This is very interesting case. Let's try to solve it together because one day it can catch me or another forum member.
https://media.digikey.com/pdf/Data%20Sh ... P81022.pdf

Page 25:

So there is a current mirror inside the chip. Check ROSC pin and its resistor. Maybe it is open or shorted? Just lets start from the beginning. Any digital chip needs a clock source. What would happen if we have a short on this pin or open circuit? How to test if internal oscillator is running or not running?

Page 28:

So it should have initial slow ramp rate of voltage. It should start with a very narrow pulses to inductors in order to make a soft start. There is something wrong with clocking of the chip. Maybe the chip does not get power on reset of some kind. I don't know.

Let's dive deeper into the PDF... What is the power on reset procedure and what should happen? Let me read it...

Page 3:
Table with pin description:

What will happen with PWM signals when you pull EN to ground in some way? Just force disable the chip. They should be tristated in my opinion..

Page 3: table and the block diagram

Maybe the slope is too fast when capacitors are not charged yet? Maybe not.

In general add a 0 ohm resistor 0603 or even 0402 or a small fuse when testing in series with power supply. You will not lose transistors.
Posted after 1 hour 10 minutes 14 seconds:

I have such card now Asus Rx580.
Never repaired, never in a mining rig, very clean inside as new, doesn't look like cleaned. The card works, it gives picture, tserver can load all phases are good, pulses to inductors are correct, tserver memory test gives "PASS" but a little load on the GPU and there is a black screen. Even when I increase GPU clock from tserver from the command line it immediately becomes black and all voltages are gone. Only 5V stays there and 3.3V from the LDO to the PWM controller.
What I noticed on the oscilloscope - the PWM duty cycle does not increase and stays at 7.6-7.8% duty cycle and the VCORE voltage is around 0.8V and does not go up. The chip is DIGI+ ASP1300.
Maybe no communication between GPU and the PWM. I will investigate it...

If you speak English this is the best you can get:
https://www.youtube.com/watch?v=6_QooqL3-tI

One of my faulty cards doesn't have any adjustment of the PWM, maybe the GPU is dead?
Where could I download the complete IR3537B PDF from?

I had some time to rethink this case again because I have similar problem on the card which was never dead. It started to behave like that from the owner description.
I analyzed schematic of MSI card from the repository here. I have Asus but looks like with the same controller remarked by Asus to Digi+.

These pins form i2c bus
24 - SM_DIO
25 - SM_CLK

These pins form SVI serial bus
16 - VDDIO (I/O power supply for logic levels for 18 and 19)
18 - SV_CLK
19 - SV_DIO


Maybe something is blocking the bus! Failed capacitor, there are some pico farad caps, maybe one or another device on the bus, maybe pull-up resistors or maybe just the master's pins are dead. Follow labels on the schematic starting from the PWM controller. I have two identical cards and I will investigate it because one is faulty exactly the same way.
Mine problem is not a 3d engine I think because it happens immediately after I increase a core clock from the command line of tserver in 2d mode in linux when using tserver. It also does not react on commands to control the speed of the fan.

About SVI page 26:
https://www.renesas.com/us/en/document/ ... -datasheet

There must be a sequence of events.

1. PWM chip must receive positive enable
2. It reads PRE_PWR_GOOD 2 bits from data and clock as a logic levels
3. It responds with power-good of some kind
4. It receives back powerok from the GPU
5. Now it starts to actively monitor commands from the GPU to increase/decrease voltage

If this fails the card can't adjust power to the clock frequency and load, simple.

Thanks for your support! It is very important.

Perhaps an important point that can give you an idea: according to the history of the card owner, the card died after the 12V wire and the ground wire on the additional power were shorted. The card was turned on at that moment. After that, the card was no longer worked. Earlier, by some miracle, I managed to launch it on one power phase and see a picture, which means the GPU and memory survived.


I checked the resistor on this line, it is 71kOhm which is the norm for this resistor, so on working card there is the same 71kOhm resistor here.
The circuit is not broken between the pin and the resistor, then the resistor goes to ground. In this moment, everything is fine. But on ROSC pin i got 0v, i don't know is there ok and maybe I should find the same resistor and replace it just in case because a few days ago I had another RX570 card, there was no power except 5v, 3.3v and 12v. I found out that the Enable signal does not come to the PWM controller of the PCI-E bus, therefore 0.95v does not appear and with it all other power supplies. The Enable signal goes there directly from the 3.3v bus through a 10kΩ resistor. I checked the lines for damage - there was no damage, then I unsoldered out and checked the resistor - it was in absolute order! Resistance 10kOhm, no short circuits, but in fact, 3.3V goes to it, and at the output of the resistor - 14mV instead of 3.3v. I replaced the resistor with the same one and the card working fine. My point is that even a good resistor that shows the correct resistance can be damaged. How exactly - I do not understand. So, I lead to the fact that if on my card the problem is somewhere in small resistors and capacitors on the power/PWM circuits, then I will have to replace everything that is there in order to understand.
The second point and my fear is if there is any problem in the PCB layers. I will probably never find this problem.

