tehsiggi wrote on 2025-06-24, 08:56:

Hey there, as promised, here are some results. […]
Show full quote

Hey there, as promised, here are some results.

I did measure under idle and load.

Temperatures:

Ambient: 24°C
Cooling plate: 46°C in idle, 50°C under load.

Regulator output voltages:

Voltages were basically the same under load and idle.

VDDC: 1.505V
MVDDC: 2.847V
MVDDQ: 2.498V
VTT: 1.255V

I added a drawing that shows you where I measured which voltage.
All voltages were using the slot bracket as ground reference (using a clip). That'd be good enough..

The attachment measure.jpeg is no longer available

Power input:

Average values taken, you could easily measure the EXT currents using a current clamp, if available.

IDLE:
3.3V AGP: 1.456A / 4.663W
5V AGP: 0.502A / 2.571W
12V AGP: 0.164A / 2.018W

5V EXT: 2.311A / 11.742W
12V EXT: 0.213A / 2.629W

LOAD:
3.3V AGP: 1.925A / 6.169W
5V AGP: 0.612A / 3.131W
12V AGP: 0.164A / 2.020W

5V EXT: 2.498A / 12.660W
12V EXT: 0.267A / 3.229W

Thank you very much for this detailed post and the time that you took to get the information.
I'll work through this and get the comparative values for my faulty card.

tehsiggi wrote on 2025-06-24, 08:56:

Hey there, as promised, here are some results. […]
Show full quote

Hey there, as promised, here are some results.

I did measure under idle and load.

Temperatures:

Ambient: 24°C
Cooling plate: 46°C in idle, 50°C under load.

Regulator output voltages:

Voltages were basically the same under load and idle.

VDDC: 1.505V
MVDDC: 2.847V
MVDDQ: 2.498V
VTT: 1.255V

I added a drawing that shows you where I measured which voltage.
All voltages were using the slot bracket as ground reference (using a clip). That'd be good enough..

The attachment measure.jpeg is no longer available

Power input:

Average values taken, you could easily measure the EXT currents using a current clamp, if available.

IDLE:
3.3V AGP: 1.456A / 4.663W
5V AGP: 0.502A / 2.571W
12V AGP: 0.164A / 2.018W

5V EXT: 2.311A / 11.742W
12V EXT: 0.213A / 2.629W

LOAD:
3.3V AGP: 1.925A / 6.169W
5V AGP: 0.612A / 3.131W
12V AGP: 0.164A / 2.020W

5V EXT: 2.498A / 12.660W
12V EXT: 0.267A / 3.229W

Okay, the voltages on my card are:

VDDC: 1.35V
MVDDC: 2.89V
MVDDQ: 2.54V
VTT: 1.30V

That heatsink plate gets crazy hot very fast.
I don't have anything to measure the temperature but it's definitely well above 46°C. (Maybe +70°C in less than a minute. I cannot hold my finger on it after about 40 seconds)

I also confirmed that the core gets hot. It takes about 40 seconds to get too hot to touch and I didn't risk running the PC any longer.
I have a cooler for the card and obviously had it mounted while testing the voltages.

Thanks for measuring. VDDC appears to be too low imho. Things getting this hot this fast isn't normal either.
Can you measure the resistance between: 5V EXT and Ground as well as VDDC and Ground?
I'll get the corresponding values tomorrow morning.

It might well be helpful to remove the heatplate. It has two push pins and some thermal compound (usually sticky). It gets off best, if it's warmed up.
That'd make it possible to check which part specifically gets so hot.

Resistance was measured with the PC turned off.

5V EXT +/- 2.5KΩ
VDDC is a dead short (getting 0.03 Ω which is the resistance of my test leads)

Yes, I’ll remove the plate and identify the chip that is getting so hot.

It is these two chips that are getting so hot. Without the heatsink they burn me within 2 seconds

Ouch, with 0.03Ohms, things are not looking good.
It appears the SC1175 tries to pump power into it, thus the MOSFETs getting hot. Both of them are part of the SC1175 circuitry..

Now the big question: Is it a short in the GPU?
The easiest way to find that out: Desolder both Mosfets and check the resistance of VDDC again. If nothing changes from a value perspective, it can be assumed that the GPU has a short.

It also could be one of the capacitors on that rail, though I'd be surprised. Does any of the C's get hot on the back of the GPU? (like quicker than anything else?)

