EEVblog Electronics Community Forum
Electronics => Repair => Topic started by: more_stars on March 28, 2020, 10:45:36 pm
-
Hello guys,
My name is Alex and firstly I would like to say how much I will appreciate all the help and effort in diagnosing my problem.
I have a r9 390 nitro gpu that I bought as broken because I really wanted to upgrade my existing hd6850 gpu. I do repair some basic staff and troubleshoot tv's xbox controllers and power bricks but never attempting to repair a pcb with only smd s and so many components.
Firstly I plugged the card in my computer and checked what is doing. The gpu's bios switch powered on but that was it so I took the radiator off and plugged the gpu back in the motherboard. When powering the pc, the main gpu did not heat up so I tried to measure the voltages across the phases. I only came across the 12v inductor that is near the aux vrm but all the vrm's did not display any voltage what so ever. I tested the 3553m mosfet and they all seem to work fine but the 4c10n ware shorted on the drain and source so I tooked them out and I have seen that the mosfet are perfectly fine but the pads are still shorted. I checked with the ohm meter and there is an 0.1 ohm in there betueen D and S. The ohm reading from ground to the other phases are 2.2 ohm for the gpu vrm, 0.2 for aux vrm, 0.2 for mem vrm and 0.1 for 0.98v rail.
Also, I can't take the inductor off from the memory vrm to see if the short is from the memory chips or from the mosfet controller. I have seen some guys using ipa and applying a small voltage to see whether there is a component that is heating up but I don't know where to apply that power and I also don't want to destroy the whole card.
I also have another xfx rx 470 with 8gb that has two thick lines with green and pink throughout the screen. This was given to mt by a friend who used it in mining. Is there any hope for this card?
thank you in advance!
Stay safe!
-
You talk about plugging the GPU into the motherboard but you never mention the two 8 pin power connectors. Did you connect those as well? If so and none of the VRMs are getting the 12V rail I would check around the two power connectors for broken fuses, missing resistors (is this the card they did overclocking with by removing a resistor on the 12V input and shorting it out?),... It would be nice to get an image of the setup as well as the card itself and the top righr corner also.
Another advice. If this is the only way you can afford upgrading (no disrespect ment here), you shouldn't just plug known broken stuff into your only PC and hope it doesn't break you motherboard aswell. Maybe get a cheap setup for testing. It only needs PCIE support so any old 775 socket mb with core 2 duo or similar will do. And when you're certain it works at least hardware wise you can plug it into a better machine. Just my 2 cents.
-
The first time I tested the video card, I dindn't plug the two 8 pin pci-e connectors as I was thinking it will work without those but after some research I did understand that the gpu needs power from all the places so I connected the pci-e connectors as well but still nothing. I see that there are 2 fuses and two 0 ohm resistors near the pci-e conectors and another 0ohm resistor near the bvi video connector.
Also, I was not thinking that this can cause anything to my computer as the worst I could think of is shorting my psu. Actually I have another motherboard laying around with 4gb ddr3 and I will use this one for further testings.
-
Also the full card
-
Oh boy, you are missing 4 chunky components to the left of the power connectors (https://linustechtips.com/main/uploads/monthly_2016_03/03_bigee.jpg.08dd8bc014a70d2f64f637c9d3db546f.jpg). Check out Actually Hardcore Overclocking's video on the R9 390 Nitro. According to him, those are voltage regulators (4C10N ON Semiconductor) for VRAM *gulp*. Let's hope VRAMs didn't get zapped. This still doesn't explain why you don't have 12V on other voltage regulators tho. Did you check the fuses and 0 ohm resistors for continuity?
I would suggest you start by finding a high res image of both sides of the PCB and carefully check if there is anything else missing. I wouldn't be surprised if someone used it as a parts card (or you might be lucky and someone before you already found the issue in the missing parts and only removed those...).
