Help dead vega 64

Hi there, I have recently just built my first pc.
I bought a used gigabyte reference vega64 off ebay and the heatsink for it was completely damaged, causing overheating. After ripping on it a little, it looked like this.

So I then decided to take matters into my own hands and I attached my cpu liquid cooler to it.


I plugged the card in and it ran far better. ~50 degrees C at ~150W load of furmark for a few mins

Sadly, afterwards , I decided to leave the computer unattended and turned on at idle for over an hour and by the time I come back the card is just dead. No video output, No GPUTach LEDs, it wonā€™t get past BIOS and I canā€™t feel the card warm up when i turn on the PC

Is there any way I can fix this issue and of not, will I be able to get gigabyte to RMA the card?

1 Like

Did you cool the memory and vrms at all? If the power distribution cooked then you have probably toasted the card.

You say you bought it ā€œoff eBayā€. I take it this wasnā€™t a retailer? For RMA you usually need proof of original purchase.

5 Likes

your VRMs overheated, you need heatsinks and fans on them
try turning it on now that its had a chance to cool off

2 Likes

Iā€™ve already tried rebooting the card. And nothing happens. I highly suspect youā€™re right in that I toasted my VRMs since they were running at about 96 degrees C under 150W load when I first booted the card up.

Guess I shouldā€™ve been more patient and waited for a delivery of heatsinks to come in.
Im the UK and weā€™re currently under lockdown again :joy:, so the local hardware shops are shut.

Anyways, thank you, I have messaged the eBay seller for the purchase invoice and also gigabytes UK rma dept ([email protected]). Hopefully they get back to me and iā€™ll see what I can do.
Cheers

If only they had fully integrated voltage regulators (FIVR) on gpus :joy::joy::joy:. I seen some patents from amd about that though for future zen products

Would make them uncoolable. Imagine if the 3080s 600mmĀ² die with its 300W power draw was to output another 50W from VRM switching.
Run it in oil submersion of not run it at all.

Well, unsurprisingly, gigabyte refused to repair the gpu, and now iā€™m left here with a dead vega card and very little idea on how to fix it :frowning:
Iā€™m thinking I could probe it with a multimeter, but mine is only in the k Ohm to M Ohm range.
I just donā€™t really know where to begin with this whole thing as coming upon pcb schematics for these cards is, well near impossible.

Here, you can check out, how guy that has no clue is trying to repair gpu, under proā€™s supervision :wink:

Note the equipment used, even if you knew what to replace :wink:

But seriously, with ordinary soldering iron you could probably do some burned resistors or fuses. Multipin stuff usually needs hot air at least. Microscope can be substituted with magnifying glass, but its painful to use. And of course various items like proper tweezers, solder, flux, all this quickly adds up.

I donā€™t want to discourage you, but realistically, if this is one-off repair then equipment you need, probably will be more expensive than new Vega.

1 Like

Iā€™m not really too concerned about equipment costs as I just started a uni course in engineering and when I go back in January I can just abuse the departmentā€™s tools in order to fix this thing.
Guess iā€™ll just have to wait it out a month or so. :joy:

Oh, iā€™ll also need to find replacement screws for the card :joy:

1 Like

The part about using server PSUs for a whole system :zap: :zap: :zap:
(links included)

HELLO. well as much as I wanted to go back to uni and diagnose/repair the GPU, Iā€™ve been stuck at home for the past month and will probably continue to do so :frowning: Well anyway, I hooked up my dead Vega GPU to a cheap eBay test rig in order to supply it power
.

All the power phases measured at 0V except for the PEX rail (the small one next to the PCIe slot) and the middle of three smaller rails on the left, which correctly measured at 1.73V (should be ~1.8V).

Now although Iā€™ve measured these voltages, I canā€™t find any public datasheet on the IR35217 pwm controller, and you obviously canā€™t get GPU PCB schematics soooā€¦ idk what to do from here. I mean I found the datasheets for the smaller driver ICs but given both the 3V and 0.8V rails are at 0.

@wendell On a good note though, this whole journey did lead me to discover a method in order to cheaply power a PC with cheap HP server PSU units. Found this out from this video (1:55):
Building a 2X GTX 1060 6gb Mining Rig! - YouTube

Also Here:
Mining Rig Server PSU with ATX Motherboard PICO PSU - YouTube

As the server PSU is 12V only, all the 12V devices (CPU, GPU) can be powered off the server PSU itself directly with 6 pin PCIe to 6+2 pin PCIe cables male-male along with PCIe to EPS male-male cables for the CPU. Though youā€™ll probably have to DIY those EPS cables. This lets you power only the motherboard and other lesser components with the power adapter (PicoPSU, HDplex, or whatever Power Distribution Board used).

Components needed:

All in all, this is all available on Ebay should probably cost somewhere around Ā£60 for a 750W unit total if you go hunting and probably makes for a fun ITX power solution.

Any help?

So all of the VDCC rails have 0 volts at the output of the inductor, and the same is true for MVDD?

Do you have a thermal camera to see what parts are getting hot? If not, utilizing rubbing alcohol could help identify abnormally hot spots (ex. Caps shouldnā€™t get hot unless they are shorting to ground or really heavily pushed, which I doubt would happen on a GPU).

How much power is the GPU pulling from the 8pin connectors? You can use a current clamp to check (or a very DIY multimeter setup).

yeah, thanks for the advice. That would help with the diagnosis of my GPU. However, I just decided to send off the GPU to Buildzoid and hopefully, he can manage to resurrect it as I donā€™t know nearly enough about GPU VRMs or PCB schematics to have any good chance at it. So fingers crossed I guess, although I donā€™t have much hope thoughā€¦

1 Like

Update. Buildzoid is a legend and turns out, all that was wrong with the card was one dead enable transistor. He fixed it and is sending it back to me.

7 Likes

Just to close off this thread: Buildzoid actually did repair the card. Iā€™ve seen it on his twitter. Sadly, when it got to me, I took it to Uni w/ me, and it still wouldnā€™t work. Same dead as before.

So I tried to replace the labelled ā€˜Q301ā€™ enable transistor. However, there was no details on part numbers or whatever.
So I just bought a pack of tiny BJT transistors based off the markings on the teeny transistor.


Even though I knew it was probably the wrong one, I did manage to replace it and solder it on. God knows how I did it. But I think that murked the card because I saw the mboard lights flicker between GPU and something else.
At this point I just had to call it, so I sold it off for Ā£170 on eBay. Almost as much as I paid for it.

Thanks

This is possibly the best emotional rollercoaster Iā€™ve ever been on.

Tops op.

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.