The 3.3V from my post refer to the power rail, just to make sure. Not the AGP signal voltage.

But yes, I've seen cards with no obvious issue that just didn't like to run in a certain way, just like your Ti4200. The spirits of hardware I guess.

What board/platform is this ?
What drivers did you used for chipset ?
Do you get into Windows in VGA mode ?
Did you tried both DVI ports ?
Do you tried VGA (on both ports) ?

agent_x007 wrote on 2025-05-31, 21:45:

What board/platform is this ? What drivers did you used for chipset ? Do you get into Windows in VGA mode ? Did you tried both D […]
Show full quote

What board/platform is this ?
What drivers did you used for chipset ?
Do you get into Windows in VGA mode ?
Did you tried both DVI ports ?
Do you tried VGA (on both ports) ?

What board/platform is this ? - > I've tried a Biostar M7VIT 800 (in the picture), ABIT AB-NF7-S V2.0 and a MSI K8T Neo.
What drivers did you used for chipset ? -> Hyperion Pro v5.24a and nForce v5.10
Do you get into Windows in VGA mode ? -> Yes VGA mode works fine as it sets the screen resolution to 640x480
Did you tried both DVI ports ? -> Yes, both have been tested
Do you tried VGA (on both ports) ? -> There are no VGA ports

Archer57 wrote on 2025-05-31, 10:22:

zuldan wrote on 2025-05-31, 09:17:

Here are some basic voltages

Are those still fine when the screen goes blank?

Yes voltages remain the same after the screen goes blank

tehsiggi wrote on 2025-05-31, 10:31:

I just had a check. Your card has the same layout as the MSI Card MS8872 - 200. I looked at the schematic, from your marked vol […]
Show full quote

I just had a check. Your card has the same layout as the MSI Card MS8872 - 200.
I looked at the schematic, from your marked voltages from top to bottom we have:

VTT (Termination voltage)
FBVDD (Framebuffer voltage aka. Memory voltage
NVVDD (GPU)

FBVDD and NVVDD are on point, supposed to be 2.8V and 1.7V respectively. So that is good.
VTT however is supposed to be exact half of FBVDD, meaning it should be 1.4V instead of 1.2V.

R1038 and R1039 should form the reference voltage divider, coming from FBVDD. There are test-points around:

TP29 should measure 1.4V. You can find it directly next to the two inductors for FBVDD and VTT. It's also marked as "TP29". The resistors are on the backside of the board, next to the SC1175. According to the schematic they're both 1K 1%.

TP29 measures 1.2v

tehsiggi wrote on 2025-05-31, 11:01:

Just some things I've noticed as well: […]
Show full quote

Just some things I've noticed as well:

The card appears to derive the main voltages the following way:
FBVDD is generated from the 12V from the AGP slot. VTT is generated from FBVDD.

NVVDD is generated from the 3.3V from the AGP slot. The core takes up to 10A at 1.7V, meaning around 17W, this would translate to ~5A on the 3.3V rail at the AGP slot.

Then there is a voltage call 3.3VL - which is generated from the 5V from the AGP slot. It appears to be mainly used by DAC circuitry and is linear, being very low noise.

If somehow possible for you, a look at the 3.3V on the AGP slot can be interesting. Don't risk too much though. Depending on the mainboard there could be a significant drop at those currents, leading to potential issues for the regulator.

This issue is occurring on all on three different motherboards I have. Is it still worth looking at the AGP voltage of this particular motherboard? I also found new (old stock power supply) with huge amount of amps on 5v but still no luck.

zuldan wrote on 2025-05-31, 09:49:

Found an old version of MATS. Thanks Vogons! It supports Geforce 1 up to the FX series. It looks like the memory is 100% OK.

Re: GeForce3 - mixing memory modules

The attachment MATS.JPG is no longer available

Was the memory at stock?

It sounds similar to a dud mosfet/driver.
As soon as the load goes up, the mosfet/driver can't deal, and the core stops.
Of course, with 20 year old capacitors, it's more than likely they're the cause.

Switching power supplies are exactly what they sound like.
They rapidly turn the "mains" on and off to get a lower voltage on the downstream. This is obviously very abrupt, so inductors and capacitors are used to smooth it out and give the practical appearance of a smooth lower voltage.

Go to your nearest light switch and turn it on and off really fast.
Instant, isn't it?
Now if you add a capacitor (assuming dc) it will be less instant.
The peak voltage will be lower, and the lowest voltage will be higher.
If you size the capacitor perfectly and time the switching perfectly, the light will stay on without flickering.