From this point on I'll use the markings from the above video as descriptors. At this point, we are testing the power supply for the card so there's no need to plug the card into the PCI-E slot (I believe it doesn't use the slot voltage supply for any functionality to the right of the VRAM modules, but I might be wrong...).
1) You need to find out if the 12V rail is present at all the voltage regulators of the 6 phases of Vcore (this should be the case whenever the 8-pin plugs are connected) or there is continuity from the 8-pin plug to the VRMs. Each of the 8-pin plugs has 3 +12V rails which should go directly to the Vcore VRMs. If the voltage isn't present you need to check the fuses.
2) Find out if the VRMs (IR3550) are being supplied with a PWM signal by the 6+2 phase controller (IR3567B) as well as Vcc. If there is no Vcc there probably isn't a PWM signal either so we look into a logic voltage regulation part of the board under the 8-pin connectors. If there's no PWM signal but the Vcc is present we move to the phase controller. If the signal is present the core should be running (or at least have the voltage needed). You could check the voltage on the capacitors to the left of the Vcore VRMs.
3) Find the 5V and 3.3V supply. The VRMs use 5V and the phase controller uses 3.3V as Vcc so both should be present on caps around the corresponding chips. If one is missing find the voltage regulator for those and at that point, you've probably located the fault.
EDIT: I just noticed something isn't quite right with your C4139 (bottom left of the card next to the last VRAM chip)
-
The four missing mosfets are because I desoldered them as I said in the initl post. The drain and source ware shorted so I figured that I might desolder those and see if the mosfets are OK or not and after checking, the mosfets are OK but the pads are shorted or they seem to be shorted as the rrzistancecis 0.1 ohm.
The cap at the bottom left is like that because I also desoldered it but not with the hot air station. I will resoldervit properly thus time.
The 0 ohm resistors and fuses are all OK.
I did purchased another broken r9 390 nitro with 30$ but it will arrive Thursday. This way, if they don't have the same problem, I will have another pcb with working parts to check from.
I do think that the problem is with the mosfet driver but I don't know how to test is as I don't have an oscilloscope at hand.
-
My bad, I completely forgot about the detailed part of your description :/ Did you check for the 5V and 3.3V supplies tho (point 3 of the above post)?
-
Ohh I haven't seen the point on your reply. I will check them now and let you know.
Also, thank you very much for this help. I wrote to hardcore overclocking but he blocked me right away :)) I think I was a bit to derailed on explaining.
-
Also if you check the IR3550 datasheet you can see a typical configuration for MOSFETs, coils, caps, etc. for the supply. The only way I could see the MOSFET pads being shorted was if one of the caps around them is bad or, the bad outcome, which is that the VRAMs are shorted. From your description of all the MOSFETs being shorted it, unfortunately, sounds like the latter.
Can you maybe figure out how the 2.2 ohm resistor under the coil is connected to the circuit? If it is inline you could desolder it to see which side is shorting.
In short, there should be two pairs of MOSFETs. Each pair should have one source on 12V rail, one drain on GND and one source and one drain connected together. Where the two connect the coil should also be connected as well as some capacitors.
-
So, this is what I come across:
C658=1.7v
C650=3.2v
C714=0.3v
C4136=12v
C4118=12v
C4037=6v
C327=5v
C4034=12v
C4139=1.7v
C320=12v
At the 6 gpu phases the drain is not shorted with the source and the source have indeed 12v but I don't seem to have ir3550 but ir3553.
The mosfet from the memory VRM are the ones that have the source shorted to drain. Those mosfets are 4c10n.
I don't have an oscilloscope to check the pwm of the driver so is there any way to check the driver or I need to take it out fro the board and see if the shirt is still there. Also, I think that the ir3567b driver is only for the 6 phases for the gpu and I don't know where is the driver for the memory phase.
Also, the inductor is just too big to be desoldered with my basic soldering iron.
-
At this point you have all the power rails I would expect present so if there's nothing wrong with the PWM you should have output from the Vcore VRMs. I would check voltages on all 10 capacitors to the left of the coils (measure on the back of the PCB).
