This looks a lot like bad RAM on the primary VSA100.
You can run Video Memory Stress Test to further investigate. The archive also contains a .iso file you can burn to CD to use the test without Windows.

Best of luck! 😀

Edit: I think I found a picure of your card on twitter. What's all that white stuff between the left RAM chips?
If it's still there try cleaning the area and inspecting the RAM chips' legs for failed solder joints.

Edit2: Yup, that's a reflection. ^^; Disregard.

Doornkaat wrote on 2020-08-11, 17:12:

This looks a lot like bad RAM on the primary VSA100. You can run Video Memory Stress Test to further investigate. The archive al […]
Show full quote

This looks a lot like bad RAM on the primary VSA100.
You can run Video Memory Stress Test to further investigate. The archive also contains a .iso file you can burn to CD to use the test without Windows.

Best of luck! 😀

Edit: I think I found a picure of your card on twitter. What's all that white stuff between the left RAM chips?
If it's still there try cleaning the area and inspecting the RAM chips' legs for failed solder joints.

Edit2: Yup, that's a reflection. ^^; Disregard.

Yep! That's my card, sadly I spoke to soon in thinking it was working on my tweet because I didn't have the right JST headers to connect the 5v fans so I only tested as far as post to prevent any chance of overheat.

Currently running the stress test from bootable floppy, and it's reporting errors from 01FFFFFF all the way to 0201265E currently and still climbing, during the "Ones" test.

On the 5500 does the address range reporting issues point to a particular chip?

Will update this post when the test completes

[Update]
So an odd development, I was getting so many errors I decided to swap in my voodoo 3 3000 to see if it was some sort of main board issue, the Voodoo 3 3000 worked fine and completed multiple passes without error, but as I went to reach for the 5500 to reinstall it my cat got spooked and knocked it off the table, upon installing it,it completed without error, so I let it boot into windows and it didn't have artifacting, device manager showed the drivers were installed, so I booted quake 2 in 800x600 opengx 3dfx mode and it ran for about 20 seconds before artifacting returned, and now I'm back to square one and getting a constant stream of errors in video memory stress tester for 'Ones' test.

I'm thinking the light knock was just coincidental and it was just the card cooling briefly that made the difference,. And it's either still the memory or I need to replace all of the capacitors.

Any ideas?

avxstudios wrote on 2020-08-11, 17:55:

[Update] So an odd development, I was getting so many errors I decided to swap in my voodoo 3 3000 to see if it was some sort o […]
Show full quote

[Update]
So an odd development, I was getting so many errors I decided to swap in my voodoo 3 3000 to see if it was some sort of main board issue, the Voodoo 3 3000 worked fine and completed multiple passes without error, but as I went to reach for the 5500 to reinstall it my cat got spooked and knocked it off the table, upon installing it,it completed without error, so I let it boot into windows and it didn't have artifacting, device manager showed the drivers were installed, so I booted quake 2 in 800x600 opengx 3dfx mode and it ran for about 20 seconds before artifacting returned, and now I'm back to square one and getting a constant stream of errors in video memory stress tester for 'Ones' test.

I'm thinking the light knock was just coincidental and it was just the card cooling briefly that made the difference,. And it's either still the memory or I need to replace all of the capacitors.

Any ideas?

That sounds a lot like a cracked BGA solder joint under the VSA100 chip.
Try carefully but firmly pressing the chip against the PCB. If this changes the artifacts (or makes them disappear completely) it's 99% a bad solder connection. In this case the card might simply need a reflow. Note that this is different from chip-substrate-bonds failing on newer GPUs which can not be permanently fixed with reflowing!
The bad connection may also be on the RAM chips so carefully inspect all legs on those and see wether they're making good contact. If one of those is loose you're in luck; this can be fixed with a steady hand and a soldering iron.

So some really weird developments.

I dipped a new toothbrush in isopropyl and scrubbed down the contacts of all the ram chips incase there was any debris or contaminates causing partial shorts or full shorts, and now the card boots normally to windows, and 5 minutes of quake 2 didnt have any issues, silky smooth.

Video memory tester however does not like it, in dos (bootable floppy) it does not get the cards video memory correct, if you try to specify 64mb, it floods you with errors (im guessing because its dual gpu and each gpu manages 32mb each), if you specify 32mb it works a little better but still throws out errors, however, if you set the ram size to 28,708,864 bytes, it completes the full suite of tests just fine with no errors, in Windows the windows version of the test suite states 28,708,864bytes of ram, and completes fine too, despite 3dfx tools reporting 64mb.

I'm guessing it's just a matter of VMT not playing ball with a dual GPU voodoo card.

https://streamable.com/am34hi
Here's some one handed Quake 2 action, despite the video there isn't actually any flickering it's just my phone not liking CRT.

Test in windows:

Test in DOS:

avxstudios wrote on 2020-08-11, 21:00:

So some really weird developments. […]
Show full quote

So some really weird developments.

