.png
File size
32.78 KiB
Views
429 views
File license
CC-BY-4.0

Looks weird

https://dosreloaded.de - The German Retro DOS PC Community
https://www.retroianer.de - under constructing since ever

Co2 - for a endless Summer

Reply 4 of 10, by Doornkaat

User metadata
Rank l33t
Rank
l33t

Thr Maxi Gamer 3D is the pickiest and wonkiest Voodoo I have tested.
Mine locked up at CPU speeds >333MHz and FSB speeds >66MHz iirc. It also had issues with different motherboards and PC cases.
Mine was an early card though while your's is a much later PCB.

Have you tried a different (slower) system when troubleshooting?

Reply 5 of 10, by adalbert

User metadata
Rank Oldbie
Rank
Oldbie

Recently I fixed a Trident card, which had video chip in exactly same form factor as Voodoo. I was getting video glitches too. The main chip just desoldered itself off the board, pins were floating above the PCB. You can see how it looks here in close-up (40 sec. in the video): https://youtu.be/syAhMC4H5aY?t=40

It was not visible with naked eye. Resoldering helped and the card now works fine, so you might want to check that (you can carefully poke the pins with a sharp tool and see if they are moving, hovewer take care not to mess things up). I was surprised that the only fault was just the solder itself and nothing else.

Repair/electronic stuff videos: https://www.youtube.com/c/adalbertfix
ISA Wi-fi + USB in T3200SXC: https://www.youtube.com/watch?v=WX30t3lYezs

Reply 7 of 10, by grantek

User metadata
Rank Newbie
Rank
Newbie
matze79 wrote on 2022-10-24, 08:16:

Looks weird

adalbert wrote on 2022-10-24, 12:21:

The main chip just desoldered itself off the board

rasteri wrote on 2022-10-24, 12:33:

Yep I had to resolder a bunch of pins on my maxi gamer 3d. I suspect they were never soldered in the first place, just they made enough contact with the pads that it worked for a while.

Hmm, interesting. I pulled the larger heatsink off the FBI chip, took some higher res shots, and ran the tip of a scalpel across the pins. I couldn't see any evidence of loose pins, and the legs look "wet" to the pads, but given the detail in the video linked (I watched a similar repair video here with similar issues) I think I'll need to get it under a microscope later this week.

Yesterday I did a bit more testing with Quake 2 and Unreal Tournament. With the cooling going, I got through the UT99 intro with no driver overrides, but had a lockup later on. With SST_MEMFIFO_TEX=0 and cooling it was stable, but I had noticeably worse performance in UT, with audio stuttering. The game is barely playable on this card anyway, but it seems like a good test (and might be more playable on a good card with driver optimisations). There were still occasional texture glitches, adding SST_MEMFIFO=0 and SST_MEMFIFO_LFB=0 seemed to get rid of them completely but make the audio stuttering worse.

Given the overall SST_MEMFIFO option I don't know if there's one hardware FIFO per chip or if one chip is routing its memory access through another with one or the other options set. Either way, I don't know if the driver settings are helping because it's accessing the memory through different pins, or in a different pattern that's more tolerant to glitches in the soldered connections, the RAM chips, or the 3DFX silicon. Or it's possibly just improving tolerance by slowing things down.

Doornkaat wrote on 2022-10-24, 09:22:

Have you tried a different (slower) system when troubleshooting?

Thanks for the tip, but this system is a Pentium 200MMX (66*3). I have another socket 7 motherboard but it's got basically the same speeds available.

Reply 8 of 10, by Imperious

User metadata
Rank Oldbie
Rank
Oldbie

I have almost the same problem with my rev E Diamond Monster. The FBI and TMU ic's get very hot when the card doesn't work.
On the rare occasion that it does work the ic's do not get anywhere as hot.
I have tried moving all pins on both ic's with a scalpel blade, none are loose. I haven't resoldered them yet though.
I have resoldered all ram pins which didn't make any difference.
I have checked and tested where possible all resistors, capacitors, coils, all seem good.
Downclocking to 20mhz I haven't tried but the card isn't designed to run at that speed so artifacts at 20mhz don't prove anything.

So what i suspect is the FBI chip is damaged due to overclocking or age or whatever and that's why it may work at lower speeds.
If You can fix it properly let us know. At some point I'll resolder the TMU and FBI chips and we'll see what if any difference that makes.

Atari 2600, TI994a, Vic20, c64, ZX Spectrum 128, Amstrad CPC464, Atari 65XE, Commodore Plus/4, Amiga 500
PC's from XT 8088, 486, Pentium MMX, K6, Athlon, P3, P4, 775, to current Ryzen 5600x.

Reply 9 of 10, by rasteri

User metadata
Rank Member
Rank
Member
Imperious wrote on 2022-10-25, 07:31:

I have tried moving all pins on both ic's with a scalpel blade, none are loose. I haven't resoldered them yet though.

Yeah the pins on mine were obviously loose, although I did have to look under a loupe to see it properly so maybe you'll see better under a microscope

It's very possible that un-soldered pins could cause these overheating problems, like if some of the GND or VCC pins were missing then it'll draw too much current from the remaining pins. Or go into an undefined state that causes latchup or something

Reply 10 of 10, by Doornkaat

User metadata
Rank l33t
Rank
l33t
grantek wrote on 2022-10-25, 00:33:
Doornkaat wrote on 2022-10-24, 09:22:

Have you tried a different (slower) system when troubleshooting?

Thanks for the tip, but this system is a Pentium 200MMX (66*3). I have another socket 7 motherboard but it's got basically the same speeds available.

I agree, that does not sound like it would be a CPU/FSB speed issue.
You may want to try the other motherboard nonetheless. In my experience some Voodoos are picky about their system.
Since this is a late Voodoo I tend to agree with your suspicion of some form of defect though. Sorry.