Also when I was talking about the structure of the two mosfet driving the phases I was talking about the VRAM VRMs and comparing them to the Vcore structure.
We should expect a similar structure and as the phase driver is a 6+2 I would expect the 4 VRAM mosfets to just be discreet versions of the Vcore VRM and function somewhat similarly.
Let's first figure out if Vcore is good and we'll take care of VRAMs after :) Too many different things to talk about at the same time.
-
That LEFT regulator showing ZERO volts
is the top primary issue
A know problem in these boards which
should be checked is U10 IC (PCI regulator)
U10 vary across boards usually you may
find those in pics. Both should be checked
until working properly so VCore VRM will be next
-
After managing to de solder one inductor from the Vcore, I was able to see that the voltage got a bit higher from 0v to 0.22v
Also, after investigating the pcb I was unable to find the u10 ic. All I could find was u3, u4010, u1704, u350, u4017, u4018, u4, u2, u700, u201 and u1000.
Near the PCI are two big ic s apl5930 and 25l0010 be1611
-
I also took a photo at that side of the biard
-
That LEFT regulator showing ZERO volts
is the top primary issue
A know problem in these boards which
should be checked is U10 IC (PCI regulator)
U10 vary across boards usually you may
find those in pics. Both should be checked
until working properly so VCore VRM will be next
Do you think (or know even) that the PWM is being suppressed/disabled because some of the PCIE side of things aren't working?
To the OP: Can you take a closer image of the left side of the pcb around where you measured 0V as well as one of just the board left of the VRAMs is visible?
-
Also, the card is not recognised by the computer at all because with the card inserted, the computer boots up normally with the normal beep sound.
I also attached three more photos
-
Do you think (or know even) that the PWM is being suppressed/disabled because some of the PCIE side of things aren't working?
nah no clue.
Usually I go every test possible out board to check
suspected too low impedance
But from the photo only? That zero regulator
may be preventing a dead shorted U10 to burn
Fairly common issue.
Unless the mosfets are already toasted
-
I also have two of those rx470 8gb from xfx. One is only for parts and the other one have some artefacts but knowing that they ware used in mining, I thing that the problem is with some of the gddr5 memory chips.
If you think that I have more chances to repair this rx470 one, then, I will go with this one but also I would like to repair the r9 one as I already purchased another one for parts.
-
Ok, let's say the right side of the card is working (except the VRAM MOSFETs) and the PWM is being disabled/not enabled by the left side of the card.
At this point, I would suggest measuring the voltages on the three silver capacitors on the left side of the board. According to their rating, I'd expect 12V on the 16V one and 1.something V on the other two.
-
On the 16v cap I can measure 12v but nothing on the other 3v caps.
-
On the 16v cap I can measure 12v but nothing on the other 3v caps.
Ok at this point just one image is missing. Could you post an image of the backside of the PCB around these caps? At this point, there is definitely a power supply problem somewhere on the card. We need to find which voltage regulator should be feeding the two caps that have no voltage on them. Please also measure if these two rails are shorted to ground (that would be another reason for 0V.
EDIT: NVM it's definitely the other IR3553 that is right there. Please look at the pinout of the IR3553 (https://www.infineon.com/dgdl/ir3553.pdf?fileId=5546d462533600a4015355cd94ee1767) and confirm that it has 12V rail present on Vin, 5V Vcc present, and the SW isn't shorted to ground.
-
The rail is also shorted to the ground for those caps.
-
Any chance you could pull the coil and remeasure if the rails are still shorted?
Also the big chip (78M05G) you can see on the back is a 5V regulator supplying Vcc to the IR3553 it should have 12V on one side, ground in the middle and 5V on the other side of it.
