RX570 dies completely, then recovers.

[fiveseventy]

Hello all,

I have a Sapphire Radeon RX 570 Pulse ITX 4GB (single fan) which has started misbehaving recently.
The card was bought used maybe 5 years ago and has worked without issue until recently.
I have no reason to believe it has been worked on or physically damaged.


For periods of time it will be 100% perfect.
Perfect temperatures, no glitches or artefacts, solid FPS.
I can game, run benchmarks, do whatever without any problem.
I can reboot the machine as many times as I like and nothing changes

however, most times when I shut down the PC the card will appear dead on next power up,
even if that's only seconds later.
This is repeatable in three separate PCs with no other common components.


By 'dead' I mean completely invisible to the system, and no fan spin.
My system will boot up with no output to monitor and I can use RDP from another computer to get in,
but the card is not detected at all - Nothing shows in device manager (not even unknown),
and all the fields in GPU-Z are blank. It's like there's nothing in the slot.


Many times the card just came back to life after some time powered down.
Sometimes an hour or two, sometimes overnight. Sometimes several days.


On a hunch I considered the bios chip and related components.
When the card is up and running I can successfully save and/or flash the bios without a problem.

I have monitored VCC at bios chip pin 8 (tacked a wire on) with a DMM and determined that
when the card works the reading will dip around 0.1v a second after power on and hold there,
whereas when the card does not work the reading will remain steady without dipping.

This appears to be perfectly consistent.

I've also noted that when the card has died it could be dead for hours, or days,
but usually if I take it out and handle it…maybe probe to test the fuses or check some caps to ground,
generally when I plug it in again it will work straight away.



I feel I can rule out a few things.

• It's not overheating.
• I don't think it's related to the card edge connector (spotless) or anything mechanical.
  I've seen no pattern to suggest that.
• The card being warm, or cold, doesn't seem to be important.
• All other hardware/peripherals seem irrelevant, as the symptoms persist in three machines.
• I'm ruling out video cables, monitors, etc, as this is all reproducible with none connected.

The only concrete reproducible thing is that VCC measured at bios pin 8 dips 0.1v when the card works,
but doesn't when it doesn't.
I'm not sure what that points to and was hoping someone here might have some ideas.


So you know where I am, this is my first experience attempting to troubleshoot a GPU.
I have a basic level of diy/tech repair knowledge and ability. I could replace a cap or resistor but you won't find me reballing VRAM or anything like that.


Thanks in advance for reading, and for any thoughts.

[DynaxSC]

First get a boardview of the card and measure all main voltages when card is powered and in non-operating mode. Check is any voltage is missing then.
Might be some electrolytic cap failing, they are able to recover sometimes. Could be also any other device failing and recovering. Unfortunately this kind of failure is the most difficult one to trace. U might try using a freezing spray on working and powered card to narrow down the area, but this must not necessarilly work. Heating an area with hot air is also an option, but temperature must be reasonalble, especially for electrolytic caps. Attaching some smaller CPU cooler instead of the normal one can help access the devices. Don't feeze/heat the GPU itself. This might end up with disaster.
Bending the card during operation might show if this is some cold solder point.
Measure all low ohm resistors for decalibration (0, 1R0, 2R2, ... up to 50 ohms if exist).
Using a riser is also very helpful for diagnostics.

[fiveseventy]

Thank you for the reply, and suggestions.
I've been able to confirm that 5v and 1.8v rails are present, whether it works or not,
but going any further may be difficult without a riser or smaller cooler, as you say.

I'll see what else I can check, and go round the low ohm resistors as you suggest.
Thanks again.

[fiveseventy]

I've had a little bit of time to poke around.

I'm looking at the RX580/570/480/470 2.1 boardview - Thank you for letting me know about those.

I've already reported pin8 at bios chip reads constant value when the card doesn't work, but drops 0.1v almost immediately when it does work.
I observed the same pattern at u100 pin 4, PERSTB_BUF.

Output at 5V regulator is 5v under all circumstances.

VOut at U300 reads 1.8v under all cirumstances.

One interesting thing I did find is that around Q2604, providing VDDC_VDDCI_OE_VR,
the two resistors at +MVDD input read wrong values.

