Ahoy!

Off to the next race. The card is using the TI 4800 reference layout. So you can look for MS-8900 as a reference.

The resistors you've measured are not that off. You got a bit mislead by their label:

R107 is 221kOhms in the schematic and marked with 34D.
R109 is 301kOhms in the schematic and marked with 47D.

These are 1% tolerance resistors which use the EIA-96 marking which can be a bit confusing.
From what I'm seeing they are installed in the wrong way (switched the values).
They are part of the AGP reference voltage generation, so having them wrong isn't too good 😉

C110 is also a buffer cap for AGP reference voltages, good that it's back in place.

No voltages at all can mean that NVVDD_PWRGD is not in the right state, which in the case of that card would lead to FBVDD (memory) and VTT (memory) not being enabled. It would also prevent 3.3VL (used by all the logic on the board) from being enabled.
All of them wait for the PG signal from the GPU voltage, otherwise they do not enable and will not run.

So everything around U814 U831 is of interest. NVVDD rail resistance etc is worth looking out for.
You have the BOM that uses U831 in form of a ISL6525CB.

I hope that gives you something to start with.

By the way. 3.1V -> 3.0V at the input is a bit on the low side. Probably due to that (very nice!) flexible riser.

Cheers.

tehsiggi wrote on 2025-10-26, 06:05:

Ahoy! […]
Show full quote

Ahoy!

Off to the next race. The card is using the TI 4800 reference layout. So you can look for MS-8900 as a reference.

The attachment 8900-100.pdf is no longer available

The resistors you've measured are not that off. You got a bit mislead by their label:

R107 is 221kOhms in the schematic and marked with 34D.
R109 is 301kOhms in the schematic and marked with 47D.

These are 1% tolerance resistors which use the EIA-96 marking which can be a bit confusing.
From what I'm seeing they are installed in the wrong way (switched the values).
They are part of the AGP reference voltage generation, so having them wrong isn't too good 😉

C110 is also a buffer cap for AGP reference voltages, good that it's back in place.

No voltages at all can mean that NVVDD_PWRGD is not in the right state, which in the case of that card would lead to FBVDD (memory) and VTT (memory) not being enabled. It would also prevent 3.3VL (used by all the logic on the board) from being enabled.
All of them wait for the PG signal from the GPU voltage, otherwise they do not enable and will not run.

So everything around U814 U831 is of interest. NVVDD rail resistance etc is worth looking out for.
You have the BOM that uses U831 in form of a ISL6525CB.

The attachment ISL6525.PDF is no longer available

I hope that gives you something to start with.

By the way. 3.1V -> 3.0V at the input is a bit on the low side. Probably due to that (very nice!) flexible riser.

Cheers.

Did some measurements around this chip. 0 ohm resistor R1303 for UGATE / NVVDD is 73 ohms, eek!

Any other pins in particular you would like the resistance of?

That is very weird indeed. It's even labeled 000, which safe a 0 Ohm resistor. I wonder if there is flux or anything similar obstructing the measurement? It's quote odd.

What about the resistance between NVVDD and GND? somewhat in the 10s of ohms region? R1262 should have around 12V in turned on mode on both sides of the resistor. BOOT (10) and PVCC (13) should have 12V as well.
What voltage is there at NVVDD_PWRGD (7) during operation? That'd be the PG signal.

tehsiggi wrote on 2025-10-26, 11:01:

That is very weird indeed. It's even labeled 000, which safe a 0 Ohm resistor. I wonder if there is flux or anything similar obstructing the measurement? It's quote odd.

What about the resistance between NVVDD and GND? somewhat in the 10s of ohms region? R1262 should have around 12V in turned on mode on both sides of the resistor. BOOT (10) and PVCC (13) should have 12V as well.
What voltage is there at NVVDD_PWRGD (7) during operation? That'd be the PG signal.

I measured that 0 Ohm resistor again and its 0.6 ohms. You're right, it must of had gunk on it.

Here you go!

NVVDD and GND = 8.7 ohms (used C1301 to measure)
R1262 = 12v both sides
BOOT (10) = 12v
PVCC (13) =12v
NVVDD_PWRGD = 0.37mV

NVVDD_PWRGD is low, meaning the ISL6525CB is not happy with its output (yikes).
That in my mind would already explain why all the other voltages are missing.
The question is now, why it isn't happy. C1301 for measurement of the NVVDD is a good choice. Though the 8.7ohms appears to be on quite the low side. I'm currently taking care of the little one, who's ill 😁 I can get you a measurement of a regular Ti4200 later on, so we can compare.

