RX570 completely dead, then not.

[fiveseventy]

Hello all,
I have a Sapphire Radeon RX 570 Pulse ITX 4GB (single fan) which has recently developed an issue.

Most times when I boot up it will be totally dead.
Its fan don't spin and I see nothing but a black screen, but the system does post and windows boots.
I can confirm this via RDP, and can see that the GPU is not listed in device manager (not even as unknown), or detected by GPU-Z.


Interestingly, without changing anything, the card sometimes just works again after some time powered down.

I've noted that when it's up and running fine I can reboot as many times as I like and it continues working,
but if I shut down and start up again, even immediately, the card will be dead the vast majority of times.

When it does work it works 100% perfectly. No glitches or artefacts, low temperatures.
I've let benchmarks run and played games - Just perfect.


I feel I've ruled out multiple things.

It's not overheating - It's very clean and runs cool.
It's not the power cable to the GPU, or power supply - All confirmed working fine with other more taxing cards.
It's nothing to do with the rest of the system - Issue is reproducible on two other computers.
It's not as simple as "only works when it's cold" or "only works when it's warmed up" - it's not that consistent.
It doesn't seem to be related to bending/warping/pressure being applied, so I'm ruling out dodgy BGA joints.
I can't see anything physically wrong on the card, like a burned component or bulging/leaky cap,
although I appreciate that many component level issues won't be visible.

I considered self-healing fuses but, then, sometimes it took days before it came back to life and other times just minutes.
I don't even know if it has those.


Today I've determined that any time the card fails I can revive it 9 times out of 10 by removing it from the machine and handling it.
Any time I took it out to check fuses or probe anything it would just work the next time.
Over time I realised that simply touching the solder joints on the back of the card (ESD safety FTW) usually results it in working the next time.

Could that suggest that something is holding charge and shouldn't be?
I'm manually grounding/draining something?

Given that it's just not detected at all when the issue is present, I'm guessing that means it's something relating to, or affecting, the bios?
That made me think maybe there's faulty cap to ground on the 3.3v, possibly, but that's really as far as my best guessing goes.

I'd be very grateful to hear any thoughts on it!
Thanks for reading, if you got this far.

 

[VuurVOS]

You did quite an investigation! Did you try different HDMI/DP ports on the card? Different HDMI/DP cables? Different monitor?

 

[ty_ger]

I would suspect a bad motherboard slot connection. Either due to mounting pressure, misalignment, or corroded contacts.

 

[fiveseventy]

You did quite an investigation! Did you try different HDMI/DP ports on the card? Different HDMI/DP cables? Different monitor?

I have tried with both HDMI and DVI outputs.
I doubt that would matter as the card isn't recognised by the system at all but anything's worth trying!

I would suspect a bad motherboard slot connection. Either due to mounting pressure, misalignment, or corroded contacts.

The contacts were already spotless although I did give them a light clean with a pencil eraser, then an alcohol rub down.
Mounting pressure is possible, I suppose,
although when the card works fine and I shut down / power up again, it will cease to work.
This is very reproducible and happens without touching or moving anything.

Thank you both for the replies!

 

[VuurVOS]

I have tried with both HDMI and DVI outputs.
I doubt that would matter as the card isn't recognised by the system at all but anything's worth trying!

Does your system hang or beep when the monitor isn't connected during power on self test?

 

[fiveseventy]

Neither - It carries on and boots up without issue, even when the card has failed.
I can use RDP from another machine to verify that, and to confirm that the card is not detected by the system at all.

 

[screet87]

Although you rule it out it sounds to me like cracked solder joints under the core or within.

 

[fiveseventy]

I suppose I can't 100% rule it out but the fact that I can reboot forever, it seems, without issue but power cycling once kills the card made me think there was something more to this.
I am here to ask rather than tell, though, so if you can elaborate at all I'd be very grateful.


I had hoped that the card being completely invisible, when in broken state, might have been a big clue?
I don't know this to be true but had hoped that would point to the bios chip or related components, as many other faults would allow the card to be seen but not function, or function with issues?

Yesterday I did save the bios when the system was running and, for good measure, flashed a known good bios from here,
mostly to see if it would complete both operations without issue.
It did.

It seems to me quite likely that whatever the issue is, it takes place when power is temporarily shut off, and only then.



...a few moments later...
Pursuing that idea I tacked a wire on to vcc (8) of the bios chip so I could monitor the voltage reading there with a DMM.

When the card works I read around 3.2v as soon as I power on but that falls to 3.1 within a second or so.
When the card does not work I'll read around 3.2v but it stays there.

I'm finding it's somewhat difficult to provoke a failure with the DMM connected.