Hacking GPU cards back into operation because I need something to do....

[deadsix]

As soon as I turned on my Rig, I saw sparks fly off the top edge of the back-plate and I immediately yanked the power chord.
On removing the backplate this is what I saw :



Surprisingly though, i plugged it back in after a while to flash stock bios, and saw that the card still works flawlessly. Sent it in for RMA anyways.
Thought this might interest people.

Also, the VRM area just above the highlighted circle, the backplate has a kinda cushiony thermal pad there that makes contact between the backplate and the pcb, but it starts oozing some kinda sticky liquid after 4-5 months of operation (atleast for a few of my cards).
I believe one other card died earlier due to that liquid, unsure about that though, had gotten it RMA'ed.

Anyone else having similar experiences on the RX 470's? Investigative diagnosis from the masters? Suggestions and Warnings, pitfalls to avoid in the future? For the record, my cards are all at or below stock speeds and heavily undervolted.

[1714833809]

1714833809

[1714833809]

1714833809

[adaseb]

Anyone else having similar experiences on the RX 470's? Investigative diagnosis from the masters? Suggestions and Warnings, pitfalls to avoid in the future? For the record, my cards are all at or below stock speeds and heavily undervolted.

That's what usually happens, something shorts out, your PSU cuts power. Then when you turn it back on, its when it usually starts sparking/smoking.

this is a good tip, i just started work on that gigabyte we've been pm'ing back and forth about.

preparing to get out the good ole solder sucker tho ;p coupled with my hakko of course.



any idea why all sapphire gpu's have this thermal pad stuck over various smd components? also leaving that single ram chip completely uncooled.

Its not only Sapphire, ASUS does that also. No idea why. Its not an issue until the >2GB games came out and people started getting artifacts due to certain ram chips overheating.


By the way, I had the exact same Sapphire Dual-X 280X go on fire also, it wasn't in that spot, it was one of them VRMs. It pretty much messed up the PCB completed, I desoldered that burnt VRM, checked for shorts and ran undervolted for a while, and then another VRM ended up blowing.

Apparently these VRMs had issues where they easily ran at 120C and higher and didn't throttle properly and ended up frying themselves. I think the issue was solved with the later Vapor-X 280X and everybody complained about throttling issues but those at least didn't blow up.

If you search the old Litecoin forum, that GPU and the Gigabyte Windforce Tahiti's were among the worst for mining because they had VRMs blow up and Capacitors.

[joerazor]

Very nice seeing this thread

I have 2 dead GPUs (R9390x) (Sitting on the shelf for the past 5 months and wish I could put them to use)

I wonder if there is any easy fix to get them working. I will explain what caused the problem for each:

GPU 1 (R9 390x Sapphire): I mistakenly inserted the powered PCI Riser in the wrong direction. Immediately after powering my PC on, I saw a spark and some smoke (Not sure if it came from the riser or GPU).... Since then, my GPU would never get detected in "device manager" and the GPU would never output anything on any connected monitor... Tried the GPU on 5 different computers!


GPU 2 (R9 390x Asus Strixx): Fans on this GPU failed, so I purchased an Arctic Fan Cooler from amazon and managed to replace the stock asus fans with the arctic ones.. Everything worked fine for a few days then suddenly, my PSU would switch off immediately after a second of switching the computer on.

After fiddling around, I discovered that the problem was with this 390x Asus Strix card... I guess it is causing a short circuit or something causing the PSU to switch off immediately.


Is there any way that I can resurrect these cards from the dead? Please let me know your thoughts! If any of these card work, I will gladly donate 10ETH as a thank you.

PS: One of my friends is an advanced Motherboard technician, so I could pass the info you provide to him.

[smaxz]

Alright so part 1 of my 7950 Gigabyte Windforce 2 tear down and overhaul.

The windforce cards are well documented as not the greatest for mining. Particularly early revisions where the VRM had no heat sinks or cooling provided.

I answered an add and picked up a card that was "on its last legs, not overclocking and displaying severe artifacts"..  the seller admitted to overvolting and mining early scrypt coins with this but fortunately the price was right for a chance to mess around with and learn more about these cards.

Plugging it in at home it mined surprisingly well off the bat, but any overclock put it unstable and crashed the OS shortly after. Flashing the ROM to something more tame helped but not by much and even at factory clocks there was artifacting under load and crashing after sustained mining.

I did notice the VRM area was equipped with a heatsink but still got incredibly warm, too hot to touch infact and the back of the vrm just had this flat bar across it as a heatsink which was also uncomfortably hot.

Time for disassembly to apply some VRM cooling mods..



