hello

i have this chip in a laptop, after some modding attempt things went south and it shows up with code 43 (during failure first time i seen 43 i shut it down, powered on again then it came fine, but afterwards is 43 always)

involved was on-time resistor change to lower the switching freq of the memory VRM, during test it was able to withstand furmark at 1940MHz, 1.35V (with default there were always crashes at 1900MHz)
memory is 4 x H5GC8H24MJR-R0C
VRM was G5335 10A converter, i replaced it with AOZ2264QI 15A - but there is no change

VDDC resistance is about 2ohm - same as before
VDDCI resistance is about 33ohm - same as before
MVDD resistance went bad it is 3.2ohm, before it was 56ohm - and also i have noticed slight voltage drop about 30mV on 1.35_GFX rail

GPU dead, or memory, or both?

Use one of near-short-circuit finding method. Put 1V voltage with external supply and look if some of memory ICs is getting hot (thermovision/Isopropanol evaporation/etc)

there are 12 VREF pins in total, on 11 of them there is 0.946, 0.947 or 0.948 V, but on one pin there is 0.955 V. does that mean anything?

vram module closest to the vrm output heats up quite faster than other modules. interesting, but it is not the one with the higher VREF voltage although it is on the same channel
also, found there is 0V on low side gate of VDDCI. does that indicate damaged mosfet?

It seems that at least this the fast-heating module is damaged. However, quite often all modules are damaged, not the only one. Desolder it to verify.

About VREF - Where and in what conditions the VRef was measured?

Sometimes, some gates show low or zero 0OHm to ground while they are completely fine. It looks to be related to inner workings of the gate drivers. After 1-2 days wothout ANY, including measurments electrical activity - such gates become fine.

So, if this gate is the only "non-typical" thing for VDDCI, and overall VDDCI is fine - it may be completely fine.

As far as I understand the only obvious problem is MVDD line.

see attachment, on VREFD2 side of C1033, VRM output 1.35V

any ideas how do the laptops behave without one or two memory modules? are they still able to boot?
already tried removing inductors of the VRAM/GPU core, but as there is no PWROK the controller does not allow boot - need to know if the same would happen without some memory modules, as i can still use it with CPU graphics

decided to take more measurements, as i was under impression that something in the sensing circuit might have failed, mostly because of the GPU while tested with modified on-time was sometimes reporting abnormal current values even in idle state (139.72A, exactly twice of the limited current) as can be seen in screenshot

the voltages measured were:

COMP 1.43V
COMP_NB 1.27V
FB 0.83V
FB_NB 0.80V
NTC 1.06V
NTC_NB 1.05V
IMON 41mV
IMON_NB 33mV
ISEN1 0.75V 0.83V
ISEN2 0.75V 0.83V
RTN 0.4mV
PGOOD 3.32V
PGOOD_NB 3.32V
PWROK 3.32V
ENABLE 3.32V
VSEN 0.83V
VSEN_NB 0.81V
ISUMN 0.83V
ISUMP 0.83V
VR_HOT_L 30mV
SVD 1.8V
SVT 1.81V
ISUMN_NB 0.81V
ISUMP_NB 0.81V
VDD 5.16V
VDDP 5.16V
LGATE_NB 0V
PHASE_NB 0.81V
UGATE_NB 0.9V
BOOT_NB 1.10V (4.45V???) - one measurement gave 4.45V on boot capacitor but i think it had no effect on output voltage to GMC
LGATE1 4.67V
PHASE1 0.763V / 0.83V - one measurement was giving 0.83V on both sides of the core inductor, otherwise it is about 0.75V on the VRM side
UGATE1 0.9V
BOOT1 4.81V 4.81V (5.01V 4.98V) - measured both sides of boot capacitor, sometimes there is about 4.8V on both, other times is the other value
LGATE2 4.66V
PHASE2 0.756V / 0.83V - one measurement was giving 0.83V on both sides of the core inductor, otherwise it is about 0.75V on the VRM side
UGATE2 0.9V
BOOT2 4.81V 4.81V (5V 4.71V) - measured both sides of boot capacitor, sometimes there is about 4.8V on both, other times is the other value


also resistor on 3 controllers in the system (all 3 are ISL62771)


PR390 is VSEN_NB pullup to VDDCI
PR438 is VSEN pullup to VDDC
PR446 is RTN pulldown to GND

maybe 0.6 is fine on them, but it doesn't look so when compared to integrated graphics or cpu controller

VR_HOT_L is on? 30mV on GPU only, on CPU it is 3.28V

"To open MOSFET the gate to source voltage is between 0 and 1V (Vgs). So zero volts on the gate will keep mosfet shut. the SMPS will not work!"

desoldered the mosfet, resistance between its pins is large (in megaohms) as well as the resistance from ground to the LGATE pin on the board. VRM should drive voltage on this pin? if yes then it's either VRM failure, or VRHOT disabled this pin.