But let's not despair for now :)


My card got logical TRUE on PWM's EN (3.3v). But there is no PWROK and PWRGD. Can you suggest why, what's missing from the PWM controller?
There is some measurements of PWM's pins:
1 - PWROK: 0V
8 - VDDNB_PWRGD: 2MV
9 - VDD_PWRGD: 14MV
10 - EN: 3.2V
11 - VCC: 5V
12 - ROSC: 0V
13 - VRMP: 11.7V
51 - VDD: 0V

do you suggest adding a zero resistor to the PWM-in line before the drmos?


A month ago i got Asus RX470 Mining Edition, i've repaired the GPU power supply and oscilloscope shows perfect picture, card goes live. But if you start the Furmark, there is immediately black screen but it seems to be rendering the test, because fan and temperature was high. But if you start mining test - its working ok. i've tested it for 5 days and everything is fine with mining test, but some graphics stresst test - card goes blackscreen. :)

No, I suggest 0hm to the 12V to protect drmoses. Put fuse on the yellow cables or zero ohm somewhere temporary, but the small one. You will not loose transistors. I have a cable with small fuse specially prepared. I use 1-3A just in case or power it from protected laboratory psu with defensive mode.

PowerOK comes from the GPU when the PWM controller gives PowerGOOD. So when you don't have running VRM you will not have both signals.
At least it is exactly written in PDF.

The rest in a moment....

ROSC szhould be <= 2V but not 0. Even if it is current limited it is still 2V and 71k. 71k is a very big resistor so it should be some voltage here, not zero.

Oh, okay, i get it. This card has fuses on both 12v lines, but them 2x10A on additional power and 10A on PCI-E 12v. Put a little fuse make sense, thanks.


Thank you, I suspected that this was related and at the moment, due to the disabled PWM, it does not make sense.
---------- Добавлено спустя 1 минуту 9 секунд: ----------

i will try to find 71kOhm and replace it, maybe this will be the answer

If it measures 71k it is 71k. From any ground to the pin.

Yes. But as i say early about another card:

I've replaced the 10k resistor (measured as 10k as well) on the another 10k resistor and the EN signal up to 3.3v and the card came to life. why her own resistor was sagging voltage up to 14mV, I don't know. Never mind

I have read it again. Nowhere is written how this 12 - ROSC should behave but I think it is pure analog regulator and must be present immediately after powering on the card. This looks like an analog signal and the dropout on this resistor is measured or there is any other kind of measurement but should start after powering the chip. And regardless of enable state.
No voltage here so internal regulator doesn't provide 2V so the chip is ded.
Remove doublers and leave only this chip and check other card with the same PWM if you have one.
When you remove 71k measure it without first.

Hello.
So, when resistor placed on board its 71kOhm on ROSC pin and 0.2Ohm on ground side of the resistor. And no voltage on ROSC pin.
I've unsoldered the resistor, checked it - its 71kOhm, resistance on ROSC pin without resistor is 6.2MOhm.
So, i turned on the card without 71kOhm resistor and there is about 2.3-2.5v on ROSC pin! The chip's internal regulator provide ~2v

I will try to find another 71kOhm resistor and replace it on board then i'll check the voltage on ROSC pin with the new resistor

Strange. This has to be a resistor for setting the frequency. They say:

Whatever it means the 71k connected to ground should not interfere with the voltage if it is a constant voltage source. 50-100k is nothing compared to the 2V voltage source. Maybe the circuit is different or it is not a voltage source but 2uA or 2nA current source. Maybe it is disinformation.
Maybe there is anybody with working such chip and can measure?