I found some Enzotech low profile copper heatsinks for GPU ram that looked really promising and clear the height allotted by the factory windforce heatsink.

With a steady hand a thin dremel cutting disk I prepared some heatsinks with intent to fix to the top of the VRM mosfet's with some fujipoly thermal pads.

Then the factory aluminum heatsink for the vrm mosfets could be repurposed for the underside instead of that flat bar to keep things chill.



Unfortunately while cleaning the thick caked on putty/paste used as a thermal interface medium (TIM) on the underside of the VRM components, a problem became apparent.




can you guys spot it?




There was a little SMD/SMT (Surface Mount Technology/Device) capacitor that either disintegrated under extreme heat, had a crack in its solder and came off while I was cleaning the board, or the previous owner attempted to reapply thermal grease here and damaged the board unknowingly.

So with an idea what the issue was, time to figure out how to remedy it.



SMD capacitors come in various sizes, these particularly are extremely small, getting out the digital caliper and accounting for the extra solder on the end of the cap, the dimensions are consistent with 0402 size SMD capacitors.

Knowing the size, the value of the cap is not as easy to determine. Checking with a multimeter is really not so accurate in this instance. Ideally the manufacturer will have the specs on hand and can identify against the board assigned "Cxxx" designation printed beside the capacitor.

I emailed gigabyte support asking for a schematic of the board or value of the capacitor in question and received a not so prompt reply stating this is proprietary information. Further prodding revealed that I would not get far with gigabyte/amd for this issue so decided to dig further on my own.

Although the schematics are kept quite secretive, the VRM portion of these cards are usually sourced out. To find which VRM system is used on your GPU you must locate the PWM (Pulse Width Modular) which is the VRM master control chip of sorts, and read the printing on the chip surface.



It will be a larger chip, off near where the VRM mosfets are all lined up. A magnifying glass may be helpful at this point. If you're lucky the text will still be legible.

Mine read:

ADP 4100
JCPZ
TAB49003
#1249
KOREA

Which lead me to ADP's online schematic for their 4100 system at http://www.onsemi.com/pub_link/Collateral/ADP4100-D.PDF

Where the helpful data was identified below:



Which puts me to where I'm at now.

Stay tuned for how to resolder/reflow smd board repairs, application of upgraded VRM cooling on Gigabyte Windforce 7950 GPU and the performance gains afterwards.

[lightfoot]

preparing to get out the good ole solder sucker tho ;p coupled with my hakko of course.

Another solution to removing SMD chips with an iron is to try chipquik. It lowers the melting point of the solder to where you can remove most components (not QFN IIRC) with a simple iron, then clean off the stuff, tin with solder, and put the new component on.

any idea why all sapphire gpu's have this thermal pad stuck over various smd components? also leaving that single ram chip completely uncooled.

Trying to move heat. However thermal pads are kind of useless but if you have no airflow over the chips I guess it's better than nothing.

[smaxz]

alright so after a long wait for parts to arrive and research/prep on how exactly to work with SMD, i spent the end of the weekend getting my fix on.



check out how the 0402 caps are shipped, encapsulated in that little strip.. grabbed some solder paste and since i have a surplus of these components now, i removed the other of these same variety caps in the mosfet step circuitry as they probably got subjected to excessive heat over their life span which is why the card was in its sorry state to begin with and it was parted with so cheaply.



you really have to be on your game with these tiny components.. i should have bought those visor goggles with magnification in them, but think i fared alright with the naked eye.. close up for scrutiny provided (: started with a toothpick but ended up using the head of a pin to apply paste, and worked great at picking up the capacitors also.



the paste is sticky enough to hold the parts in place but i really wanted to get some smush on these components to ensure a solid connection.. i grabbed some foil tape, 3M foil tape.. it has an acrylic adhesive that holds up at least until the paste starts to work its magic. polyimide tape is the way to go here if you got it.. or simply omit the step, the residue left over made me think twice about going this route in the future.



next its off to my designated reflow toaster oven. this paste melts at around 420F so i went 425 for 4 minutes and spiked it to 450 the last 2 minutes.