I dipped a new toothbrush in isopropyl and scrubbed down the contacts of all the ram chips incase there was any debris or contaminates causing partial shorts or full shorts, and now the card boots normally to windows, and 5 minutes of quake 2 didnt have any issues, silky smooth.

Video memory tester however does not like it, in dos (bootable floppy) it does not get the cards video memory correct, if you try to specify 64mb, it floods you with errors (im guessing because its dual gpu and each gpu manages 32mb each), if you specify 32mb it works a little better but still throws out errors, however, if you set the ram size to 28,708,864 bytes, it completes the full suite of tests just fine with no errors, in Windows the windows version of the test suite states 28,708,864bytes of ram, and completes fine too, despite 3dfx tools reporting 64mb.

I'm guessing it's just a matter of VMT not playing ball with a dual GPU voodoo card.

Sorry to spit in your soup but this is not normal behaviour for that card. 🙁 My Voodoo 5500 tested fine with VMT CE. I have the PC in storage right now so I can't post proof but maybe somebody else has one on hand? Forcing the test to run with 64MB won't work as each chip only has 32MB and VMT CE can only test the primary chip - you're totally right about that. But forcing the test to limit itself to a known good memory area doesn't prove it works fine. 😉
From what you describe I really think this problem has to do with bad contact between the primary VSA100 and its memory.

Doornkaat wrote on 2020-08-12, 11:40:

avxstudios wrote on 2020-08-11, 21:00:

So some really weird developments. […]
Show full quote

So some really weird developments.

I dipped a new toothbrush in isopropyl and scrubbed down the contacts of all the ram chips incase there was any debris or contaminates causing partial shorts or full shorts, and now the card boots normally to windows, and 5 minutes of quake 2 didnt have any issues, silky smooth.

Video memory tester however does not like it, in dos (bootable floppy) it does not get the cards video memory correct, if you try to specify 64mb, it floods you with errors (im guessing because its dual gpu and each gpu manages 32mb each), if you specify 32mb it works a little better but still throws out errors, however, if you set the ram size to 28,708,864 bytes, it completes the full suite of tests just fine with no errors, in Windows the windows version of the test suite states 28,708,864bytes of ram, and completes fine too, despite 3dfx tools reporting 64mb.

I'm guessing it's just a matter of VMT not playing ball with a dual GPU voodoo card.

Sorry to spit in your soup but this is not normal behaviour for that card. 🙁 My Voodoo 5500 tested fine with VMT CE. I have the PC in storage right now so I can't post proof but maybe somebody else has one on hand? Forcing the test to run with 64MB won't work as each chip only has 32MB and VMT CE can only test the primary chip - you're totally right about that. But forcing the test to limit itself to a known good memory area doesn't prove it works fine. 😉
From what you describe I really think this problem has to do with bad contact between the primary VSA100 and its memory.

Puzzling,
I checked in Windows and VMT sees it as having 31,599,680 bytes (31.6mb), setting it to that in VMTCE in dos completes the test successfully, not sure why it doesn't see the full capacity in Windows, and I've never really had an issue with memory before where the capacity has reduced a small amount on one chip but tested fine at the smaller amount 🤔

I suppose it might be the system itself but I don't have any other AGP based systems on hand to test with to see if that's the case.

Does anyone here have an agp 5500 they could run VMT on Windows with to see what capacity it says is available?

avxstudios wrote on 2020-08-13, 01:20:

Does anyone here have an agp 5500 they could run VMT on Windows with to see what capacity it says is available?

I have a Voodoo 5 5500 PCI I could run if you get no AGP offers of help? Essentially the same card anyway.

paradigital wrote on 2020-08-14, 06:51:

avxstudios wrote on 2020-08-13, 01:20:

Does anyone here have an agp 5500 they could run VMT on Windows with to see what capacity it says is available?

I have a Voodoo 5 5500 PCI I could run if you get no AGP offers of help? Essentially the same card anyway.

Worth a shot! I'd imagine the memory is identical across both agp and PCI versions

So VMT on my fully functional Voodoo 5 5500 PCI shows 31,559,680 bytes.

paradigital wrote on 2020-08-17, 06:57:

So VMT on my fully functional Voodoo 5 5500 PCI shows 31,559,680 bytes. […]
Show full quote

So VMT on my fully functional Voodoo 5 5500 PCI shows 31,559,680 bytes.

Thanks!
This is relatively good news as that's how much windows VMT detects mine as having, and passes fine with, and passes in dos if that amount is manually set, I'm guessing dos version of VMT just doesn't know how to properly pull the right ram value for the ago version 🤔

Would be interested to see if anyone else with the AGP version experiences the same detection failure issue for dos VMT.

If I let VMT automatically detect in dos , it erroneously detects it as having over 135mb of ram, and manually specifying 32mb as 32mb also fails as VMT reads that as 33,554,432 bytes as apposed to the 31,559,680 bytes that the windows version sees it (and from your post, the PCI version too) of having.

