GTX970 short circuit

[andrzejj]

Hello.
Im trying to fix GTX 970 phantom 4gb
There is short circuit after the two phase regulator (i desoldered inductors and checked), im pretty sure its GDDR chips supply voltage.
Also there is short circuit on the ultra-small capacitors on the GPU chip, so I guess it somehow powers the GPU too.
I connected 35W 12V light bulb in series with this section, powered with 12v, bulb was lighting, no device on the PCB gets hot, I have no idea where to look for broken element.
There is only 2 electrolytic caps on that line and 1 tantalum, all of them are working.
Thanks in advance for your help
Andrzej

[amyk]

Modern low-voltage high-current ICs can look like dead shorts - keep in mind the core voltage of these GPUs is only ~1V, but current draw is >100A, meaning effective resistance is sub-ohm.

[andrzejj]

Of course you are right, but in this case i found on some russian site that the load resistance of bigger regulator should be around 3ohm(in mine its same) and smaller around 20 ohm(in mine load resistance 0ohm).
Also i read that 1080ti has under 0.1ohm resistance.
Anyway, I don't have rework station so I cant desolder GPU/GDDR chips to check if maybe some of it is a problem, do you have any clues what may cause that short circuit if not them?
Here are photos: https://imgur.com/a/7UCTDCH

[Rasz]

GPU is not shorted, its just low low resistance

0 ohm on Vram rail does seem bad. I dont quite follow 12V bulb experiment, I dont think you put 12V on the ram rail  so where was it? in front of ram DC-DC? that would mean shorted converter, not ram, in front of cards power socket? that would be normal, this card does take at least 30W just to power on.
if there is a short then how do you know caps are ok, did you pull them off?
Vram should be 1.5V on this card.

random video breaking down modern gpu power supply:

[andrzejj]

Hello and thanks for reply.
Yes, i pulled off caps and coils, and the short was on load side. I checked capacitors and they are ok.
I connected 12v to the SC357 through 12v 35W bulb. I guess its Vram, and its shorted.  I was hoping to get the bad element hot, although i know i could damage board if the shortage would evaporate for instance. Anyway, its still there, so i doubt i damaged anything doing that.
photos: https://imgur.com/a/PoehUU8
Parts with red X- shorted(many other ceramic caps are shorted too, i just haven't marked them)
Parts with OK checked and in good condition (at least if the red parts with the Q on them should measure 0ohm, I guess they are fuses)

[Rasz]

wait, so you DID put 12V directly on Vram rail, you crazy crazy man if it wasnt bad before it sure is now
have you tried pumping current limited 1.5V and looking for shorts with ipa?

[andrzejj]

There was no 12V on Vram as it is shorted, so there was like 0.01V/2A on Vram and 11.99v/2A on the bulb
I know its really ghetto, but IMHO it worked as much as regulated 1.5V (I guess current limiter would work and voltage would go to the lowest possible) would work and shouldn't do any harm in actual state.
I went from 2 bulbs in series with 5V to one bulb with 12V checking voltage that Vram sees, it was always near-zero value.
I agree it would damage parts if there was no shortage.
Yes, I poured some acetone on the board with my bulb and 12V connected, found no heat spots, the shortage must be very low resistance.
So i have question, is Vram directly connected to the GPU too, or the shortage must be on GDDR chips itself or some ceramic capacitor?
Im wondering about desoldering ceramic caps one by one and checking them.

[andrzejj]

Thanks guys for your reply- i kinda solved my problem, but the card is still dead.
Shortage was on the GDDR chip, i desoldered them in oven.
Sadly i damaged one pad, its C13, is it possible that it is NC, as there is no via in that place and i dont think i see any trace going from it

[Rasz]

I hereby pronounce you ghetto lord!

https://www.micron.com/~/media/documents/products/data-sheet/dram/gddr/4gb_gddr5_sgram_brief.pdf
C13 is a data pin edit: its CRC/data strobe, latter is required to work
its time for microscope and needle digging


also https://www.techinferno.com/index.php?/forums/topic/9021-hardware-mod-gtx980m-hynix-to-samsung-memory-swap/

[andrzejj]

Yep, I think repair of the pad may be possible.
Sooo 0.02mm copper wire, epoxy underneath it and UV soldermask over it is the right way to go?
Am I going to be able to access VIA hole after scraping off soldermask?
Next step: finding broken chip and manual reballing.
All GND and VCC pads on the GDDR chips should be connected internally, right? I have tested all chips and they all give 150ohm between A1[Vssq] and D1[Vddq], although no more short on the PCB, so at least one of them must be somehow dead.
https://imgur.com/a/ze6BWdQ
@edit
Repair done, copper wire from USB cable, its really thin, dont know exact diameter, but i guess its thinner than hair.
Soldered to the via and glued with epoxy on the other side.
So you say it should work?(if I add soldermask)
Do you know anything about GDDR chips from aliexpress? I've read that 50% of gpu chips are faulty.

