Never ending WHEA 18 Cache Hierarchy errors, black screens and restart woes on my PC!

[emothxughts]

Update & bump, 1 Mar 2023:

 

I put my older, also second hand, RX 580 back into my system. I described the problems I had with this RX580 a few posts back, but still, this GPU and my emergency RX 550 were proven to never ever cause WHEA 18 restarts. This will do for the games I play at the moment. (Genshin Impact, Honkai Impact 3rd)

 

As for the RX5600XT card, I concluded that it was ultimately responsible for my never ending WHEA 18s and that it wasn't repaired properly by Gigabyte when I RMA'd it, if even at all. I sent it back to the original seller again today, as per warranty terms in my country Malaysia. I never want this lemon of a card in my machine ever again.

[emothxughts]

Update & bump:

 

I called the shop where I sent the RX5600XT card to, they confirmed that the card did indeed crash in their shop. What they said next however was a facepalm moment. Long story short, they said that the first time I sent the card to them, they didn't actually send it to Gigabyte after they checked the card. (For context, in my country, GB's warranty policies says that I should return the card to the original seller) This time around, they asked me if I would like the card to be actually sent to Gigabyte. I of course said yes.

 

It seems that the card kept crashing to a WHEA 18 error even after the apparent warranty return, because the shop didn't actually send it to Gigabyte, only kept the card in the backrooms or something. It also explains why Gigabyte says they don't have the card in their systems when I asked them. Wth??

[emothxughts]

As for refund and replacement: The original seller can only refund within seven days from purchase...which was in 2020. So refund's right out.

 

And replacement is only available after they send the card to Gigabyte, which they didn't do the first time around. If replacement does go through though, the seller offered two cards, RX6600 and RX6700XT both from Sapphire. But yeah, they just asked me to confirm sending the card off to GB today, so that'll take a while...

[emothxughts]

Update, 14 March 2023:

 

Good news: Looks like the seller actually sent the RX5600XT card to Gigabyte this time around. Bad news: GB says "The unit VGA unauthorised modification, change of thermal pad by user, this unit is out of the warranty terms and condition". I myself have never ever opened the card, so it must be the previous owner that I bought it from that did it.

 

Thing is, since thermal pads are applied to VRAM, what are the chances that the WHEA 18's I've been experiencing with this card are a VRAM issue? I've never seen artifacting with this card.

[YoungBlade]

7 hours ago, emothxughts said:

Update, 14 March 2023:

 

Good news: Looks like the seller actually sent the RX5600XT card to Gigabyte this time around. Bad news: GB says "The unit VGA unauthorised modification, change of thermal pad by user, this unit is out of the warranty terms and condition". I myself have never ever opened the card, so it must be the previous owner that I bought it from that did it.

 

Thing is, since thermal pads are applied to VRAM, what are the chances that the WHEA 18's I've been experiencing with this card are a VRAM issue? I've never seen artifacting with this card.

I'm guessing that Gigabyte just wants to find any excuse not to fix the card. It's odd for the thermal pads to be replaced that early in the life cycle of a card, so I don't know why the previous owner would have done that. I'm not saying that Gigabyte is wrong about the pads being replaced, but I think it's certainly possible.

 

As for VRAM overheating causing issues, I suppose it's technically possible? You could try opening it up yourself when you get it back to see if you can fix whatever is wrong with the thermal pads.

[emothxughts]

3 minutes ago, YoungBlade said:

I'm guessing that Gigabyte just wants to find any excuse not to fix the card. It's odd for the thermal pads to be replaced that early in the life cycle of a card, so I don't know why the previous owner would have done that. I'm not saying that Gigabyte is wrong about the pads being replaced, but I think it's certainly possible.

 

As for VRAM overheating causing issues, I suppose it's technically possible? You could try opening it up yourself when you get it back to see if you can fix whatever is wrong with the thermal pads.

I might be wrong, but I suspect the previous owner mined on the card. Haven't checked the Vbios on the card, but I noticed that in Superposition it is identified as a RX 5700 XT instead of 5600 XT.

 

Speaking of pads, the other two pics below were forwarded to me by the seller from Gigabyte. This is literally the first time I've seen the insides of this card.

[YoungBlade]

1 hour ago, emothxughts said:

I might be wrong, but I suspect the previous owner mined on the card. Haven't checked the Vbios on the card, but I noticed that in Superposition it is identified as a RX 5700 XT instead of 5600 XT.

 

Speaking of pads, the other two pics below were forwarded to me by the seller from Gigabyte. This is literally the first time I've seen the insides of this card.

The misidentifying as a 5700XT could be the result of using a different VBIOS. The 5600XT, 5700, and 5700XT are all very similar in that they all use the same die (Navi 10) so it is often possible to change the VBIOS around to have the card behave like one of the other SKUs. It's not a good idea to do it with a 5600XT, though, because it has only 6GB of VRAM, so unless the VBIOS was modified, it could easily cause the card to overflow memory and crash the system, because it thinks it has 8GB to work with.

 

That Superposition result does say that the card has 6GB, though, so I would guess that's what the previous owner did - they modified the card to think it was a 5700XT, after they edited the VBIOS to say that the card only has 6GB of VRAM. However, this can still cause instability, as the card isn't actually a 5700XT, even if it does use the same die. Some of the SPs are disabled vs the 5700XT, and this often happens for a reason - some of them have usually failed validation, so are turned off and then the card is turned into a lower SKU.

 

Flashing the card with a standard 5600XT BIOS might solve the issue for you.