In any case we can do a similar test as with your Radeon 9700:
You can probe the SS pin with an oscilloscope. If it looks somewhat like this, the ISL6525 considers the current limit on the output to be exceeded. This can have different reasons, but would be a relatively safe indicator that the ISL is "healthy".

You can also probe the low gate and high gate signal to see if there's something happening here.

Just make sure to have a "steady hand" when doing it.

Cheers

tehsiggi wrote on 2025-10-27, 07:28:

I'm currently taking care of the little one, who's ill 😁

Been there, done that 😉 My little ones are a little older now. Enjoy every minute of it.

tehsiggi wrote on 2025-10-27, 07:28:

In any case we can do a similar test as with your Radeon 9700: You can probe the SS pin with an oscilloscope. If it looks somew […]
Show full quote

In any case we can do a similar test as with your Radeon 9700:
You can probe the SS pin with an oscilloscope. If it looks somewhat like this, the ISL6525 considers the current limit on the output to be exceeded. This can have different reasons, but would be a relatively safe indicator that the ISL is "healthy".

The attachment Screenshot 2025-10-27 at 08.20.03.png is no longer available

You can also probe the low gate and high gate signal to see if there's something happening here.

Just make sure to have a "steady hand" when doing it.

Cheers

I really appreciate you taking the time to look at this and explain what you're doing. I'm sure many other people here are grateful too. Every time I fix one of these cards I learn a ton! I can reball cores but I can't read schematics, so I'm currently trying to teach myself (https://www.youtube.com/watch?v=4B6feSKfLxo). I want to understand how reading the schematics lead you to this chip (ISL6525) for investigation.

Here is the oscilloscope data...

SS Pin

LGATE Pin

UGATE Pin

I'll take a look at the picture during the "mid-day-nap" 😁

Regarding how I got to the IC as our first point to look, let me describe how I got there (very high level):

• We know that the graphics card as a system consists of two major components: GPU and Memory. They both need power.
• The GPU usually has one core power rail.
• The Memory, depending on the configuration, has up three rails.
• MVDDC is used for the memory ICs main core functions
• MVDDQ is used to power the I/O sections of the memory ICs (basically driving the data signals) - it is often also shared with the GPUs memory controller
• VTT is used to terminate the I/O of the memory ICs against a known voltage. This is important for signal integrity. You'll usually find this on cards using DDR1 memory with more than 250MHz. For frequencies below and equal 250MHz, mostly no so-called "parallel termination" is required, thus no VTT exists.
• There might be configurations where MVDDC and MVDDQ are the same voltage. Some graphics cards use the same power rail in that case.
• In the case of your card, I looked for sheets in the schematic that have POWER or POWER SUPPLY in the title, pages 24, 25 and 26.
• Page 24 is power sequencing, basically an auxiliarly schematic that checks if all incoming power is correctly applied before it allows the on-board regulators to create the required power rails. The signal is PWRGD_AGP - however there is a mention that this is not used when Intersil Power Supply chips are used (which are in your case) - so this signal can be ignored. (It's probably not even populated)
• Page 25 is NVVDD power supply. NV being the GPU (NVXX) and VDD being the high side voltage for it (VDD referring to DRAIN of a mosfet, VSS referring to SOURCE of a mosfet. For NFETs Source is negative potential and Drain is the positive potential)
• Page 26 as FBVDD-VTT POWER SUPPLY. FBVDD referring to Frame Buffer VDD and VTT for the termination voltage. So there are only two rails for the memory.
• Since the sheets are inter-connected, look out for off-sheet-labels (usually --->> or <<--- with a name next to it) These show the direction of the signal and its name.
• PG or POWER GOOD labels are always interesting. These are feedback of some sort that the power rail it belongs to is in proper shape. Often these are watched by other ICs to prevent half-powered scenarios
• Page 25 has NVVDD_PWRGD as an outgoing signal, so this is being checked somewhere else, which is interesting
• Page 26 has NVVDD_PWRGD as incoming signal, so this is being consumed here somewhere. Following it you'll see that it goes to the EN1 and EN2 pins of U830 (ISL6225) which controlls FBVDD and VTT. This concludes: These voltages will not be available unless NVVDD_PWRGD is high (good), otherwise the regulator is in shutdown mode for both outputs
• Searching for other occasions of NVVDD_PWRGD in the schematic shows that it goes to EN of U819 which generates 3.3VL (for the logic on the board). This is the case if R1141 is installed (0Ohms). If R1140 is installed, this will be controlled by the power good output from the FBVDD IC U830
• We now know that the NVVDD power rail is essential for everything else to start up. Looking at page 25 shows that there are three alternate chips that can be used: U814 - SC2602, U828 - HIP6012CB or U831 - ISL6625. Checking the pictures of your card shows that you have U831 installed.
  