I realise resistors read in circuit can be misleading but R2627 reads 1k, not 10k,
and R2693 reads 0ohm. Boardview says 5.1k.

Seems like I should read at least some value on both resistors, no?
I think I'll probe the output here (which I think goes to pin21 EN) while the machine is running to see what I can find
then, depending on the result, maybe temporarily remove those two resistors for isolated testing and/or replacement.

Very much stabbing in the dark here so any thoughts are truly appreciated.

[DynaxSC]

R2627 - that can be normal.
R2693 is strange, but sometimes can happen if a mosfet is switched on, mosfets have some gate memory and sometimes it takes some time they switch off, even unpowered. They behave a little like ordinary switches and the "on" resistance is very low (miliohms). U may look if there is a voltage on the resistor when card is powered.
Sometimes changing the probes when measuring in-circuit can give different results (polarity dependencies).

I suggest before u go on with repair, that u read as much as possible other posts about GPU repair, there is still a lot knowledge to find on the forum, so u get more clarity whats going on in these GPU's. They are all quite similar in architecture, so does not very much matter if it is another model/make.

[fiveseventy]

I can't see any circumstances that would make a 10k resistor in that environment read 0ohms other than
a short across it - solder blob, for example,
a wrong value or damaged resistor
a mistake in the schematic.

As it turns out I think it was the last option, at least according to readings taken by another user on another thread.
They described the cited resistors as 1k and jumper, so that's good enough for me.
Shame.

"Sometimes changing the probes when measuring in-circuit can give different results (polarity dependencies)."
What you're describing is often the effect of diodes in circuit. They greatly restrict the flow of current in one direction.

Well, ok. I'll keep digging and reading.
I started this thread hoping that the seemingly unusual nature of the problem might be an enticing mystery for someone with experience.
I've yet to see another thread where the person's card flips between 100% dead and 100% perfect.

Regardless, I can continue to apply general troubleshooting techniques as you suggest and see where that goes.
Thanks for the reply.

[Fachrudin_Ahmad]

I also have the same issue. I have an RX 570 Sapphire Pulse. I bought the VGA in used condition. I only used it for 2 days without gaming or any editing, but suddenly there was no display, and the fans stopped working.

Then, I started to investigate the root cause of the problem. Initially, I had no knowledge about repairing GPUs, relying solely on YouTube and this forum.

Straight to the point.

First, I checked the condition of the fuses. Fortunately, on the Sapphire Pulse, if there is a short, the fuse will immediately blow, preventing the problem from spreading. The Sapphire Pulse has 3 fuses: one near the PCI and two near the power pins.

The condition I found was that the two fuses near the power pins were blown. After researching from several references, I found that these two fuses are connected to the 3 vcore. See the attached image.

Then I checked the 3 v-core SIC632 with a multimeter in short/beep mode.

Specific Short Checks See attachment:

• Short between V_IN and GND: Check for short between pins 8 to 11, 34 and 12 to 15, 28, 35.
• Short in Internal Logic Circuitry: Check for shorts between V_CCIN (pin 3) and C_GND (pins 4, 32).
• Gate Driver Short: Check between V_DRV (pin 29) and C_GND (pins 4, 32).

I found that the SIC632 was shorted and produced a beep, and the resistance was very low, approaching 0.

I then compared it with the vcore that had an intact fuse, and the result was high resistance and no beep. So, I concluded that there was a problem or short in the 3 SIC632s connected to the two blown fuses.

Finally, I replaced the 3 SIC632s and the 2 fuses. My RX 570 came back to life.

However, I am not sure what exactly caused the vcore to short.

1. Was it due to overheating?
2. Was it due to a problematic power supply?
3. Was there an issue with the VRAM?
4. Was there a problem with the circuit?
5. Or other issues?

I hope my diagnosis and repair are correct, because if the VRAM or circuit condition is problematic, the short is likely to happen again.

[fiveseventy]

I've since passed my card on to someone with more time and knowledge,
and I can say that my fuses were not blown,
but I appreciate you adding to the thread and sharing your findings.

Hopefully your card keeps on trucking and this info will be useful to other!

[Fachrudin_Ahmad]

It's OK, brother. Sharing your problem and repair efforts is very helpful for me as well.