I tested something different today on another card. I removed 2 inductors from the memory VRM and watched on the scope what is on the output of a good and working VRM. After powering up there are 20-30 very short, short as a 1-2 pixels, 12V pulses to the inductor and they stop. Looks like VRM tests for a short or open circuit first.
Some VRMs get feedback from both sides of the inductor. As you may know the current does not flow through inductor immediately but first the magnetic field must build up until the current stops rising or to the saturation of the inductor.
If there is a failed inductor, with a short between turns it can give unpredictable results. Maybe these inductors already overheated and the wire insulation is broken so the inductance is smaller or even none and such element can't accumulate energy in the magnetic field and the output current is not skewed in phase to the voltage. It requires further investigation. Maybe drmos or controller require/expect a certain phase shift between voltage and current on the inductor? Shorted inductor will work as a short circuit, not as an energy storage and a current delay.
In general the inductance can be measured and should be for example R22 = 0.22uH.
You can measure it by oscilloscope and a sine wave generator. Nothing else is required.
Without inductance the drmos pushes current to the empty capacitor and the current it infinite at the beginning so maybe it is a cause of failure of the one phase, a short in the inductor? You will not measure inductor's resistance, you must validate component's inductance. In pure capacitive circuit the current leads voltage and maybe it kills the transistor.

yesterday I got exactly the same rx470 with the same board and elements and this card works great. Now I can compare measurements.


About the ROSC pin: I compared, and the working card also has 0v on this pin.


I dont have sine wave generator, but i have the same four R15 inductors. So, I replaced the inductors and this gave nothing.

I soldered one drmos for the first phase of the power supply, which is powered by PCI-E, the card does not start. all power lines are working fine except for the GPU power. PWM in = 2.5v in phase with soldered drmos, and 5v on three phases without drmos. and flat line on the osciloscope anywhere here. and doublers have got no pulsations on oscilloscope too

I think what else I can check.

-compare BIOS with the working card
-compare pulsations on SVI2 data/clock

SVI2 has nothing to do with it. Even if you don't have pulses on both data and clk then there is still default voltage output set by the power on states of clock and data treated as logic levels

00 800mV
01 900mV
10 1.000V
11 1.100V

Whatever appears on these pins will set something, at least should set, when the PWM chip starts. Later after some handshakes, some chipselects, power_goods, power_ok in return the SVI2 becomes a transmission line. Until then it only acts as 2 wire GPIO for initial setup.

---

Posted after 6 minutes 34 seconds:


Now you may have dead PWM or dead doublers But it was running on one phase, wasn't it? Something happened that it stopped working.

If i understand correctly he gets powerok, afterwards its SVI2 in charge to raise voltage with load increase.
A fact that its failing at slighthly higher load and PWM having straight 5v on several outputs should be a reason enough to check whether GPU is requesting VID properly

Yes, POWER_OK is the last signal coming from a GPU -> to the PWM chip, informing that GPU acknowledged good power status nad GPU will start controlling the power itself. After that GPU starts sending serial commands to the PWM chip and/or to a fan controllers and so on.

Again.

1. Logic gates and small transistors set enable signals to the PWM, small VRMs and GPU
2. PWM controller and probably small VRMs assert POWER_GOOD to some additional logic and finally it goes to the GPU
3. GPU answers by asserting PWR_OK
4. GPU starts sending commands to the VRMs/PWM controllers.

Any VRM at any time can disable it's POWER_GOOD signal informing other chips and GPU about power failure.

BIOS is ok. This card got 4 BIOS and it was run on one phase early.


Yes, i may have this. I'm waiting for parcel from china with 2N7002DW chips (K72 marking doublers that used on this board), i'll replace all of them. The PWM controller could die during one of the tests when the drmos burned out. I will replace it too, I have 4 more new PWM controllers in stock.


Yes. Previously, I managed to start it on one of the four power phases. At the same time, the oscilloscope showed an atypical, "wrong" picture of the signal on this drmos. and even so the video card gave me an image.

I checked the resistances on all elements (resistors, capacitors) on the power phase, I did not find anything particularly strange, except that in a couple of places there was a slight deviation of about 1-5% from the working video card. I think this is a normal deviation. I have never met two identical video cards with exactly the same values ​​on the elements. Maybe I'm wrong and it matters.

Now I will look for K72 doublers on donor video cards, perhaps I will find and be able to test the card's performance with them until I received the package from China.

I will write back here with results. Thanks.

2N7002DW can also be found on laptop motherboards

This is always a problem with parts. They can't be purchased for tomorrow from the EU supplier. All takes weeks from China. This is very painful.

Thanks.


Agree.
Do you know if there will be generations on doublers if no drmoses are installed? or the PWM controller will give PWM signals only when the drmos is installed on the power phase?

It should be a small burst of pulses and they stop. On the scope, very narrow pwm pulses something like sampling if the circuit is in order or not.

Hello. Got some news here.

So, i've received parcel with K72 (but this chinese chips marked 72K and got a dot under the "7").

I've soldered the new PMW controller and ONE doubler on the first phase.


And checked this. There is nothing at all. I connected the oscilloscope to the PWM IN contact and looked at the moment the video card was started - there was not even the slightest signal. And there was nothing on the doubler.