No, nothing gets hot at the back of the GPU, just slightly warm around the area without being concentrated in a single spot, which is quite normal I think.

Got some isopropyl?
Without a thermal camera, iso will evaporative fastest around the dead thing.
With luck it'll just be a cap.

I’ve removed the two mosfets but it hasn’t changed anything as far as resistance measurements are concerned.

I’m aware that these cards have a reputation for dead cores due to poor cooling but it could also be likely that a cap has failed or perhaps even one of the smd components.

Wouldn’t a shorted core have extremely hot spots and heat up very fast? That isn’t my experience here.

Before I give up on it, I’d like to work though the components and rule out as much as I can to be as sure as possible that it’s a dead core.

Maybe the SC1175 failed?
Maybe the cap where I’m measuring VDDC is faulty?
Maybe an SMD cap, resistor or something else in that circuit above?

Smd caps are renowned for failing short.
I dare say half of all ewaste is because a single smd cap failed.

If you measure that low of a resistance, you can rule out part by part. There is not much aside the GPU that is on that rail. Mostly caps. Anything that has the short and is on that rail would get hot. However if you measured 1.3V at the regulator output, it would seem that the regulator is able to shove some current into whatever is shorted.

That's why dm- proposed to use isoprop, it would disappear much faster one the warmed up component.

I'll mark you all parts that are on the VDDC rail by checking on my good card, so you can see which parts are on that rail.

If you reinstall the fets you could also check if there is activity coming from the SC1175 on their gates using your scope. Also the isoprop method is worth a shot.

tehsiggi wrote on 2025-06-25, 06:00:

I'll mark you all parts that are on the VDDC rail by checking on my good card, so you can see which parts are on that rail.

If you reinstall the fets you could also check if there is activity coming from the SC1175 on their gates using your scope. Also the isoprop method is worth a shot.

I checked that I was able to get replacement mosfets and then used a small side-cutter to snip off the legs of the mosfets. It was the fastest and easiest way to remove them but obviously means I cannot reinstall them. (It will take a while for the replacements to arrive though)

Thank you, having those components marked would be very helpful.

I’ll try the IPA option to see if anything gets hot.
Will that still work without the mosfets on the board?

Without the Mosfets there will be no power on the rail, so that test will have to wait then.

gpu resistance must be around 6-10 ohms. in case of short there will be no voltage at all and mosfets most likely will explode.

i saw a couple of times situation when small capacitor in pwm sense line may cause weird behaviour on mosfets causing overheat. detectable by oscilloscope.

connect a LAB power source to vgpu and detect heated areas.

Is 5v still grounded from the molex with the fets removed?

byte_76 wrote on 2025-06-24, 18:40:

Resistance was measured with the PC turned off. […]
Show full quote

Resistance was measured with the PC turned off.

5V EXT +/- 2.5KΩ
VDDC is a dead short (getting 0.03 Ω which is the resistance of my test leads)

Yes, I’ll remove the plate and identify the chip that is getting so hot.

The 5V EXT were never shorted.

12v then. Whatever feeds vddc upstream

myne wrote on 2025-06-25, 08:38:

12v then. Whatever feeds vddc upstream

VDDC is fed through 5V EXT and 3.3V AGP, none of which appear to have been shorted, since the system powered up and the short circuit protection did not kick in.

The SC1175 uses both rails to generate VDDC, it is a neat little chip..

It looks like the SC1175 is programmable, so even if it is faulty, replacing it might not be a simple task.

byte_76 wrote on 2025-06-25, 12:44:

It looks like the SC1175 is programmable, so even if it is faulty, replacing it might not be a simple task.

The SC1175 is not programmable. It's a two input fixed function PWM buck regulator IC. Any "programming" mentioned in the datasheet refers to configuration via external components, e.g. voltage limits or current sharing. But that has nothing to do with software.
If you switch that IC, it's just a 1to1 replacement.

I attached two pictures that mark all components which could potentially short VDDC to ground (apart from the GPU).
It's all capacitors + the SC1175.

Also, here's the whole segment of the SC1175 generation, if you look at it, there are no parts that could short the output to ground apart from the MOSFETs, the SC1175 and the capacitors. (Rb1 and Rb2 are very unlikely to be affected, they'd burn out directly)

Furthermore here's the whole segment that uses +VDDC:

The resistance of VDDC against ground is around 20Ohm - 30Ohms on my cards.