VOGONS


First post, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I was wondering what are the top 10 failure modes for Voodoo2 cards?

I have a Diamond Voodoo2 12 MB card that has been sitting in my box of dead PCB's for a number of years now and decided it would be a good time to figure out what's wrong with it. The initial symptoms were such that Windows 98SE wouldn't even detect a PCI Multimedia Device. So I started to poke the pins on one of the QFP chips to see if I could easily break them free from their solder pads. 3 such pins moved without any effort on the 500-0009-01 chip. I brushed on some flux and took a fine soldering iron tip and reflowed them. I also reflowed several other adjacent pins because the flux was already there.

Putting the Voodoo2 card back into the system, Windows 98SE now sees the card and I install the drivers. However, I receive an error, _GlideInitEnvironment: glide2x.dll expected Voodoo^2, none detected. I've placed a know working Voodoo2 into the same system and it works fine, so I don't think there is a configuration or driver issue.

What are some other common failure modes on these cards? Should I reflow all the pins on the 500-0009-01 chip? What about the two 500-0010-001 chips? Does the AS2830 voltage regulator commonly fail? How about the GENDAC ICS5342 clock generator? I checked all the caps and none were shorted.

A scan of the card is attached below.

Diamond_Voodoo2_top.jpg
Filename
Diamond_Voodoo2_top.jpg
File size
1.86 MiB
Views
905 views
File license
Public domain
Diamond_Voodoo2_bottom.jpg
Filename
Diamond_Voodoo2_bottom.jpg
File size
1.45 MiB
Views
905 views
File license
Public domain

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 1 of 50, by The Serpent Rider

User metadata
Rank l33t
Rank
l33t

Some cards will refuse to work at default clock. Probably due to prolonged thermal stress.

Get up, come on get down with the sickness
Open up your hate, and let it flow into me

Reply 2 of 50, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Alright. My plan is to reflow the whole 500-009-01 chip then see if I can reduce the card's operating freq.

It seems like there's a lot of dead Voodoo2 cards out there and figured there would have been more respondents with their fixes. I guess they all have stayed broken, which isn't very encouraging.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 3 of 50, by kolderman

User metadata
Rank Oldbie
Rank
Oldbie

Here's an uncommon failure mode on my current sli setup - one of the cards had a tiny cap fall off, the result being it can *only be used as the slave card*. Not alone, not as primary. But it is perfect as the slave in sli!

Reply 4 of 50, by feipoa

User metadata
Rank l33t++
Rank
l33t++
kolderman wrote on 2020-04-04, 02:21:

Here's an uncommon failure mode on my current sli setup - one of the cards had a tiny cap fall off, the result being it can *only be used as the slave card*. Not alone, not as primary. But it is perfect as the slave in sli!

That's something I had been wondering about, and is good news at least. Unfortunately, it requires another Voodoo2 card, which I suppose I should start looking for.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 5 of 50, by kolderman

User metadata
Rank Oldbie
Rank
Oldbie
feipoa wrote on 2020-04-04, 02:38:
kolderman wrote on 2020-04-04, 02:21:

Here's an uncommon failure mode on my current sli setup - one of the cards had a tiny cap fall off, the result being it can *only be used as the slave card*. Not alone, not as primary. But it is perfect as the slave in sli!

That's something I had been wondering about, and is good news at least. Unfortunately, it requires another Voodoo2 card, which I suppose I should start looking for.

Not just any, but same model/revision if possible.

Reply 6 of 50, by meljor

User metadata
Rank Oldbie
Rank
Oldbie

Your problem is on the backside: c151 , c51,c53, c61 are knocked off.... as far as I know the pads should all be populated (not at home and no voodoo at hand now...)

It`s fairly common that these get damaged/knocked off by other cards when kept in a bin/box. Had a couple from a recycler that had the same problem. Easy fix when you have spare parts or defective voodoo`s laying around. Errors seem to match yours... sometimes they work fine without a few and sometimes they give those errors you mentioned. Worth a try!

asus tx97-e, 233mmx, voodoo1, s3 virge ,sb16
asus p5a, k6-3+ @ 550mhz, voodoo2 12mb sli, gf2 gts, awe32
asus p3b-f, p3-700, voodoo3 3500TV agp, awe64
asus tusl2-c, p3-S 1,4ghz, voodoo5 5500, live!
asus a7n8x DL, barton cpu, 6800ultra, Voodoo3 pci, audigy1

Reply 8 of 50, by feipoa

User metadata
Rank l33t++
Rank
l33t++
The Serpent Rider wrote on 2020-04-03, 12:10:

Some cards will refuse to work at default clock. Probably due to prolonged thermal stress.