Thanks for the links

[Rasz]

I used to work in laptop sweatshop, we stopped using alliexpress GPUs hmm I want to say 5 years ago, it was never ending string of fake 'brand new' ones. Dont know about ram, those guys on techinferno might know more.

I have a bad feeling about not finding shorted ram. Using oven might of been your doom. Whole bake your card bro myth strikes again, it could be one of the small ceramic caps after all, was shorted before baking, heat relieved internal stress and now its pretending to be ok.

What is the cheapest GPU using GDDR5? hmm GTX 750 uses 4Gbit(same) gddr5 (slower) ram chips
https://prohardver.hu/teszt/nvidia_geforce_gtx_750_ti_itt_a_maxwell/evga_geforce_gtx_750_ti_ftw.html
650 uses 2Gbit ones. Maybe find cheap used one just to test your ram on.
and borrow/steal hotair, or hit nearest hackerspace.

[andrzejj]

I have HD5770 that is using K4G10325FG, you think I can swap them too?
Hotair would be a great tool, although I have no money for it now, Ill ask my friend if he has one.
Yep, I'm thinking the same with baking, is there possiblity GPU was shorted and now it seems ok? Or if the GDDR's are ok it must have been one of ceramic caps?
If yes, than card will be working it should fail sooner or later, and I will find it after another failure.

[Mr. Scram]

Modern low-voltage high-current ICs can look like dead shorts - keep in mind the core voltage of these GPUs is only ~1V, but current draw is >100A, meaning effective resistance is sub-ohm.

I've been wondering about this. Do these cards really do that much amps and are the power supplies aboard these things apparently capable of providing this?

[Rasz]

I have HD5770 that is using K4G10325FG, you think I can swap them too?

750 uses exact same chip so its a sure thing, no idea about K4G10325FG
but I wouldnt do anything without at least hotair (not even gonna mention preheater)

Hotair would be a great tool, although I have no money for it now, Ill ask my friend if he has one.

you are into ghetto, get yourself http://allegro.pl/kolba-hotair-do-stacji-wep-yihua-8786d-wentylator-i6815165567.html
+ triac dimmer(broken vacuum cleaner) or arduino/555 &triac, and build one.
or even cheaper http://allegro.pl/grzalka-do-stacji-hotair-wep-yihua-wenty-4przewody-i6913999737.html + computer fan and spare metal pipe.
Edit: actually gonna mention preheaters, one http://allegro.pl/promiennik-bga-grzalka-podczerwieni-400w-230v-x2-i7233690261.html should be big enough for a graphic card, its ECP-4 from www.elcer.com.pl, they are little cheaper brand new from manufacturer, but logistics might be a problem

Its not that extraordinary to make own tools, shop I worked at had homemade ~5KW rework station, similar to Ukrainian friends at

Yep, I'm thinking the same with baking, is there possiblity GPU was shorted and now it seems ok?

unlikely, how would that manifest on ram power rail?

Or if the GDDR's are ok it must have been one of ceramic caps?

most likely

If yes, than card will be working it should fail sooner or later, and I will find it after another failure.

yeah, from the photos it looked like you only pulled 2 ram chips? I would put them back and check again

I've been wondering about this. Do these cards really do that much amps and are the power supplies aboard these things apparently capable of providing this?

up to 200A, just like modern motherboards are build to survive ~200Amps. X299 boards throttle at ~300Watt to cpu.
but crazy people hit ~600Amps thru CPU socket at 1.45V https://www.pcgamer.com/overclocked-core-i9-7980xe-hits-61ghz-and-pulls-1000-watts/

[andrzejj]

unlikely, how would that manifest on ram power rail?

I found that russian schematic saying that its powering GDDR controller in the GPU, and also marked caps (at least some of them, haven't measured all) were 0ohm before, now they have same resistance as ram rail.
Here it is: https://imgur.com/a/45tlzMd

I will buy hot air for sure I recently got myself good soldering station (Iron is chinese Hakko T12 copy, regulator is based on STM32, they sell it on aliexpress, works like a dream 20-300*C takes max 10 seconds), hot air station will be next for sure.
Can I reball the GDDR chips in oven?
I desoldered 7 of them.