Well, it sounds like your gpu is flickering.

zuldan wrote on 2025-06-01, 00:13:

tehsiggi wrote on 2025-05-31, 10:31:

I just had a check. Your card has the same layout as the MSI Card MS8872 - 200. I looked at the schematic, from your marked vol […]
Show full quote

I just had a check. Your card has the same layout as the MSI Card MS8872 - 200.
I looked at the schematic, from your marked voltages from top to bottom we have:

VTT (Termination voltage)
FBVDD (Framebuffer voltage aka. Memory voltage
NVVDD (GPU)

FBVDD and NVVDD are on point, supposed to be 2.8V and 1.7V respectively. So that is good.
VTT however is supposed to be exact half of FBVDD, meaning it should be 1.4V instead of 1.2V.

R1038 and R1039 should form the reference voltage divider, coming from FBVDD. There are test-points around:

TP29 should measure 1.4V. You can find it directly next to the two inductors for FBVDD and VTT. It's also marked as "TP29". The resistors are on the backside of the board, next to the SC1175. According to the schematic they're both 1K 1%.

TP29 measures 1.2v

Now i am curious. Could you check the markings on r1038 ans r1039? Perhaps gainward uses different values.

tehsiggi wrote on 2025-06-01, 04:56:

Now i am curious. Could you check the markings on r1038 ans r1039? Perhaps gainward uses different values.

I'm having trouble locating r1038 ans r1039 on the PCB 😉 Going to have another look...

zuldan wrote on 2025-06-01, 05:10:

tehsiggi wrote on 2025-06-01, 04:56:

Now i am curious. Could you check the markings on r1038 ans r1039? Perhaps gainward uses different values.

I'm having trouble locating r1038 ans r1039 on the PCB 😉 Going to have another look...

Never mind, I found them...

myne wrote on 2025-06-01, 01:21:

Was the memory at stock? […]
Show full quote

Was the memory at stock?

It sounds similar to a dud mosfet/driver.
As soon as the load goes up, the mosfet/driver can't deal, and the core stops.
Of course, with 20 year old capacitors, it's more than likely they're the cause.

Switching power supplies are exactly what they sound like.
They rapidly turn the "mains" on and off to get a lower voltage on the downstream. This is obviously very abrupt, so inductors and capacitors are used to smooth it out and give the practical appearance of a smooth lower voltage.

Go to your nearest light switch and turn it on and off really fast.
Instant, isn't it?
Now if you add a capacitor (assuming dc) it will be less instant.
The peak voltage will be lower, and the lowest voltage will be higher.
If you size the capacitor perfectly and time the switching perfectly, the light will stay on without flickering.

Well, it sounds like your gpu is flickering.

Yes I restored the BIOS back to the original with standard clock speed then ran the test.

I'm not 100% convinced it's the capacitors yet. I took a few out and tested their capacitance and ESR. All were 100%, although you just need 1 bad capacitor to make things go bad. A full recap is on the table...

There were also reports of people having trouble connecting GeForce 4 cards to modern monitors using DVI to HDMI adapters.

Not sure if that's the case with your system, but I would try using an older monitor with actual DVI input, if available. Try both DVI ports on the card too.

zuldan wrote on 2025-06-01, 05:20:

zuldan wrote on 2025-06-01, 05:10:

tehsiggi wrote on 2025-06-01, 04:56:

Now i am curious. Could you check the markings on r1038 ans r1039? Perhaps gainward uses different values.

I'm having trouble locating r1038 ans r1039 on the PCB 😉 Going to have another look...

Never mind, I found them...

The attachment Resistors.JPG is no longer available

Glad we're checking this. So apparently Gainward does use a slightly different BOM. They are using 1k for R1038 (as MS8872) but 750Ohms for R1039 (instead of 1k) - this leads to the reference voltage for the second channel of the SC1175 to be exactly 1.2V, which is spot on in your testing.

I wonder: You did set lower clocks via the BIOS and were able to get into windows. Might trying only changing one clock at a time? Either GPU or Memory?

Transients on the NVVDD line could be the case. But especially since you tested the caps incl. ESR, that would appear odd. If the equipment is there, there's no harm in doing a recap. 560µ instead 510µ ones should work, as long as their ESR is similar.
For a definitive analysis a look with an oscilloscope on NVVDD at the time of load (perhaps single trigger, falling flank, trigger voltage 1.6V) could shed some light.