 

As for the thermal pads, I would guess those are not original to the card, although they look to be in good condition. Usually, manufacturers use one long thermal pad for memory modules in a row like that, as it makes the assembly process faster. So odds are, yes, the previous owner modified the card.

[emothxughts]

2 hours ago, YoungBlade said:

So odds are, yes, the previous owner modified the card.

Gigabyte thought so too, which was their grounds for voiding the warranty on this card. Next move is to have the original seller post this RX5600XT card...not back to me, but to a repair shop I know on their side of my country.

 

About Superposition though, on newer runs after I got it back from the seller the first time around, it identifes the card as a Gigabyte Windforce 6GB OC 6 GB, still doesn't actually identify the card as an RX5600XT. And still no GPU degrees C readings either. I wonder what the seller did to this card back then...

[emothxughts]

Bump:

 

To sum up this thread so far, I bought a Gigabyte Windforce OC Rev 2 RX5600XT second hand to replace my dying RX580 card and my still living but weak RX550 card, but unfortunately my PC kept crashing and restarting with a WHEA 18 Cache Hierarchy Error. After stress testing and even swapping out nearly every piece of hardware in my PC, I concluded that the RX5600XT is causing all of these crashes. I sent it to the original seller to claim warranty as per my country's warranty policies, but the first time around the seller didn't even send the card to Gigabyte and sent me back the card unrepaired and still crashing on me. Took me sending it a second time to finally have the card sent to Gigabyte. Unfortunately, after I sent it out under warranty, I was then denied warranty repairs (this is not in the USA), because Gigabyte found out that the thermal pads were replaced (by the previous owner, I never even opened the card). I also noticed that when I run Superposition, it identifies the card as an RX5700XT, though between the first and second time I sent this card out, it then identifies the card as Gigabyte Windforce 6GB OC 6GB (Navi 10).

 

The card is currently not with me, I had it arranged to be sent to a third part repair shop (might as well, warranty's void). At this point, I am suspecting that the original owner did something to the VBIOS that also made it necessary to change the thermal pads on the VRAM, methinks it's mining. So for this bump, I'd like to ask, can a bad VBIOS cause the card to crash to a WHEA 18 error?

[emothxughts]

Bump and update:

 

The repair shop that I sent the RX5600XT card to reported that it did crash on their test rigs. Then they gave me a bombshell reveal, that (to paraphrase) "one of the memory chips has a different code". They showed me this pic:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Seems that one of the chips is less worn out and therefore newer than the rest. Also there are some...marks above the chips. What do you think about this?

[YoungBlade]

1 minute ago, emothxughts said:

Bump and update:

 

The repair shop that I sent the RX5600XT card to reported that it did crash on their test rigs. Then they gave me a bombshell reveal, that (to paraphrase) "one of the memory chips has a different code". They showed me this pic:

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Seems that one of the chips is less worn out and therefore newer than the rest. Also there are some...marks above the chips. What do you think about this?

It could be that a memory chip died and the previous owner had it replaced. That would explain why they put on all new thermal pads - a repair shop would generally do that. Compared to the cost of the repair, a few thermal pads is nothing, and it generally would improve the customer experience.

 

Just like with mixing RAM, mixing VRAM modules can cause issues for the memory controller. It's possible that these memory modules aren't playing nice with each other.

[emothxughts]

Update & bump, 14 April

 

The repair shop has confirmed that VRAM and power connector are OK, yet the card still crashes in games. They now say the only thing left to do is to reball the core, which they say is a risky proposition that may kill off the GPU for real.

[emothxughts]

Bump! And some updates:

 

The repair shop appears to have successfully reballed the GPU, it output display and passes benchmarks and doesn't crash in games. But then a few days later, the repair shop run those tests again and unfortunately the GPU green screened in benchmarks and in games, and they have to restart the system by force. They offered to return the card to me for a much smaller charge than if the repair was successful. I finally got the card back today, I am running Unigine Superposition and Heaven on a loop as I am writing this, and no crashes so far. Weird.

 

Thing is, the system that crashed and the system that didn't aren't the same, as pictured. I hear that PCIe gen 3.0 risers don't play nice with PCIe gen 4.0 cards like this RX5600XT (notice where the other end of the riser cable is connected to), could the riser be responsible for the green screen crash and that the card itself be okay?

 

And as for the other card mentioned in the very beginning of this thread the RX580, it is unfortunately dead, it blackscreens under any kind of load heavier than idle. I sold it off for parts. That's the end of the story for the RX580, bought used for $292 in 2021, sold as faulty for $13.

[emothxughts]

Update, final one perhaps:

 

So I got my RX5600XT cards back 2 weeks ago, and I installed it in my PC anyway despite the previous green screen issue. Surprisingly it runs well crash free and green screen free...at first. Then a week later, while running the Unigine Superposition benchmark, my PC crashed yet again, and when I checked the Event Viewer afterwards, it's the same WHEA 18 error as always. That was when I decided this card is a lost cause.

 

Sometime before I got the RX5600XT back, I ordered yet another used card, a Zotac Mini GTX1070, for $81. It came in the mail a week after the RX5600XT, I installed the 1070, and I'm having no crashes ever since.

 

In conclusion:

• The RX5600XT card is a lost cause, I still don't know what causes it to cause the WHEA 18 errors to this day, and neither does the repair shop.
• The RX580 card is outright virtually dead now. Sold for parts.
• I replaced both these cards with a GTX1070. No crashes ever since.

Sorry for those who has followed this topic, but there is no real solution here, only replacing the cards with another one.