[Rasz]

derp, of course, it needs common IO buffer voltage. I even saw that picture before 

still, heating older broken gpu usually makes it reconnect broken connections, and doesnt "fix" shorts. Thats the problem with baking whole card, cant be sure of anything at this point

setting fresh balls using oven will be a challenge to say the least
http://allegro.pl/sito-bga-do-ukladu-ram-ddr5-playstation-4-i7295481432.html + solder paste / http://allegro.pl/kulki-bga-olowiowe-pmtc-0-40mm-0-40-0-40mm-10-tys-i4939486033.html + even shittiest flux
unless you are crazy enough to build 8x170 solder blobs by hand with soldering iron and expect it to work


edit: you might find this forum helpful https://www.elvikom.pl, learned a lot there while fixing laptops

[andrzejj]

Ok, im waiting for the parts, i plugged in GTX970 with one memory chip, 0,9V at GPU, 1,6V at GDDR, seems legit, isn't it? DVI socked deformed during baking, and i also dont have mini-HDMI cable, so I cant check if there is any image at the moment.
 In the meantime i diagnosed Aerocool templarius 1150W PSU, There must have been flashover betweend rectifier bridges terminals and wire jumper placed 1mm or less from them. Maybe some dust with moisture from the air caused that, or maybe it was hit by thunder. Also PFC mosfets are broken, when tested with low voltage from DMM they seem ok, but when 320V is applied they cause shortage. Mosfet drivers and all resistors around them seem OK, I'm going to measure caps ESR and capacitance, just to be sure.
So im waiting for balls, stencil etc for GTX970, and mosfets and bridges for PSU.
Also repaired GTX550TI- there was blown/broken off SOT-23 n-mosfet in the external power supply sensing circuit.
There was also GTX580 in my closet, PC boots with it(display connected to integrated card), 1V at GPU, 1,57V at GDDR, 4.95V at +5V in DVI socket.
GPU core resistance 0.5ohm, GDDR over 20 ohms.
There was 3 ceramic caps missing- one exploded, and 2 were just broken off. I replaced all of them with same dimensions caps from other graphics card.
I guess CHL8266 is GPU VRM controller and APW7088 is GDDR VRM controller.
CHL8266 datasheet: https://www.marutsu.co.jp/contents/shop/marutsu/ds/CHL8266_v1.05_10-12-11.pdf
APW7088 datasheet: http://www.anpec.com.tw/ashx_prod_file.ashx?prod_id=643&file_path=20131217184206967.pdf&original_name=APW7088.pdf

I think the controllers should send some kind of PWR_good signal, and its P_ok on APW7088, going to check that, but is there anything similar in CHL8266?
I've also read that these GPU-s tend to fail internally, maybe that is the case and there is no sense to investigate further?
Thanks for all of your help and knowledge!

And one more question, i saw ZXW dongle for mobile phones, there are many schematics/boardviews for laptops in the web, is there any tool/site to obtain desktop schematics? graphic cards, motherboards?

[andrzejj]

New balls set: https://imgur.com/a/YWgDbku
Waiting for the glue (UHU endfest 300)  to repair pad.
Seems to be great glue for that job, as it can be cured in temperature up to 200*C, so I guess it should withstand 220*C-240*C when cured without any problem, although haven't found any info about that.

[andrzejj]

And the GTX970 is alive
I used NC-599 flux, not sure if original or fake.
Waiting for new caps (Kemet 12mOhm ESR should be good?) as old ones got swollen during baking

Cheers

[Rasz]

Im honestly surprised you got it on the first try, nice
does your 559 say "known to the state of Colifomia"? counterfeit but not fake

[andrzejj]

Its in replacement package (syringe), ordered online, no way to tell if it is counterfeit or not, I'm guessing it is, as the cheapest soldering paste/flux(for tht and cables) did better job when I was setting fresh balls.
That is output image of card with 7 out of 8 memory chips installed, I'm guessing it would be the same if one of the chips would get damaged: https://imgur.com/a/M9YiTRG
Also, with 7 of 8 GDDR chips installed card works in low res as shown, when windows installed some drivers it went all black, and after a while PC restarted.
After installing 8th chip card is working great.
I don't know anything from the past about the card, is there possibility it had shortage on the BGA balls under GDDR chips from the factory? Do things like that happen?
I diagnosed the ATX, it had weirdly damaged PFC mosfets(not shorted, not wide open, had some kind of gate-source leak), and burned bridges, everything except that was working. There also was a transistor with one lead unsoldered from the factory, it was driving the broken mosfet, i think that was the cause.

I've found DPS5020, and It seems great. I'm guessing 20A, even with solid short cap would make it visible with IPA/ freeze spray?
Definitely next-buy.