  So U831 is apparently the one that has to be happy first, before anything else goes on. This is how I went through that. Not so easy to note down all thoughts in that process, but perhaps that helps a little.
  
  Also, if you have a red label on a signal like on page 25 and it appears anywhere else on that page a gain, those signals are connected via the Label, not necessarily by having a wire drawn between them. This i can be see a lot on page 25 and the alternate BOM versions (the three different ICs that can be populated.)

tehsiggi wrote on 2025-10-27, 10:05:

I'll take a look at the picture during the "mid-day-nap" :D […]
Show full quote

I'll take a look at the picture during the "mid-day-nap" 😁

Regarding how I got to the IC as our first point to look, let me describe how I got there (very high level):

• We know that the graphics card as a system consists of two major components: GPU and Memory. They both need power.
• The GPU usually has one core power rail.
• The Memory, depending on the configuration, has up three rails.
• MVDDC is used for the memory ICs main core functions
• MVDDQ is used to power the I/O sections of the memory ICs (basically driving the data signals) - it is often also shared with the GPUs memory controller
• VTT is used to terminate the I/O of the memory ICs against a known voltage. This is important for signal integrity. You'll usually find this on cards using DDR1 memory with more than 250MHz. For frequencies below and equal 250MHz, mostly no so-called "parallel termination" is required, thus no VTT exists.
• There might be configurations where MVDDC and MVDDQ are the same voltage. Some graphics cards use the same power rail in that case.
• In the case of your card, I looked for sheets in the schematic that have POWER or POWER SUPPLY in the title, pages 24, 25 and 26.
• Page 24 is power sequencing, basically an auxiliarly schematic that checks if all incoming power is correctly applied before it allows the on-board regulators to create the required power rails. The signal is PWRGD_AGP - however there is a mention that this is not used when Intersil Power Supply chips are used (which are in your case) - so this signal can be ignored. (It's probably not even populated)
• Page 25 is NVVDD power supply. NV being the GPU (NVXX) and VDD being the high side voltage for it (VDD referring to DRAIN of a mosfet, VSS referring to SOURCE of a mosfet. For NFETs Source is negative potential and Drain is the positive potential)
• Page 26 as FBVDD-VTT POWER SUPPLY. FBVDD referring to Frame Buffer VDD and VTT for the termination voltage. So there are only two rails for the memory.
• Since the sheets are inter-connected, look out for off-sheet-labels (usually --->> or <<--- with a name next to it) These show the direction of the signal and its name.
• PG or POWER GOOD labels are always interesting. These are feedback of some sort that the power rail it belongs to is in proper shape. Often these are watched by other ICs to prevent half-powered scenarios
• Page 25 has NVVDD_PWRGD as an outgoing signal, so this is being checked somewhere else, which is interesting
• Page 26 has NVVDD_PWRGD as incoming signal, so this is being consumed here somewhere. Following it you'll see that it goes to the EN1 and EN2 pins of U830 (ISL6225) which controlls FBVDD and VTT. This concludes: These voltages will not be available unless NVVDD_PWRGD is high (good), otherwise the regulator is in shutdown mode for both outputs
• Searching for other occasions of NVVDD_PWRGD in the schematic shows that it goes to EN of U819 which generates 3.3VL (for the logic on the board). This is the case if R1141 is installed (0Ohms). If R1140 is installed, this will be controlled by the power good output from the FBVDD IC U830
• We now know that the NVVDD power rail is essential for everything else to start up. Looking at page 25 shows that there are three alternate chips that can be used: U814 - SC2602, U828 - HIP6012CB or U831 - ISL6625. Checking the pictures of your card shows that you have U831 installed.
  
  So U831 is apparently the one that has to be happy first, before anything else goes on. This is how I went through that. Not so easy to note down all thoughts in that process, but perhaps that helps a little.
  