I remember trying a 4MB ASUS Riva 128 PCI card with VMT CE and getting lots of errors. Turns out somehow the program assumed the card had 8MB VRAM.
You can select different memory detection methods when configuring the test manually.

Well, I tried to run the DOS version of VMT, but the top section of the screen disappears seconds after starting (I can see the statistics section, but nothing above that, it just goes blank white). The test continues to run, but clearly it's testing invalid memory areas as the test quickly starts racking up thousands of errors.

In Windows the test completes successfully, and in actual real world and synthetic benchmarks the card performs flawlessly, so I'm not particularly convinced by VMT, especially the DOS VMTCE version.

paradigital wrote on 2020-08-18, 07:06:

Well, I tried to run the DOS version of VMT, but the top section of the screen disappears seconds after starting (I can see the statistics section, but nothing above that, it just goes blank white). The test continues to run, but clearly it's testing invalid memory areas as the test quickly starts racking up thousands of errors.

In Windows the test completes successfully, and in actual real world and synthetic benchmarks the card performs flawlessly, so I'm not particularly convinced by VMT, especially the DOS VMTCE version.

Interesting. This is the same behaviour as with that PCI Riva 128. What board were you using and did you manually specify any test parameters?

paradigital wrote on 2020-08-18, 07:06:

Well, I tried to run the DOS version of VMT, but the top section of the screen disappears seconds after starting (I can see the statistics section, but nothing above that, it just goes blank white). The test continues to run, but clearly it's testing invalid memory areas as the test quickly starts racking up thousands of errors.

In Windows the test completes successfully, and in actual real world and synthetic benchmarks the card performs flawlessly, so I'm not particularly convinced by VMT, especially the DOS VMTCE version.

This is the exact behaviour exhibited with my voodoo 5, I recorded the screen and was able to go back and look at the frames right before the top of the screen disappeared and saw that it had detected the card as 135mb, if I manually set the card to 32m, the exact same portion of the screen just be slightly corrupts instead, and gets a little further through the test before streaming a long list of errors.

When manually setting the exact same number of bytes as the windows version detects, it runs perfectly fine without error, I'd say the memory detection for the DOS version isn't as good with detecting the correct ram amount on cards, it works for some but it seems there's a fair few cards it will only correctly auto detect in the windows version.

Hi guys.
Do you know if mojo is working with voodoo 5 5500 cards?
I've tried 2 versions, one tells that no voodoo card was found, the other sees something but stops with errors.
Is there any other good DOS testing tool available?
I'm strugglyng with a card that shows vertical artifacts and lots of garbage.
I'm concerned right now about the primary GPU.
I've measured every single pin of the first 4 RAM chips, checked numerous times every connection.

VMT throws thousands of errors. There is some pattern in them.
Here I attach some saves made by VMT.
If the lower byte is either 0xD or 0xE, and the 4th byte is odd, I get repeating 0x79 0xCF instead of 0xFF.
If the lower byte is either 0xA 0xB 0xC or 0xF, and the 4th byte is even, I get some other pattern.
Is there a good way to find out which RAM is faulty?

Thanks.

szabi_o wrote on 2023-08-02, 16:01:

I'm concerned right now about the primary GPU.

Don't be. Until you have hard evidence that this is the case, your first assumption should be that it's a faulty memory IC (which is a very frequent occurrence).

Admittedly, I've never tested Mojo with cards newer than Voodoo 2, but I have a hard time believing that it would work (the version I have is from 1997, so hard to believe that it can detect a card / chip released in the year 2000).

szabi_o wrote on 2023-08-02, 16:01:

Is there a good way to find out which RAM is faulty?

Unfortunately, not (nothing like a modern nVIDIA MATS/MODS equivalent)... unless you're lucky and the damaged chip has a short which causes a temperature increase (which can be detected with an IR thermometer / thermal camera / good ol' finger test).
You say that you already verified the pins of each of the four memory ICs (just as an FYI, the primary VSA-100 and its memory ICs are the ones closest to the bracket).
Fortunately, it's just 4 memory chips, so there's a good chance that you could get away with just replacing one or two until the problem goes away.

"Once I tested tsop sdram and I detected what tsop was causing the problems. I started to reemplace it for a new one and voila..."
Someone wrote this here: http://www.voodoo3dfx.com/Foro/viewtopic.php? … a2515fffe968165
So I guess there is a way to tell which one is bad.
Where would you source replacement IC's? They are not that common. You cannot get them from some sdram modules.
I don't have a thermal camera, but all 4 are heating up quite a bit. Including the VSA.
btw, the card isn't working under Windows. Freezes after drivers are installed.

As far as I know, he doesnt have a method to test individual chips, he does have a lot of experience for educated guessing.

One utility exists for 3dfx cards, but no one compiled it so far, it contains various hardware tests. Well I did, it just compiled wrong.. eventually when I take the time 😜

I usually identify faulty chips by running the card outside of the computer and probing the RAM ICs with an oscilloscope.