Cool, I have a voodoo 2 8 mb I never got going, mojo gave no clue. When the 3dfx logo shows up, it just locks up. Think I replaced some caps on it but no luck. Might try this and make a thread about it someday. 👍
Many thanks on all your work on this! I am looking forward to testing my non-functional Voodoo2s. Cheers!
Yesterday, I tested two of my broken V2s. The first one is missing 2 framebuffer memory chips, TMU0 and several TMU memory chips. The test shows correctly 2 bad memory chips, the rest seems to be ok (I know it doesn't test TMUs yet).
The second one has nothing missing but doest't work properly. I don't remember right now how it behaved, just that it was defective. Witchery suggests error at FB-DAC bus (pin 37 of FBI to pin 20 of DAC). And after measurement it indeed is not connected. So I guess bad VIAs in the PCB as this track is not visible on the surface. Well, I guess it is time to get a thin wire and repair the damaged trace 😀
Log file of the second card.
HW museum.cz - my collection of PC hardware
Thanks a lot, that's great !
I just tried it on my K6, which run Debian Etch, so 2.4.27-3-k6 kernel.
Of course it doesn't work on it, since there's no sysfs 😀
Not a problem, I was just curious.
havli wrote on 2026-02-02, 16:37:The first one is missing 2 framebuffer memory chips, TMU0 and several TMU memory chips. The test shows correctly 2 bad memory chips, the rest seems to be ok (I know it doesn't test TMUs yet).
If there is no TMU #0, Witchery should've failed on STAGE #7 (TMU #0 connectivity)...
havli wrote on 2026-02-02, 16:37:The second one has nothing missing but doest't work properly. I don't remember right now how it behaved, just that it was defective. Witchery suggests error at FB-DAC bus (pin 37 of FBI to pin 20 of DAC). And after measurement it indeed is not connected. So I guess bad VIAs in the PCB as this track is not visible on the surface.
Great! At least one board could be fixed with help of Witchery.
I'd check soldering under RAMDAC pin an FBI pin. Those can crack and not show any visible signs of that. But vias could also be a problem.
Hi,
i'm to stupid to run this on one of my Voodoo2 build, so i need help. I received "no place after kernel for initrd" when i run witchery in dos 6.22?!? Do You have some advice for such "smart" guy?
Thanks in advance
butjer1010 wrote on 2026-02-02, 19:54:Hi,
i'm to stupid to run this on one of my Voodoo2 build, so i need help. I received "no place after kernel for initrd" when i run witchery in dos 6.22?!? Do You have some advice for such "smart" guy?
Thanks in advance
What hardware are you trying to run it on?
Pentium+ and 32+ MiB RAM is recommended.
Fagear wrote on 2026-02-02, 19:59:butjer1010 wrote on 2026-02-02, 19:54:Hi,
i'm to stupid to run this on one of my Voodoo2 build, so i need help. I received "no place after kernel for initrd" when i run witchery in dos 6.22?!? Do You have some advice for such "smart" guy?
Thanks in advanceWhat hardware are you trying to run it on?
Pentium+ and 32+ MiB RAM is recommended.
Pentium 75 (AMD) + 32MB RAM
Or maybe 16MB??? Is it to low?
Added 64MB, now it's working. I didn't look for the system requirements 🙁
Thanks a lot man
Edit: something was wrong on this machine, add this hdd to K6-500 with 128MB of RAM, now it is ok. Working.... 😀
I don't know how to copy log file to disk, i can only view (F3) them, but when i copy them (F5), no matter where i wrote to copy them, they are not there?
butjer1010, you can copy your file to a mountpoint. You can navigate between them using "Disk" menu (F2).
Once you are done copying, make sure to do a clean reboot (either by using a "reboot" command or by pressing Ctrl-Alt-Del) or perform a flush (by the "sync" command) manually, so the buffers would be written out to your disk.
kmeaw wrote on 2026-02-02, 21:56:butjer1010, you can copy your file to a mountpoint. You can navigate between them using "Disk" menu (F2).