I used Voodoo2 Tweaker to set the glide and D3D pixel clock to 75 MHz, which is down from 90 MHz, but I receive the same error when opening a game.

meljor wrote on 2020-04-04, 08:49:

Your problem is on the backside: c151 , c51,c53, c61 are knocked off.... as far as I know the pads should all be populated

Good eye. What capacitances should these be?

EDIT: I soldered c61 back on properly, but I don't have another two-sided Voodoo2 card, so I cannot measure the values of c51, c53, and c151 when removed. As such, I soldered on 1.0 nF capacitors and put the card back in to the system. The error remains. Could the capacitor values be incorrect, or is there something else wrong with the card?

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 9 of 50, by meljor

User metadata
Rank Oldbie
Rank
Oldbie

Look at a high res picture online. That way you can determine which caps are the same size/color. Desolder those and measure.

Good luck!

asus tx97-e, 233mmx, voodoo1, s3 virge ,sb16
asus p5a, k6-3+ @ 550mhz, voodoo2 12mb sli, gf2 gts, awe32
asus p3b-f, p3-700, voodoo3 3500TV agp, awe64
asus tusl2-c, p3-S 1,4ghz, voodoo5 5500, live!
asus a7n8x DL, barton cpu, 6800ultra, Voodoo3 pci, audigy1

Reply 10 of 50, by feipoa

User metadata
Rank l33t++
Rank
l33t++
meljor wrote on 2020-04-04, 14:46:

Look at a high res picture online. That way you can determine which caps are the same size/color. Desolder those and measure.

I did this prior to selecting the 1 nF values for the 3 missing caps. The image I used was http://www.vgamuseum.info/images/palcal/diamo … a_bottom_hq.jpg . c61, c53, c51, c151 are all the same colour and size, however I don't know if this fact is sufficient enough to deduce the capacitance. In finding three 1.0 nF caps of this size, I desoldered a few dozen caps from an old HDD controller board. Caps of the same size and colour didn't necessarily have the same capacitances. Most capacitors were 100 nF, next common was 10 nF, then 1 nF and a few 0.5 nF and 4.5 nF caps.

The question I have in my mind is if c53, c51, and c151 are off by some variation, e.g. they should be 0.5, 10, or 100 nF instead of 1 nF, is that sufficient to result in the observed error when trying to use the card? If so, then I need to find these capacitance values, if not, what else could be the problem?

This is curious. On the PowerColor version of this card, C151 has a 0-ohm resistor instead of a capacitor as on the Diamond card. See these two images:

Diamond
http://www.vgamuseum.info/images/palcal/diamo … a_bottom_hq.jpg

PowerColor
http://ohwc.narod.ru/video/3dfx/v2_pc110_b.jpg

Why is this? Should I try shorting the terminals for c151? Is there anyone with a Diamond Voodoo2 rev. E who would be willing to desolder and measure c51, c53, and c151?

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 11 of 50, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I reflowed the two 500-0010-01 chips pin by pin, but it did not correct the problem.

I measured the voltages on the VRM, AS2830 (datasheet attached) and Vin = 5V; Vout = 3.5V; ADJ = 2.27V. Are these within range and what components on the PCB need 3.5 V to run?

I also took a look at the clock signals from the GENDAC ICS5342-3 clock generator. According to the datasheet (attached), the -3 model is the 135 MHz version. Should this frequency be visible at CLK0 or CLK1? I'm not getting anything near that freq. at these outputs. Has the 14.318 MHz crystal gone bad?

ISC5342-3_clk_output.png
Filename
ISC5342-3_clk_output.png
File size
8.46 KiB
Views
672 views
File license
CC-BY-4.0

Attachments

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 12 of 50, by Tiido

User metadata
Rank Oldbie
Rank
Oldbie

Freq depends on whatever resolution is needed, 1024x768 @ 60Hz isn't gonna need more than 60MHz pixel clock. the 30something MHz seems to be what is expected for 800x600.

The traces look pretty rounded, your probes or scope isn't up for the task.

T-04YBSC, a new YMF71x based sound card & Official VOGONS thread about it
Newly made 4MB 60ns 30pin SIMMs ~
mida sa loed ? nagunii aru ei saa 😜

Reply 13 of 50, by sdz

User metadata
Rank Newbie
Rank
Newbie

On V2 cards, AS2830 regulator (or equivalent) should output 3.5V, so that's fine on your card.
The RAMDAC should not output 135MHz back to the FBI, CLK1 should output 90MHz (when the cards isn't idle) and CLK0 varies depending on the selected video mode. In idle CLK1 should be ~44MHz, and CLK0 ~30MHz.
The 14.318MHz crystal is most likely OK, you can measure it if your oscilloscope probe doesn't load it too much.
The "glide2x.dll expected Voodoo^2, none detected" error indicates an error with the FBI, so you can probably ignore the two TMUs for now (unless they have shorted pins).
For missing 0603 caps @ FBI/TMUs, it's pretty safe to use 100nF 25V for the lighter color ones, and ~4.7uF >=6.3V for the darker ones.
If the FBI is properly soldered, check for broken traces, there are a few scratches on your card.