  Also, if you have a red label on a signal like on page 25 and it appears anywhere else on that page a gain, those signals are connected via the Label, not necessarily by having a wire drawn between them. This i can be see a lot on page 25 and the alternate BOM versions (the three different ICs that can be populated.)

Oh wow what brilliant information. I would have never figured this out by watching “Learn schematics” videos. I need to process this all and read the schematics again.

It's experience, that's hard to get from those videos. But it is however always a good starting point. Just don't try to boil the ocean!

Coming back to your scope pictures. It appears that you're in the realm of nS as your horizontal time base. That'd be MHz of frequency. We don't expect any interesting signals in that realm. Your measurements are basically down to noise right now.

I'd go with something in the 5-20µS / DIV for the gates and for the soft-start pin you're good with 1-20ms to see ramping. For voltage I'd go with 2-5 V/div and trigger on a rising edge around 2V, that'll keep the noise out.
If your scope permits, enable bandwidth limitation on the channel, it'll keep some garbage out.

We can calculate the actual switching frequency:

We'd just need to know if R1271 or R1263 is populated.

The nominal switching frequency of the chip is 200KHz, which equals a period of 5µS.

zuldan wrote on 2025-10-27, 09:20:

Here is the oscilloscope data...

The attachment SS Pin.jpg is no longer available

This is a classic demonstration of noise propagation across a physically distributed ground plane.
If you don't want to see something unknown every time, remove the ground alligator clips from the probe and discard them.
Train yourself to always use the Ground Spring for Oscilloscope Probe
Like this
https://www.aliexpress.com/item/4000494249484.html

Yup. And to make matters worse, the scope and PC usually are connected not only through the probe, but also through PE (earth) on the power supply side of things.

If you have to stick to the croc-clips, make sure to have a solid ground point on the card and don't let the ground cable of the probe flap around in the breeze. For those "laughable" measurements that we try here though, you'll be able to get proper results even with the croc clips..

Did the same in my measurements a while ago..

---

I just measured a regular Ti4200s NVVDD rail and I measured 9.8Ohms - so not far off from yours. So doesn't look too bad your measurement zuldan.

tehsiggi wrote on 2025-10-27, 14:47:

but also through PE (earth) on the power supply side of things.

If disconnecting the ground wire is impossible (including due to the risk of electric shock), there is a method that works even if you're trying to obtain certification for the device.
Connect the ground wire through two pairs of high-power diodes, connected in opposite directions and in series. (A 10+ Amp bridge rectifier with DC (+ and -) shorted outputs.)
Wrap the ground wire itself around a large ferrite ring.

If you have to stick to the croc-clips, make sure to have a solid ground point on the card and don't let the ground cable of the probe flap around in the breeze. For those "laughable" measurements that we try here though, you'll be able to get proper results even with the croc clips..

Short-circuit the tip and the alligator clip and bring the resulting magnetic antenna to the shielded inductor of the DC/DC converter.
Or touch the tip of the probe to the ground, even if it is close to the crocodile.
To hell with it, just grounding with a spring. 😀

shevalier wrote on 2025-10-27, 14:11:

This is a classic demonstration of noise propagation across a physically distributed ground plane. If you don't want to see some […]
Show full quote

This is a classic demonstration of noise propagation across a physically distributed ground plane.
If you don't want to see something unknown every time, remove the ground alligator clips from the probe and discard them.
Train yourself to always use the Ground Spring for Oscilloscope Probe
Like this
https://www.aliexpress.com/item/4000494249484.html

Thanks for the tip. I didn't even know that spring existed. I looked in the Rigol box and found a heap of them. I couldn't find a ground that was the correct distance for the spring to reach so I solderered a little wire and attached that to the spring but the signal still looks messy / noisy. I think it's something I'll have to play around with to get right.

tehsiggi wrote on 2025-10-27, 14:47:

Yup. And to make matters worse, the scope and PC usually are connected not only through the probe, but also through PE (earth) o […]
Show full quote

Yup. And to make matters worse, the scope and PC usually are connected not only through the probe, but also through PE (earth) on the power supply side of things.

If you have to stick to the croc-clips, make sure to have a solid ground point on the card and don't let the ground cable of the probe flap around in the breeze. For those "laughable" measurements that we try here though, you'll be able to get proper results even with the croc clips..

Did the same in my measurements a while ago..