Once you are done copying, make sure to do a clean reboot (either by using a "reboot" command or by pressing Ctrl-Alt-Del) or perform a flush (by the "sync" command) manually, so the buffers would be written out to your disk.
I finally did it!
But the problem is, this time, everything passed the tests, but every other time i did the tests, results were different. One time it said that tests cannot be done because there is no FBI connection with PCI, once there was 2 RAM IC failed (U6 and U19), every time different errors, but this time everything is ok.
This Voodoo2 worked normally, but problem is somewhere in memory or somewhere else. It showed 4MB FB memory, and 2MB total Texture memory? Now, after 4 or 5 times i have run Witchery, it shows normally 4MB of total texture memory, and the log is ok as You can see. I will try run it few more times to get errors, and than see in system info amount of memory (so far, it always showed 2MB, it has never showed 4MB like now).
Looks like witchery has repaired my Voodoo 😀
I will try witchery again tomorrow, when the card "cools down" to see the results.
Very nice! Just came back from workshop after frustrating battle with my sli setup. No sli shown in Mojo, both cards are detected, in windows I can use 1024x768, but performance is single card like! Remade new sli cables few times , no good whatsoever! Question: I would like to try witchery in DOS, are the control commands the same, as in Linux package? Btw, should I use the program on slower CPU? Like and k6 500? As now my system is running on tualeron 1500+mhz?
butjer1010 wrote on 2026-02-02, 22:09:One time it said that tests cannot be done because there is no FBI connection with PCI
It could be caused by dirty Voodoo connector or PCI slot on MoBo.
butjer1010 wrote on 2026-02-02, 22:09:once there was 2 RAM IC failed (U6 and U19)
...
so far, it always showed 2MB, it has never showed 4MB like now
Probably you should look into what Witchery said in that log file. You probably have a bad connection on FBI chip at the top right corner or a marginal resistor nearby.
butjer1010 wrote on 2026-02-02, 22:09:It showed 4MB FB memory, and 2MB total Texture memory?
Witchery can not work with TMUs memory yet, unfortunately.
Poliux3dfx wrote on 2026-02-02, 23:00:No sli shown in Mojo, both cards are detected, in windows I can use 1024x768, but performance is single card like! Remade new sli cables few times , no good whatsoever!
At the momeny Witchery doesn't diagnose SLI issues and adding that is a low-priority task.
Poliux3dfx wrote on 2026-02-02, 23:00:I would like to try witchery in DOS, are the control commands the same, as in Linux package?
From DOS it boots into Linux environment through loadlin.exe. Witchery binary is the same.
Poliux3dfx wrote on 2026-02-02, 23:00:Btw, should I use the program on slower CPU? Like and k6 500? As now my system is running on tualeron 1500+mhz?
Witchery doesn't care about CPU speed. As long as Linux doesn't screw up event timer (which it did on ULi chipset).
It works on 300 MHz, it works on 2200 MHz.
Thanks Fagear! I will do testing as soon as possible, as there is some suspicion about FBI interface not initialising correctly.
Fagear wrote on 2026-02-02, 19:17:If there is no TMU #0, Witchery should've failed on STAGE #7 (TMU #0 connectivity)... […]
havli wrote on 2026-02-02, 16:37:The first one is missing 2 framebuffer memory chips, TMU0 and several TMU memory chips. The test shows correctly 2 bad memory chips, the rest seems to be ok (I know it doesn't test TMUs yet).
If there is no TMU #0, Witchery should've failed on STAGE #7 (TMU #0 connectivity)...
havli wrote on 2026-02-02, 16:37:The second one has nothing missing but doest't work properly. I don't remember right now how it behaved, just that it was defective. Witchery suggests error at FB-DAC bus (pin 37 of FBI to pin 20 of DAC). And after measurement it indeed is not connected. So I guess bad VIAs in the PCB as this track is not visible on the surface.