Reply 14 of 50, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I measured CLK0 and CLK1 again this morning at the DOS prompt and this is what I got. So its sounds like the GENDAC is outputing in the expected range.

ISC5342-3_clk_output_again.png
Filename
ISC5342-3_clk_output_again.png
File size
7.49 KiB
Views
626 views
File license
Public domain

I'm not sure what else to check for. Any stockpile of NOS 500-0009-01 FBI chips?

I did look at those few scratches on the back of the card to ensure the traces are through.

sdz> Do you think using 100 nF instead of 1 nF caps for c51, c53, c151 will make a difference? If so, I can change them. c61 was 1 nF though. I traced out c151 - one lead goes to GND, while the other lead goes to M66EN on the PCI slot. On the PowerColor card, these leads are simply shorted, and judging by the pin description for PCI, shorting M66EN to GND tells the system that 66 MHz operation is not possible.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 15 of 50, by sdz

User metadata
Rank Newbie
Rank
Newbie

I doubt changing those caps to 100nF will do anything. As for the cap between M66EN and GND, if the card doesn't connect M66EN to GND for 33MHz operation and leaves it floating/pulls it up for 66MHz operation, the cap is added there to make the pin act as a signal GND. IIRC this cap should be 10nF as per PCI specs. Not that it matters that much for a V2 or any other PCI card. As a side note, one could even leave M66EN pin completely disconnected on a V2 card and it would work just fine, most motherboards support only 33MHz, and if you plug it in a system that supports 33/66MHz the V2 will work just fine at 66MHz (in fact it can work fine on a quite faster bus than 66MHz).

If the FBI is soldered OK and the traces are OK, before replacing the FBI maybe it's worth checking the VIAs in the FBI area. I had a rare defect on a V2, it was missing HSYNC, because of a corroded VIA which was covered by a sticker.
Most likely not the issue, but you could also probe D0-D7 @ the RAMDAC to see if they are shorted to ground or the 5V rail. Before running any app there is quite a bit of bidirectional traffic there, among other things the FBI configures the clocks that it wants to receive. These 8 pins are also shared with the FBIs RAM, so in theory a bad RAM chip could prevent the FBI from initializing the RAMDAC.

Reply 16 of 50, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I can reflow the FBI again, but I did try to pry each pin with a scalpel to ensure it was secure.

Was it visibly obvious that the via on your card was corroded, e.g. it was covered in a greenish soot? How did you correct the problem?

While I can scratch off the protective layer and check for conductivity of the vias, I will not be able to determine the terminus of inner layer traces to ensure they are connected to the via.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 17 of 50, by sdz

User metadata
Rank Newbie
Rank
Newbie

The via just looked a bit weird. I managed to insert a small piece of wire in the via and soldered it at both ends.
You don't need to remove the soldermask for probing the VIAs, just push slightly harder with the probes and it should be OK. If the outer layers are connected, I wouldn't worry about the inner ones.

Reply 18 of 50, by feipoa

User metadata
Rank l33t++
Rank
l33t++

What brand of Voodoo2 were you troubleshooting? The solder mask / acrylic coating, or whatever it is, on the Diamond has been pretty resistant to pressure from my DMM probes, so I would be doing a lot of scrapping to verify each pin on the FBI. That or buy some probes with a sharper tip to break through.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486

Reply 19 of 50, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I was thinking of pulling my STB Voodoo2 out of my AMD K5-200 system to test the dead Diamond in SLI mode. The STB is 12 MB, but has only half the memory populated, so I'm guessing each IC is double density, while the Diamond has all the memory spots populated. Will these two cards work together in SLI and is there anything special I need to do to get them working together?

I don't have an SLI cable, but I recall reading that floppy cables work. Is there any modification I need to do to them? I've located my shortest floppy cable. It is about 9" long. Image attached.

Short_floppy_cable.jpg
Filename
Short_floppy_cable.jpg
File size
144.23 KiB
Views
519 views
File license
CC-BY-4.0
Last edited by feipoa on 2020-04-07, 01:17. Edited 1 time in total.

Ultimate 486 Benchmark | Ultimate 686 Benchmark | Cyrix 5x86 Enhancements | 486 Overkill Graphics | Worlds Fastest 486