-
The coil is just to big and I am afraid that if I put to much heat into the board, I will brake something. If you know haw to do it, tell me and I'm gonna try
-
The coil is just to big and I am afraid that if I put to much heat into the board, I will brake something. If you know haw to do it, tell me and I'm gonna try
It should be easy enough with a reasonably large tip. First, you need to add some new solder (preferably with Pb) to lower the melting point. Then you use the solder sucker to pull away as much of the solder as possible. If done properly you should be able to clean it enough to pull the coil off without using the iron in the end.
Also check the end of my previous post.
-
I removed the coil with the help of both the iron and the hot air and now I can see that the caps are still shorted to ground and the aux vrm does give 0.22v and the 5v regulator works fine as there is 12vvan5v at the sides.
-
So showing low resistance 0.1 ohms? which caps?
-
At this point, I would remove the IR3553. If that's not causing the short then one of the caps is shorting. Nothing else is there as far as I can see.
-
So showing low resistance 0.1 ohms? which caps?
Yes 0.1ohm on the filtering caps of the aux vrm
-
So showing low resistance 0.1 ohms? which caps?
Yes 0.1ohm on the filtering caps of the aux vrm
Can you show exactly where you measured the 0.1 ohm?
-
At this point, I would remove the IR3553. If that's not causing the short then one of the caps is shorting. Nothing else is there as far as I can see.
If the gpu itself was shorted, it would heat up right?
I don't think that the mosfet is shorted as there us no short between drain and source but is a resistance of 5.8 kohm.
-
Those 2 electrolitic chaps and 4 ceramic ones.
-
Those 2 electrolitic chaps and 4 ceramic ones.
And you are sure that your ohms measurements are valid? If that's the case something has to be shorted as a voltage rail shouldn't have a 0.1 Ohms resistance to ground. Something also had to get fried at some point (I think it's the IR3553) if nothing is heating up but the voltage rail isn't present. You never confirmed if the IR3553 has the 12V present.
-
The 12v is present at the IR3553 and befor removing it, the voltage at the coil was 0.22v but now that I have removed it, the voltage us 0.03v. At this point I think that the problem is with the control ic s.
Also let me know what you guys think about the rx470 gpu that I posted earlier.
-
I don't see how it would be a problem with the control as you still have a short... I would still remove the IR 3553 and the caps until the short clears. Basically, you have isolated a very small area of the board between the IR3553 and the coil which only has capacitors on it. As there is a short here it's either caused by the IR3553 or the caps.
-
I don't see how it would be a problem with the control as you still have a short... I would still remove the IR 3553 and the caps until the short clears. Basically, you have isolated a very small area of the board between the IR3553 and the coil which only has capacitors on it. As there is a short here it's either caused by the IR3553 or the caps.
I have desoldered the IR3553 and the shirt is still there.
-
I don't see how it would be a problem with the control as you still have a short... I would still remove the IR 3553 and the caps until the short clears. Basically, you have isolated a very small area of the board between the IR3553 and the coil which only has capacitors on it. As there is a short here it's either caused by the IR3553 or the caps.
I have desoldered the IR3553 and the shirt is still there.
Now do the circular caps.
-
Done that and the shirt is still there
-
I know this seems dumb but I'm 100% sure there is something wrong here. I would try removing all the small caps that are connected to this one pad one by one and if the short isn't gone just resolder it back. Go around on both sides and something has to give at this point.
If you feel like it, you could try visually figuring out if this rail goes anywhere else on the board but I'm almost certain that it shouldn't leave before going through the coil.
EDIT: Hold up, is the R646 connected to this pad? If so measure it and if it shows a short try removing it.
-
I did remove anithing that I was thinking it might been shorted but there is nothing that have been changed.
I resoldered the IR3553 but now instead of the 0.22v I get 0.10v
At this moment I think that I will wait until I get the other gpu and see I the simphtoms are similar.
Also, do you think that I have any chance to repair the rx470? I posted some pictures earlyer and I was wandering if this one is easyer to repair?
-
I'll be honest... Lines across the screen can't be a good sign... But I don't know enough about it to help you out. You could try different types of outputs (if there are any - hdmi, dvi, vga). If they all exhibit the same behavior there is something wrong with the core imo.