Tricky card

tehsiggi wrote on 2025-06-01, 05:39:

Glad we're checking this. So apparently Gainward does use a slightly different BOM. They are using 1k for R1038 (as MS8872) but […]
Show full quote

Glad we're checking this. So apparently Gainward does use a slightly different BOM. They are using 1k for R1038 (as MS8872) but 750Ohms for R1039 (instead of 1k) - this leads to the reference voltage for the second channel of the SC1175 to be exactly 1.2V, which is spot on in your testing.

I wonder: You did set lower clocks via the BIOS and were able to get into windows. Might trying only changing one clock at a time? Either GPU or Memory?

Transients on the NVVDD line could be the case. But especially since you tested the caps incl. ESR, that would appear odd. If the equipment is there, there's no harm in doing a recap. 560µ instead 510µ ones should work, as long as their ESR is similar.
For a definitive analysis a look with an oscilloscope on NVVDD at the time of load (perhaps single trigger, falling flank, trigger voltage 1.6V) could shed some light.

Tricky card

Modifying the BIOS (and confirming with Z-GPU and Everest), I reverted the cards memory back to 650MHz (from 350MHz) and Windows XP booted up perfectly in 800x600. I then changed the GPU back to 300MHz (from 250MHz) and the video loss problem returned, I couldn't boot into Windows XP. I was able to go into VGA mode (640x480) and get to the desktop.

Now that I have the schematics, I'm going to order caps from Mouser closer to the actual specs. If the capacitors were causing the problem, which ones do you suggest I replace first?

Only just got my oscilloscope. It's a cheapo FNIRSI 1014D. I need to figure out if it has the capability to do what you're asking 😉

Again, it can just as easily be the mosfets or drivers.
In all seriousness, give it a wash.
Yes. Wash. In the sink.

myne wrote on 2025-06-01, 06:32:

Again, it can just as easily be the mosfets or drivers.
In all seriousness, give it a wash.
Yes. Wash. In the sink.

Will do!

zuldan wrote on 2025-06-01, 06:29:

Modifying the BIOS (and confirming with Z-GPU and Everest), I reverted the cards memory back to 650MHz (from 350MHz) and Windows […]
Show full quote

tehsiggi wrote on 2025-06-01, 05:39:

Glad we're checking this. So apparently Gainward does use a slightly different BOM. They are using 1k for R1038 (as MS8872) but […]
Show full quote

Glad we're checking this. So apparently Gainward does use a slightly different BOM. They are using 1k for R1038 (as MS8872) but 750Ohms for R1039 (instead of 1k) - this leads to the reference voltage for the second channel of the SC1175 to be exactly 1.2V, which is spot on in your testing.

I wonder: You did set lower clocks via the BIOS and were able to get into windows. Might trying only changing one clock at a time? Either GPU or Memory?

Transients on the NVVDD line could be the case. But especially since you tested the caps incl. ESR, that would appear odd. If the equipment is there, there's no harm in doing a recap. 560µ instead 510µ ones should work, as long as their ESR is similar.
For a definitive analysis a look with an oscilloscope on NVVDD at the time of load (perhaps single trigger, falling flank, trigger voltage 1.6V) could shed some light.

Tricky card

Modifying the BIOS (and confirming with Z-GPU and Everest), I reverted the cards memory back to 650MHz (from 350MHz) and Windows XP booted up perfectly in 800x600. I then changed the GPU back to 300MHz (from 250MHz) and the video loss problem returned, I couldn't boot into Windows XP. I was able to go into VGA mode (640x480) and get to the desktop.

Now that I have the schematics, I'm going to order caps from Mouser closer to the actual specs. If the capacitors were causing the problem, which ones do you suggest I replace first?

Only just got my oscilloscope. It's a cheapo FNIRSI 1014D. I need to figure out if it has the capability to do what you're asking 😉

A small update....GPU at 250Mhz and memory at 650MHz works for a minute or so but then the no signal problem returns. So, it appears reducing the memory speed is also required to make it stable.

Just to recap...

- GPU 300, Memory 650 = Cannot boot into Windows XP at 800x600
- GPU 250, Memory 650 = Can boot into Windows XP at 800x600 but after a while video signal is lost
- GPU 250, Memory 350 = Can boot into Windows XP at 800x600

* In all situations, as soon as DirectX is initialized (EG 3dMark or Quake 3 is run), I get video signal loss.

zuldan wrote on 2025-06-01, 06:29:

Modifying the BIOS (and confirming with Z-GPU and Everest), I reverted the cards memory back to 650MHz (from 350MHz) and Windows XP booted up perfectly in 800x600. I then changed the GPU back to 300MHz (from 250MHz) and the video loss problem returned, I couldn't boot into Windows XP. I was able to go into VGA mode (640x480) and get to the desktop.