Now with the repair work out of the way and the GPU settling down after its roast in the toaster.. we can focus back on the mosfet heatsink mod i had originally intended.



because i'll be using Arctic Alumina which is an epoxy adhesive with quick cure time, its vital to have your shit together while executing this mod.

above is a dry run planing the layout of my sinks and which gets designated to which chip.



since it would be a waste to permanently fix these nice little copper sinks to this gpu, i'm only going to apply the epoxy to the corners of the mosfet and a conventional tim in the center.. not only does it provide a means to repurpose these heatsinks in the future by making them removable, but the conventional paste actually dissipates heat better than the epoxy so the benefit it two fold.



we're going to use a clamp once the sinks are all in place, one could have opted for cardboard or another buffer for the underside, it's much more practical to use the repurposed heatsink that was originally on the top side though.



and all clamped up, try to use a soft wood for this if you can. and leave it a minimum of 2-3 hours. overnight is best.



while the clamp is working for us, we can focus on the fans.. there were working great beforehand but maintenence can never hurt.



many complain their fans must not come off.. sometimes you have to be forceful. i have never broken a blade performing this maintenance.. using a flat blunt edge helps with prying action if it's super stubborn. dont waste your time with the dental floss around the fan blades trick. (:

after cleaning the blade cavities and the motors, put a small drop of oil in the middle of each fan motor.



heatsinks come from factory in non optimal form as it is, but seriously.. who is the nutter that takes a razor blade to scrape old TIM off their heatsinks? i run in to this ALL the time.



properly lapping the heatsink surface does wonders. get your thermal tape ready too if thats what you plan to use on the GPU RAM.



properly applying TIM to the compute unit of the gpu is vital, depending on product you use, excess can be anywhere from catastrophic to just plain insulating heat.



better than new.



better than new.



better than new.



one thing of note, if you intend to keep the metal bracket on the side of the card (i suggest you do) and re purpose the mosfet heatsink for the underside, you will need to clearance that bracket a bit.. i just used an old file i had laying around but a dremel could do the trick just as easily. alternatively you can clearance the heatsink, as it's aluminum and softer.. this was an after thought for me ;/



here's the GPU back in action.. many solved blocks left in her, fingers crossed.

eyes peeled for next project!

[bitkilla]

Lightfoot how are you doing?

I am not real active on this forum, but I have a couple of 2 th
dragon miners with coincraft chips that conked out on me.

I Started having psu issues and even when I replaced the psu's
I have problems getting them hashing.

Can you send me an email so we can discuss possibly sending the miners to you
for repair?  Thanks for your help please check your pm box for my email or pm my

Thank you

[666mrga999]

My card totally dead after PSU fail, system would not boot with new motherboard and this GPU inside, not even if internal GPU is selected as primary... Any ideas?

[DevonMiner]

Interesting thread, only just noticed it.

Over the years I've seen loads of good mining GPUs on eBay (UK), and quite often the description says 'no video display, for spares and parts'. Always wondered if they would be OK for mining as we don't need it to display video, just hash away.

Maybe I should bite the bullet and buy one, just for fun 

[JaredKaragen]

So;  I have begun to dive into a GTX780 that my friend gave me....

He was playing a game, in his words "trying to abuse SLI" and "run them hard"... when it suddenly just bluescreened, and rebooted, showing no display on the primary video output, but the secondary card showed up.   Only one GPU in the device manager.


The back of the board looks perfect.... no signs of overheating components or anything.

The component/die side... looked fine as well.....

This leads me to want to power it on the bench, and take some voltage measurements at places.... find why it doesn't power up in the slightest (no fan, no post, nadda)

So how do you power these in order to test in such a way?  do you clamp on a rigged cooler to the GPU die for this and just let the rest run open?

Any insight into this sort of issue before?  My gut tells me to follow the power from the PCIE first....  Just by the fact the card displays the signs of no power reaching it.  Also, he said he was messing with the SLI bridge, and I believe the power piggybacks through that connector as well...

I do have a few oscilloscopes (one analog, one digital), a handful of test equipment as well as basic soldering/desoldering equipment.... if it needs heat-work done;  I'll finally have a reason to invest in a bed and temp controlled system.   I've done enough research and learning about the processes under different situations/conditions that I feel confident I can attempt an advanced repair on something.

[Nalut]

I had an RX 570 card which i could use for 3 minutes only to test the rig, after reinstalling my rig HDMI output nothing even though the card light is on "Sapphire logo's light"

is it possible it can be fixed ? i might disassemble the card later to check if the board is damaged or not.

[smaxz]

570 would likely still be under warranty.. and if you can boot it, ensure it's got oem bios and send it for exchange is what I would do.

[lightfoot]

Lightfoot how are you doing?

I am not real active on this forum, but I have a couple of 2 th
dragon miners with coincraft chips that conked out on me.

I Started having psu issues and even when I replaced the psu's
I have problems getting them hashing.

Can you send me an email so we can discuss possibly sending the miners to you
for repair?  Thanks for your help please check your pm box for my email or pm my

Thank you

Sure, will do. Been kind of busy with litecoin miner repairs and now am getting bitcoin gear again. Feel free to PM me again and we'll discuss talking a look at it.
C