You definitely can't probe it like you can the r9. The only thing you could look at are the voltage rails. If those are nominal then something's up with the core. If some are outside spec or have high ripple then it's the supply rails. But without a scope, this will be hard to figure out.
One thing before we give up on the R9 just to make sure. Which pad of where the coil used to be is shorting to ground?
-
I will attach a photo with the pads.
-
Here
-
Without knowing where the rail goes there's no way you're going to track this short down... You can check the capacitors behind the core. If any of them have ground on both sides you're most likely done. Anyway, I'm not sure I can help you any further. If you have any ideas of your own feel free to share :)
-
All the caps on the core give a reading of 0.2ohm so the core might be OK, I hope.
I really appreciate the help and the interes in my situation. You guys are awesome!
If anyone have any idea on what the problem might be, please let me know.
-
Please don't start posting multiple cards faults in the same thread! It's extremely confusing and annoying reading, start a new thread for a new repair.
According to this video below that Aux VRM section eventually goes to the GPU memory controller. There may be additional components on the GPU side or rear it feeds.
To do that low power short finding troubleshooting technique you mentioned you need to inject a low voltage on that rail and use an adjustable current limiting power supply. Or some people use a battery or simple DC voltage source and use a current limiting resistor. The idea is the low voltage and limited current means the short can only dissipate the maximum power you have predetermined and heat up rather than fuse.
Rasz and Wraper forum members here seemed to have worked on a lot of GPUs so I have messaged them for their comments.
https://www.youtube.com/watch?v=U_J8uEcGCiM (https://www.youtube.com/watch?v=U_J8uEcGCiM)
-
Just to confirm that you are looking at an actual issue. Was there no voltage on that AUX VRM output to ground and then you removed the inductor and then the VRM was suddenly showing voltage to ground again? If so that looks like a short on that VRMs output, it can mean the VRM is faulty only under load or/and a component or trace short somewhere on the output of that VRM.
-
If upper MOSFET was short, some or all RAM chips should be dead because of overvoltage. GPU memory controller might be bust as well, but they are more robust than RAM IME. I personally would attach power to RAM from external PSU and watch with IR camera. Evaporating alcohol trick will not work because heat dissipation on dead chips will be minimal due to very low resistance. They barely heat up a few degrees maximum even if you push 5A into the power rail.
-
This is what I was thinking as well that the memory chips might have gone bad as the ram vrm was shorted but after taking the mosfet out, I was able to see that they are fine and only the pads are shorted. Thus if I try to take out the inductor from that rail, it will tell me if the short is on the ram chips or at the driver part, right?
Also, I don't have a psu with curent limiting but I have some step up and step down convertor one of which actually has a curent limiting setup. Can I use that one whit a pc psu and insert 2v at 3A?
And yes, after removing the inductor on the aux vrm, from ground there was a voltage of 0.22~0.10. Also I removed one inductor fro the core vrm and as well the voltage raised from 0v to 0.22v
-
I removed the inductor from the memory vrm and the short between drain and source is gone but there is 0.2 ohm from ground to the other side of the inductor
-
I removed the inductor from the memory vrm and the short between drain and source is gone but there is 0.2 ohm from ground to the other side of the inductor
I think at this point throw it in the bin or resolder everything on as nice as you can and sell it to the next fool (please don't do this :)) You might as well desolder the RAM modules one by one until the short clears and then you can rest peacefully as there was nothing you could've done (if you don't have replacement RAM and reballing setup) :)
-
Anyway if you might as well check and rebuild the mem fets and and aux vrm so you can at least confirm you have good voltage and high resistance on the outputs up to the missing inductors.
If you still have a short on the memory side of the inductor the only thing you can do is remove caps then memory till the short goes away unless you have a thermal camera or want to power them externally like Wraper suggested. There is the high chance someone killed them rather than a graceful failure.
