Thinkpad P72 EP720 NM-B723 MB challenging error - power turns off after 2 seconds

Hi,

Try to repair a Thinkpad P72 EP720 NM-B723 MB with I7-8850H processor and N18E-Q1-KB-A1 (P3200) GPU with 6GB memory. Unfortunately there is no info what happened to the laptop. I got it disassembled with parts missing, like keyboard, battery.

Unfortunately I have no schematic available for me but got the boardview file.

Symptom: after connecting to the charger the power led light up for one-two second and switch off. After a couple secons the power led light up for half a second and switch off. No battery, no IO board, no memory, no keyboard connected to the MB.

Checked the resistance on the power rails. Measured VCCPUCore resistance to ground 2 ohm, VCCGFXCORE_D 0.1 ohm. I have no experience on these resistances but I thought the GFXCORE rail is shorted.
Thermo camera was not able to pinpoint any parts with increased heat. The only visible change was the small heatup of the VCCPUCORE inductors and drivers.

Checked the NCP81274 phase buck controller controlling the VCCGFXCORE drivers with scope. PWMs are present, VCCGFXCORE goes to 1.25V and dorps to zero withing half a second. Power good PGOOD signal goes up and down accordingly. On the other hand according to the boarview file the PGOOD signal pulled up but not routed anywhere else.

Checked the NCP81215 three output controller controlling VCCPUCORE with scope. PWMs are present VCCPUCORE goes about to 1.3V and drops zero within a second.

Injected 0.9V (5A limit) on VCCGFXCORE. I got no voltage drop and 2 amper but there was not any heat up parts on the thermo camera. The GPU slowly warmed up to 32Celsius. I do not really understand this short. If it would be short the current should have reached the limit but not.

Same injection to the VCCPUCORE rail. No voltage drop, 1A current, no heated parts.

Removed the four coils, big tantal caps from the VCCGFXCORE rail measured 0.4 ohm resistance to ground, increased from 0.1ohm, it is quite a lot. Measured the lo and the high side resistance to ground of the drivers, seems to be OK, in couple Kohm range.

At this point I am asking help how to proceed forward. Do the GPU or the CPU toasted? Injection is not supporting this theory. These amperes are within the limits, there are no under and over voltage situations therefore the controllers should operate continously.

Unfortunately I do not know the power sequence of this laptop. As far as I see the power management is controlled by one of the two Lenovo ThinkEngine chip, but there is no documentation on the net about this. It could be possible there is an another reason the rails are disabled within a short period of time.

If anybody knows these power controller ICs thsn please advise how to measure the possible reasons of the rail disabling.

Any advice is appreciated in advance.
Modify message

Mita11 написав:VCCGFXCORE_D 0.1 ohm. I have no experience on these resistances but I thought the GFXCORE rail is shorted.

Hello!
True, this resistance seems too low.

Mita11 написав:Removed the four coils, big tantal caps from the VCCGFXCORE rail measured 0.4 ohm resistance to ground

Check the tantalum capacitors out of the board.

Thank you the suggestion. I removed the tantalum capacitors at the coils but three are soldered under the GPU which is dificult to remove, la ot of tiny capacitors next to it. It requires a lot of heat to remove it. Anyway I try to remove as well and let see how the resistance is changing on the rail and check the capacitor's parameters as well.

Mita11 написав:VCCGFXCORE_D 0.1 ohm. I have no experience on these resistances but I thought the GFXCORE rail is shorted.

This resistance seems good since it is a DISCRETE GPU core supply rail. For an integrated Intel graphics supply rails the label is VCCGFXCORE_I, so it may be easily confused because of very similar name.

Mita11 написав:Removed the four coils, big tantal caps from the VCCGFXCORE rail measured 0.4 ohm resistance to ground

Please also remove the 5th coil (not sure which 4 did you remove so I have marked all 5 coils on the picture below with yellow rectangles - source of photo: aliexpress.com) as it is a 5-phase converter, and measure the resistance to ground at MOSFET switching side.