Now that I have the schematics, I'm going to order caps from Mouser closer to the actual specs. If the capacitors were causing the problem, which ones do you suggest I replace first?

Only just got my oscilloscope. It's a cheapo FNIRSI 1014D. I need to figure out if it has the capability to do what you're asking 😉

Good.

If you are considering changing caps, go with anything around NVVDD, as this is the GPU core voltage. Meaning C1302 and C1304 on the output side of things. Input would be C1264 and C1267.

C1302: 510µF - 4V - 14mOhms ESR - 4.08A Ripple current rating
C1304: 510µF - 4V - 14mOhms ESR - 4.08A Ripple current rating
C1264: 510µF - 4V - 14mOhms ESR - 4.08A Ripple current rating
C1267: 510µF - 4V - 14mOhms ESR - 4.08A Ripple current rating

There are alternate variants as well, which appear to be used on the input side of your card:
470µF - 6.3V - 25mOhms ESR - 2.09A Ripple current rating

On the input side there's also C1371, which is more of a general bulk capacitor and not as critical for the switcher as the other ones.

C1371: 1000µF - 6.3V - 170mOhm ESR - 0.45A Ripple current rating.

Just from a quick look your scope might actually be able to do some basic analysis. If you do so, don't do it "free hand", ideally solder some test point to the PCB that you can remove later, where you can attach the probe securely to. That prevents shorting something out by accident.

myne wrote on 2025-06-01, 06:32:

Again, it can just as easily be the mosfets or drivers.
In all seriousness, give it a wash.
Yes. Wash. In the sink.

I agree with the cleaning, that's always part one. However make sure to have it properly dried and any residue removed.

Flaky FETs or drivers haven't come around for me a lot the last years, so they wouldn't be my first concern.

Again, without tracing the GPU voltage in time of error, it's hard to say if it's even the issue, same for caps.

zuldan wrote on 2025-06-01, 06:43:

A small update....GPU at 250Mhz and memory at 650MHz works for a minute or so but then the no signal problem returns. So, it app […]
Show full quote

A small update....GPU at 250Mhz and memory at 650MHz works for a minute or so but then the no signal problem returns. So, it appears reducing the memory speed is also required to make it stable.

Just to recap...

- GPU 300, Memory 650 = Cannot boot into Windows XP at 800x600
- GPU 250, Memory 650 = Can boot into Windows XP at 800x600 but after a while video signal is lost
- GPU 250, Memory 350 = Can boot into Windows XP at 800x600

* In all situations, as soon as DirectX is initialized (EG 3dMark or Quake 3 is run), I get video signal loss.

Interesting. Especially that it is a "time related issue". Does the time vary between "runs" and "no signal" with memory at normal clock, if you perform many 2D draw actions? Like opening files, refreshing the desktop etc?

If the timing is fairly consistent and unrelated to that, it doesn't sound to me like it is "traffic" induced, but could be environmental.
Are you able to apply heat or cold (like serious heat ~100°C or -20°C cold) to certain components to see if it has an influence?

For starters going with the easiest: Extreme airflow on both sides of the card to see if that prolongs the "happy phase".

Cheers!

EDIT: In general, you scope should be good enough to have a simple look at ripple on NVVDD, that gives a good indication how "stable" everything is. You can find NVVDD at TP25 if you want to add a test-pin.

tehsiggi wrote on 2025-06-01, 06:57:

Interesting. Especially that it is a "time related issue". Does the time vary between "runs" and "no signal" with memory at norm […]
Show full quote

Interesting. Especially that it is a "time related issue". Does the time vary between "runs" and "no signal" with memory at normal clock, if you perform many 2D draw actions? Like opening files, refreshing the desktop etc?

If the timing is fairly consistent and unrelated to that, it doesn't sound to me like it is "traffic" induced, but could be environmental.
Are you able to apply heat or cold (like serious heat ~100°C or -20°C cold) to certain components to see if it has an influence?

For starters going with the easiest: Extreme airflow on both sides of the card to see if that prolongs the "happy phase".
Cheers!

At GPU 250, Memory 650 = The video signal loss is random, sometimes it comes after 5 seconds, 30 seconds and sometimes 60 seconds (rarely).
At GPU 250, Memory 350 = I cannot get anything 2D related to fail the card (refreshing, spamming 2d applications).

I'll run a heat gun over it tomorrow then power it up to see if anything changes. I'll use IR camera to make sure it gets close to 100c.

Actual 100°C is not "too important", it's just more important to trip anything temp related over a certain point.