I have checked resistance across boot capacitors, on both GPU core phases, on both CPU phases and northbridge phase that shares same VRM as the CPU there is about 10megaohm. Only on VDDCI phase of the GPU VRM there is 2ohms with capacitor on the board and the same 2ohms without it...

also noticed that even though one memory module heats up faster than the others, after few minutes of system powered on, all memory modules have an identical temperature (55-60c)

correction of a previous message: bold text, measurement were done on resistors that are in series with boot caps

ISL62771 removed, short on VDDCI boot-phase went away. checked the removed PWM it remained shorted.
A resistance of 0.2-0.7ohm forms along PGOOD line.
There is logic chip U24 - SN74LV1T08DCKR on 4th leg comes PEGX_RST, with the board connected to power supply, but powered off it had 68mV. This voltage should be present or not?

Took PCI resistance at C8 - C23 the result

PEG_TXP0 162K
PEG_TXN0 162K
PEG_TXP1 150K
PEG_TXN1 150K
PEG_TXP2 155K
PEG_TXN2 155K
PEG_TXP3 150K
PEG_TXN3 150K
PEG_TXP4 160K
PEG_TXN4 160K
PEG_TXP5 160K
PEG_TXN5 160K
PEG_TXP6 162K
PEG_TXN6 162K
PEG_TXP7 167K
PEG_TXN7 167K

are they acceptable or out of specification?

There was 28mV buildup on PEGX_RST even with all supplies disconnected. U24 removed, now there's nothing on PEGX_RST without supply. With supply 6mV comes.

A faulty reset logic can engage throttle on memory or not?

Isopropyl evaporation test on VRAM in the end gave inconsistent result.
With 0.8V applied, VRAM3 dried out sooner than others.
With 1.1V VRAM1, VRAM2 and VRAM3 dried out at the same time, while VRAM4 was still half wet.
Looks like there is no rule. Anybody experienced something like that?

Added 1.4V - VRAM4 hot, VRAM3 very warm, VRAM2 hotter than 3 but not as hot as 4. VRAM1 warm

Can resistances on DQ test points give any clue?

GPU heating up slightly when voltage is applied to GDDR5, there is no leakage, both VDDC and VDDCI were 0 volt

VREFD2 and PEGX_RST can be ignored

In case of PEGX_RST there was interference from too close LED light that reflected as tiny voltage on DMM

VREFD2 resistor 5.49K was out by 100ohm 5.59K out the board gave

Now after new ISL62771 is soldered nothing starts, presumably because of resistance on PGOOD. there is no even 3.3GFX or 1.8GFX.

With broken ISL62771 under same conditions there was 0.9V on VDDCI - must have been shorted so badly that it leaked to VDDCI that's why it's gate drivers burned out. Regardless of that, on the GPU side both VDDC and VDDCI resistance are still good

After having broken video card gv-rx460wf2oc-2gd bought in ebay lying around for more than a year it was finally time to unsolder GPU chip and put it on dell board.

Previuosly, there was dead hynix memory, damage on ISL62771, one dead input mosfet (near the charger IC), and lastly even though I'm unsure of GPU condition after first failure, accidentaly it was beaten to the end afterwards. Also 5.49K resistor mentioned above became completely open after few days of sitting on the table.

Since i was buying from taobao and could get anything at low price i also took all the logic gates, 1331 load switch, vddci 7400 and 7516 mos and put new ones. On one out of 5 HP compaq board i found better 22uf capacitors than those on dell board in the 1.35V output and changed those too. Other power components were replaced as planned earlier, higher rated core mosfets, two big powder inductors, which required vddci inductor to be smaller than original in order to fit next to the big two which is fine since there is small consumption of vddci. Memory inductor also changed as well as memory VRM to 15A.

Samsung HC22 memory was put together with 215-0895088 taken from gigabyte card. Whoever assembled it on gigabyte squeezed the heatsink like an insane, as a result there are permanent traces on die (at least it appears so, IPA99 did not clean it) and the chip got bent in the corners (or in the middle, appearing like it is in the corners). I spent a night reflowing it because of this, untill finally managed to align it and get consistent readings on VRAM test points.

System booted, code 43 in windows, yeah some software problems i forgot about (signature patch)
Worked fine in ubuntu 16.04, memory tested with dmgg script - no errors

After patching windows driver crossfire enabled and GPU started. Few short tests in furmark and half an hour of gaming, no issues.

Also to add, all the reading above on PCI and ISL62771 are good except lgatenb. And ASIC quality did not change, 59.9% just like with old chip which probably mean it's just dumb value from BIOS not an actual quality.