-
Ohh, thank you guys for all the helpful thoughts.
I will wait for the other r9 card and see if the problem is the same. Maybe I will use this car for spare parts. I payd 20$ for this one and 25$ for the one that will come today.
-
Anyway if you might as well check and rebuild the mem fets and and aux vrm so you can at least confirm you have good voltage and high resistance on the outputs up to the missing inductors.
You cannot do that. Inductor is part of VRM with feedback voltage taken after inductor.
-
Anyway if you might as well check and rebuild the mem fets and and aux vrm so you can at least confirm you have good voltage and high resistance on the outputs up to the missing inductors.
You cannot do that. Inductor is part of VRM with feedback voltage taken after inductor.
Both the mem supply and the aux/memory controller supply? What is considered the output after the feedback then, the inductor side filtering caps or is it feedback all the way from the memory or gpu itself?
Do you have any documentation or schematics for modern cards? I know they put out boardviews for phones and macs is there anything for gpus?
-
Ohh, thank you guys for all the helpful thoughts.
I will wait for the other r9 card and see if the problem is the same. Maybe I will use this car for spare parts. I payd 20$ for this one and 25$ for the one that will come today.
Here is an example (12m10s) of using a current limited power supply to limit the power into a short. He chose 1.2V as it was likely the lowest supply voltage on the board anyway. At 2A that is about a 0.25W into a short which was plenty to get that component warmed up.
https://www.youtube.com/watch?v=2DyEoTP303I (https://www.youtube.com/watch?v=2DyEoTP303I)
-
Perfect, then I will try that as soon as I get the other card as well. Now I will focus on seeing if I can repair the rx470 as I opened another ticket for that.
-
Cheers I'll check it out. You can say thanks to any post you find helpful by clicking on the "say thanks" button located on every post towards the top right.
-
Anyway if you might as well check and rebuild the mem fets and and aux vrm so you can at least confirm you have good voltage and high resistance on the outputs up to the missing inductors.
You cannot do that. Inductor is part of VRM with feedback voltage taken after inductor.
Both the mem supply and the aux/memory controller supply? What is considered the output after the feedback then, the inductor side filtering caps or is it feedback all the way from the memory or gpu itself?
Do you have any documentation or schematics for modern cards? I know they put out boardviews for phones and macs is there anything for gpus?
It's simply how buck converter works. It's not a linear regulator, it stores energy inside the inductor.
What is considered the output after the feedback then, the inductor side filtering caps or is it feedback all the way from the memory or gpu itself?
Wherever it's sensed, it's a single heavy power plane. Most likely connected somewhere nearby to first smoothing capacitors after inductor.
-
Yeah I know how the inductor and feedback works I was just having a blond moment while I was thinking about the best way of isolating the supply from the short and totally forgetting about the feedback on the output as well being shorted.
-
Finally the second board arrived and after some checks, all the vrm seems to be fine but the board does not power and there is no video out.
The ohm readings from ground to coil below
Aux vrm 42 ohm
Ram vrm 118 ohm
Core vrm 21 ohm
0.98v rail 2 ohm
I think that the problem with this board is the ram chips as they are covered in liquid from the thermal pads. Is thare a way to fix those or do I need to replace them?
-
I think that the problem with this board is the ram chips as they are covered in liquid from the thermal pads.
It's a non issue. Like a half of graphics cards are like this.
-
Hey Alex,
i have a short in C4136 and C4135 so i need a new cap. Do you or anybody else know where i can buy caps like this?
I don't know the specs so i can't search. Can someone help me please?
Thank you in advance,
Patrick
-
Hello i have problème with my r9 390 sapphire nitro.
Can you help me?
Thank you
I have Black screen and no boot.
I have tested all Voltage and i have no Voltage to vmem.
And two capacitor.
Thank you
After other test i connect power supply 2 mosfet 4c10n have 12v
And 2 mosfet 4c10n 0v fuse 0v capacitor 0v see picture.
Thank you