Then I soldered the drmos and damn it started only with the first phase! And one more important point: I have only one drmos and one doubler soldered. In the absence of drmoses, the PWM controller does not give any signals at all, if you solder at least one DRMOS, then the PWM controller gives a signal for all 4 phases, there is a signal even at the inputs to the doublers in the absence of the doublers themselves. there is a signal even where three doublers are not soldered.

I don't know what exactly helped. I will solder the rest of the doublers and drmoses. unfortunately, I have only three drmos left - I have no right for the error.
---------- Добавлено спустя 41 минуту 9 секунд: ----------
So now i've soldered the second phase doubler and drmos, and its burned out the first's phase drmos. i dont know.

Damn IT !!!
You were supposed to solder 0603 or 0402 R0 resistors to save output transistors.
I tested impulses on a good VRM for memory after removing inductors and pulses were present.

I was too happy to see the signals on the oscilloscope, so I forgot about the resistor.

In any case, it is not at all clear to me why the video card burns the drmos. At the same time, it burned down in the first phase, which worked without problems.

I also receive cards fatally ill. During the last days most cards with dead GPUs, probably. Nothin, nothing wrong with voltages, transients, chip selects, PCIe signals, all looks perfect even mats works on some of them but they don't accept drivers under windows.

I listen the Russian pop music to this repairs. I felled in love with Russian music hehe Hey Russian friends is Artik & Asti popular? I love their songs.

---

Posted after 19 minutes 50 seconds:

I used to write firmware in assemby, C and VHDL for almost the whole of my live and if I designed such circuit I would not allow to turn on the top side mosfet before several checks and then very slowly. Slowly I mean 1/1024 of a duty cycle and test what is going on on the analog feedback. If no change within 2-3 more steps I should assume the circuit is not sending me feedback and I must stop.

Such circuit may fail when the microprocessor clock stops or when the microprocessor doesn't do anything but there should be additional analog circuit to cope with dead PWM controller. It is insane. Half of these cards come with catastrophic failures.
If the tristated signal disables drmos so there should be a divider with internal reference voltage inside the drmos to prevent opening when it is disconnected from a driving signal or signal is not switching properly. Input of PWN signal should never go high or low on its own.

Take a picture of your card we will discuss is anyway for future readers.

hello


sometimes I also get cards with dead gpu. Can't do nothing with them


Yes, many people like Artik & Asti, especially fans of this genre. Try to listen to "Dzetta", "Noize MC", "Drummatix" maybe you will like them too.

I feel like I found a friend on a tech forum dedicated to electronics repair


I don’t know why the manufacturer didn’t foresee such things. Perhaps because the goal was to reduce the cost of production


Here is the hero of the occasion:



Sorry for quality, pics taken with phone cam.

Today i've soldered only one doubler and only one drmos.
And i got a pretty pic on this phase:


Yes, this is a pretty pic, because i got a cheap chinese oscilloscope and the pic like that - is the good pic of working phases on all my working cards.)
Card is working fine on windows, all parameters are normal:







And to be honest, I'm afraid to solder the rest of the doublers and drmoses.
And at the moment I have only one live drmos, I will have to wait for another package from China.


There is two 10A fuses here. What do you think about if i temporary solder one 0ohm resistor instead of these fuses? To save the drmoses like you said

No, I have a prepared cable with such small glass fuses like that: https://www.aam.com.pk/shop/6a-5x20mm-glass-fuse/. You can buy 3 sockets for them and each yellow cable through it. 1A on each wire or less. They are so cheap.

I will ask a stupid question, do you use exactly the same drmoses, the same symbol as they were originally? Some replacements are different, one leg or more are different.

Thanks, i'll try



Yes, exacltly the same.
FDMF3035

Sometimes the batch or production date is different, but here you can't guess what was sent from China, then I use it. previously there were no problems with different production dates on the same drmos model (FDMF3035)

You checked this on a working card?

And on PCIe phase you have pulses or straight line when it works?

Yes, its okay, the same values on the working card.


Didn't understand the question a bit. The working video card has pulses at all phases.

At the moment (in my post above) a new PWM controller, a new doubler and a new drmos for the first phase are installed on the card. And it works well on one phase, pulsations is good on drmos and on a doubler. Now I am waiting for a package from China with new FDMF3035 drmoses to continue the installation.

Nevermind, i misunderstood there is straight line on pci phase too.

You have identical working card, there is one choice:
When drmoses come from china, solder them but dont start the card. Compare resistances of both cards element by element. If hardware fault there will be difference somewhere