Hi,

Thank you your suggestions and info.
N18E-Q1-KB-A1 GPU config has four stages drive implemented, the next bigger GPU has the full five drive stages I believe. The driving circuit, coil, caps are not populated for the fifth phase on my board. Anyway all the coils and the tantalum caps are removed.
Unfortunately my previous measurement was not precise, the VCCGFXCORE resistance not 0.3ohm but remaind 0.1-02ohm after removal of the mentioned components.
I tried to measue the possible short with milliohm meter.
The bad news is I was not able to pinpoint the exact location of the short.
The good news is that the short is not at the GPU. Measuring the milliohm resistances on the caps next to the GPU gave a bit higher resistance reading than measuring the resistances at the coils' output pads. It means the short is closer to the driver-controller circuit. Therefore I focused on the traces connected to VCCGFXCORE and the driver-controller circuit.
Without schematics and boardview I would not have any chances.
Traces:
1. NCP81274 controller's CSREF pin to all of the coil's output pad. Pin is OK, no short between the pin and the outpads, serial resistors, pin input impedance are ok.
2. NCP81274 VSP pin to VCCGFXCORE_D via 100ohm PR424, OK, no short to GND. This pin has connection to GPU's VDD_SENSE pin as well.

I was not able to find any other net connected to VCCGFXCORE_D. I assume I have to accept the 1V-2A result from injecting to VCCGFXCORE_D, it is probably not (hard) short. I have no other clue how would be possible to find the short if any.

Back to the origin: have to figure out why and how the power rail disabling is happening. My first suspect is the GPU rail overcurrent protection.

-- Posts merged --

Overcurrent measurement:

VCCGFXCORE_D is porduced by four NCP302045 driver from VGA_INT (20V). The current measurement performed at VGA_INT line by measuring the voltage drops on PR512,PR513 0.005ohm resistances. The measurement done by INA32221, GU10. This ic is programmed by the GU1O GPU via i2C interface. Should check the clock and the data lines for valid communication. If the GPU is dead there will be no I2C communication and no limits will be programmed. The INA provides alert and warning signals, should check as well. I found a bit strange this config. If INA reports critical error what will the GPU do? How the GPU notify the embedded controller about this state? Should investigate further after the INA signal checking.

The enable-disable mechanism is clear from the schematics for every power rail. But one important info is missing: the timimg sequence of the power rails on switch on-off-power error condition. The system power state table is provided on page 2 of the schematic but it is far from enough.

Addition:

I was able to move forward because bought the schematics and got the boardview file from the forum.

I replaced the 330uF tantalum capacitors on the VCCGFXCORE_D rail. Soldered back the coils as well.

The situation became worst: no power at all, not for one second. I have attached the schematic of the DC-IN section to better understand the "strange" situation.
Symptom: there is no voltage at the output of the PQ1 P-channel mosfet because its gate always on DCIN_PWR20_F, ie. at 19V. It seems the Thinkengine switches off the MB immediately with PWRSHUTDOWN signal pulled to low. PQ1 pin 5, output resistance to ground is in megaohm range, no output short present.

Measured PR8 resistor's both legs with scope at the same time and got the picture attached. Yellow line on pin 2, blue line on pin 1 of PR8. Pin 1 went to high and to gnd for about 1ms and to high again. 1ms wide 3V PWRSHUTDOWN signal switches PQ8 which turns on PQ6, grounding PR8 pin1. But. How on earth the PR8 pin2 do not follow this change? PR8 pin2 is on steady 19V instead of about 6.3V for 1ms.

To simplify the situation I removed the resistors with red circles eliminating the PQ1 gate and the PQ4 effects.

The situation remained the same, PR8 pin 2 did not react the change of pin1 voltage drop to gnd. We have a voltage divider PR3 (200k)-PR8(100k) which is not working. Between PR3 and PR8 about 6.3V should be present for 1ms but not, it is 19V all the time. It is like there is no current flowing through PR3. The resistance between PR8 pin1 and PR3 pin 2 is about 300kohm wich is ok based on the schematic.

If somebody has an explanation why the voltage divider is not wokring please share with me. What would be the next try to figure out what is happening with the gate voltage?

Other question is why Thinkengine PWRSHUTDOWN output is high for 1ms only. If it is low the main power rail disabled. PQ1 is never open therefore main rail overcurrent can not be the reason to shut down the main power rail. But the first question is the PQ1 gate voltage. If the PWRSHUTDOWN signal would be high all the time the PQ1 gate voltage still remained high and the main power rail voltage would be zero.