Great! At least one board could be fixed with help of Witchery.
I'd check soldering under RAMDAC pin an FBI pin. Those can crack and not show any visible signs of that. But vias could also be a problem.
Maybe I missed part of the report.
I'll try to repair one of the cards and report back when I make some progress. I will take time, unfortunately I have too many ongoing projects and very little free time.
HW museum.cz - my collection of PC hardware
Alright, I had some time to test it out on this artefacting Voodoo2 12MB.
So I ran Witchery:
Mojo reports only 1MB for TMU1.
Next I'm going to check all the pins on this TMU.
i386 16 ⇒ i486 DX4 100 ⇒ Pentium MMX 200 ⇒ Athlon Pluto 700 ⇒ AthlonXP 1700+ ⇒ Opteron 165 ⇒ Dual Opteron 856
Hello,
what a surprise that Witchery is finally publicly available!
Thank you for that.
I immediately had my defective Gainward Dragon 3000 12MB 110MHz diagnosed.
Regarding the actual problem: The card is recognized normally in Windows 98SE and I can install the driver, in my case FastVoodoo 4.6.
My test system consists of a Pentium 2 @ 266 MHz, 96MB of memory, and the Voodoo 2 is operated together with a GeForce MX card.
But as soon as I start a benchmark like 3DMark, I get a black screen and the PC crashes.
Witchery tells me there's a problem with the FBI's memory. More specifically, with the two memory chips U18 and U20. And I'm supposed to test the connections according to the log file.
I tested the connections suggested by Witchery and all connections have continuity, so I assume there are no interruptions.
The only thing I noticed when I looked with the thermal imaging camera was a hotspot in the U20 memory chip.
Could it be that the U20 memory chip is defective?
I have attached the thermal image and only a part of the log file, as the entire file is 10MB and I can apparently only upload 5MB files.
I would be very grateful for your opinion.
Thanks in advance!
PD2JK wrote on Today, 09:43:Alright, I had some time to test it out on this artefacting Voodoo2 12MB.
Mojo reports only 1MB for TMU1.
That looks like texture mapping artefacts, yep. Witchery can not test TMUs yet.
DocStreamer wrote on Today, 12:26:I immediately had my defective Gainward Dragon 3000 12MB 110MHz diagnosed.
Witchery tells me there's a problem with the FBI's memory. More specifically, with the two memory chips U18 and U20. And I'm supposed to test the connections according to the log file.
I tested the connections suggested by Witchery and all connections have continuity, so I assume there are no interruptions.
If suggested connections are indeed ok (you can in practice probe somewhat wrong or put pressure on pins and that will temporarely restore conenction), there are two more suggestions in Witchery's log:
if all listed connections are 100% ok, reflow solder on all affected ICs
if nothing above helped, then at least one of the listed ICs might be damaged internally and should be replaced
DocStreamer wrote on Today, 12:26:The only thing I noticed when I looked with the thermal imaging camera was a hotspot in the U20 memory chip.
Could it be that the U20 memory chip is defective?
It could be.
Memory chips could get hot by themselfes if not driven properly.
Upd.: both U18 and U20 are in second bank (bank 1). Presumably it stores FIFO for PCI queue as well if initialized, it'll explain why systems crashes on active operations with Voodoo.
U18 and U20 don't share data busses, but fully share address buses. But also those are shared with U17 and U19 that were tested good.
There's one signal that is shared only by those two chips: /RAS (pin 14) that connects to FBI's pin 127 via RP9.
DocStreamer wrote on Today, 12:26:I have attached the thermal image and only a part of the log file, as the entire file is 10MB and I can apparently only upload 5MB files.
You can put the log into an archive (zip or something) and it will fit.
I'd like to see the part above what you attached. Even better if you'll re-run test with "-v" key (in verbose mode) and attach that log file.
Keep in mind that Witchery can not physically test everything on the board and can miss stuff, its not perfect. It can be wrong. More analysis is probably needed.