The attachment Screenshot 2025-09-23 at 12.48.38.png is no longer available

---

I just measured a regular Ti4200s NVVDD rail and I measured 9.8Ohms - so not far off from yours. So doesn't look too bad your measurement zuldan.

I had the croc clips attached to the AGP bracket. Will see if I can secure it better somewhere else.

That's good your resistance is very close to mine. Gives me hope in getting this card working again.

tehsiggi wrote on 2025-10-27, 11:33:

It's experience, that's hard to get from those videos. But it is however always a good starting point. Just don't try to boil th […]
Show full quote

It's experience, that's hard to get from those videos. But it is however always a good starting point. Just don't try to boil the ocean!

Coming back to your scope pictures. It appears that you're in the realm of nS as your horizontal time base. That'd be MHz of frequency. We don't expect any interesting signals in that realm. Your measurements are basically down to noise right now.

I'd go with something in the 5-20µS / DIV for the gates and for the soft-start pin you're good with 1-20ms to see ramping. For voltage I'd go with 2-5 V/div and trigger on a rising edge around 2V, that'll keep the noise out.
If your scope permits, enable bandwidth limitation on the channel, it'll keep some garbage out.

We can calculate the actual switching frequency:

The attachment Screenshot 2025-10-27 at 12.17.45.png is no longer available

We'd just need to know if R1271 or R1263 is populated.

The nominal switching frequency of the chip is 200KHz, which equals a period of 5µS.

Ok that's way over my head. I've only used an oscilloscope to look at basic stuff. I got AI to write out some instructions for me based on your suggestion. Wish me luck!

The write-up instructions don't look too bad. I don't know of course the menus of your rigol, but you'll get around them I am sure of that.

I'll see today if I can find any card in my stash that uses a similar schematic, so I can see what normal operation would look like on pins for your reference.
Fingers crossed.

The similar rail resistance gives me hope as well!

zuldan wrote on Yesterday, 01:03:

tehsiggi wrote on 2025-10-27, 14:47:

Yup. And to make matters worse, the scope and PC usually are connected not only through the probe, but also through PE (earth) o […]
Show full quote

Yup. And to make matters worse, the scope and PC usually are connected not only through the probe, but also through PE (earth) on the power supply side of things.

If you have to stick to the croc-clips, make sure to have a solid ground point on the card and don't let the ground cable of the probe flap around in the breeze. For those "laughable" measurements that we try here though, you'll be able to get proper results even with the croc clips..

Did the same in my measurements a while ago..

The attachment Screenshot 2025-09-23 at 12.48.38.png is no longer available

---

I just measured a regular Ti4200s NVVDD rail and I measured 9.8Ohms - so not far off from yours. So doesn't look too bad your measurement zuldan.

I had the croc clips attached to the AGP bracket. Will see if I can secure it better somewhere else.

That's good your resistance is very close to mine. Gives me hope in getting this card working again.

https://www.youtube.com/watch?v=fZ7tOMyi07w
https://www.youtube.com/watch?v=henwiLzSieA

The noise is around 30 mV, and the video card's DC/DC converters are not working.
When they are enabled, this noise voltage is around 100 mV.
Are the capacitors dry and need replacing, or is it noise?
Here you need to decide whether to do the measurements- or safe or correct. 🙁

@zuldan
According to the ISL6525 block diagram, its oscillator should start immediately.
Check for a sawtooth voltage on pin 1.
Typically, in these controllers with the RC-generator, capacitance is hidden inside the crystal, and there's a signal across the Rt- resistor.
(Depending on the circuit design, there may be a sawtooth voltage directly, or, if a current mirror is used to set the capacitor charging current, then artifacts from the operation of the generator itself may be visible.)
And try lift the soft start pin from the PCB so that nothing is connected to it at all.

shevalier wrote on Yesterday, 06:49:

https://www.youtube.com/watch?v=fZ7tOMyi07w
https://www.youtube.com/watch?v=henwiLzSieA

The noise is around 30 mV, and the video card's DC/DC converters are not working.
When they are enabled, this noise voltage is around 100 mV.
Are the capacitors dry and need replacing, or is it noise?
Here you need to decide whether to do the measurements- or safe or correct. 🙁
[/quote]

I watched both of those videos yesterday, thanks. The problem with the spring is if you can't find a ground that's compatible with the distance the spring can do then you're out of luck.

I'll only know if the caps are bad if I desolder them (and for SMD caps that usually means melting their base plastic casing which I don't want to do just yet).