Question RX 7900 XT Crashes - Help!

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Cableman

Member
Dec 6, 2017
78
73
91
I recently replaced my GTX 1070 with a RX 7900 XT. Finally had time to install it today. Used DDU to clear the old drivers. Ran Heaven a few times without any problems. Then started a few games to check the performance. Tried Control. Within 5 mins, I get a black screen and the PC crashes. It started doing the same in Witcher 3. Then I tried Heaven again and was running GPU-Z to monitor the temperature. It never goes beyond 76-77 degrees. After a few minutes, I get the same black screen and crash (at 75 degrees). No other apps are running.

Any ideas? I would appreciate any help.
 

Cableman

Member
Dec 6, 2017
78
73
91
Alright, I have to decide what to do. My thoughts are that the problem is either:

1) faulty GPU
2) faulty PSU
3) software/BIOS issue

The PC keeps running without any problems unless I either run Heaven or a game. It then crashes within a few seconds. No weird noise, no other issues, no overheating. I'll try the 1070 and Heaven tomorrow, maybe if it's the PSU that will stress it enough to repeat the problem. A clear reason would be nice.

Could the GPU be faulty given all the info in my posts? I got the 7900xt from work, they bought it at Best Buy over 30 days ago (I didn't have time to swap it until today) so I'm likely past the return window. If you think it could still be the GPU, I can try and return it for a replacement. Is a faulty GPU possible given the symptoms?

If it's the PSU, I'll have to order a new one soon. The PC is stable and I don't see any other issues so I could wait a bit. Maybe get a ATX 3.0 PSU...

Finally, software. Could there be some compatibility issue despite removing drivers with DDU?
 

blckgrffn

Diamond Member
May 1, 2003
9,110
3,028
136
www.teamjuchems.com
I am voting PSU. If you have access to Best Buy, they frequently have solid Corsair units 750 or 850 gold rated in store. Match to Amazon, but it local. Put it in and test, you’ll know within minutes probably if you did the right thing. And you can basically do it same day.

Worst case you ate returning both PSU and rma’ing gpu. But maybe your failing PSU is making your GPU situation worse. I would quit poking the bear. GPU rma is a bit of crap shoot, you might be getting someone else’s repaired card and that’s not high on my list of things to want.

I’ve had one bad new GPU out of the box but several flake psus out of the box and even more failures.
 

Leeea

Diamond Member
Apr 3, 2020
3,599
5,340
106
OK, so I tried all of the above.

Set the power limit to -10%. Crashed.

Lowered the fan speed to the minimum %. It didn't go past 73 degrees before crashing so it didn't even thermal throttle before crashing.
RMA the GPU. Tell your retailer it is defective. ( It literally doesn't work, and your last one did. )


You just bought it, it goes back to the retailer.


Up to you if you exchange it for another (new in box) or just get your money back.

-------------------------------------------

Older Seasonics are known to have issues with transients as they say, but I no longer feel that is your problem. The Seasonic transient issue is related mostly to rtx3090s and 3080s. Those cards have transient loads over 550 watts. PSU crashes from overload protection typically throw a thermal breaker in the PSU, and they typically take a minute or two for the PSU to reset itself before the PC will power up again.

You have not mentioned anything like that, just black screen. I think you have a bad GPU.

Defective products happen to every manufacturer. Your part of the 1.8% who were unlucky.


My 2 cents.
 
Last edited:
  • Like
Reactions: Cableman

Leeea

Diamond Member
Apr 3, 2020
3,599
5,340
106
Worst case you ate returning both PSU and rma’ing gpu. But maybe your failing PSU is making your GPU situation worse. I would quit poking the bear. GPU rma is a bit of crap shoot, you might be getting someone else’s repaired card and that’s not high on my list of things to want.
He just bought it.

It will be going back to retailer, not manufacturer. RMA is not someone else's repaired card for him.

Retailer will either give him his money or another new in the box replacement. It will be up to him.
 

Cableman

Member
Dec 6, 2017
78
73
91
one last question:

For the 8 pin connectors on the GPU, did you use two separate cable bundles coming to the GPU, or did you daisy chain one cable to both connectors?

( it should be two separate 8 pin cable bundles )


I'm using 2 separate 8 pin cables. Also, the PC can be turned back on right away after the black screen and the first second or so after the black screen, I still hear some audio.
 

blckgrffn

Diamond Member
May 1, 2003
9,110
3,028
136
www.teamjuchems.com
He just bought it.

It will be going back to retailer, not manufacturer. RMA is not someone else's repaired card for him.

Retailer will either give him his money or another new in the box replacement. It will be up to him.

He said it might already be out of the return period. Apparently, his work bought it and he got it sometime later. This probably makes returns or rmas more complicated.

Testing a PSU, even if it’s 50/50, is low risk and straight forward, you really don’t even need to take your old PSU for a quick test. The only easier thing to test would be if he had a solid second PC or workstation to at least try it in.
 
Last edited:
  • Like
Reactions: Leeea

Cableman

Member
Dec 6, 2017
78
73
91
He said it might already be out of the return period. Apparently, his work bought it and he got it sometime later. This probably makes returns or rmas more complicated.

Testing a PSU, even if it’s 50/50, is low risk and straight forward, you really don’t even need to take your old PSU for a quick test. The only easier thing to test would be if he had a solid second PC or workstation to at least try it in.
Unfortunately, I am outside the return window (checked this morning).

I did put the 1070 back in and ran Heaven without problems. Temperature went way higher than the 7900xt. No black screens. I don't know if it's because it draws so much less power or if the 7900xt is, in fact, faulty.
 

Cableman

Member
Dec 6, 2017
78
73
91
Next thing to try is to limit the power on the 7900 XT. See at what power level the black screens stop happening. If you still get black screens regardless of power, the GPU may be faulty. But if the GPU is totally stable at, say 60% power, that may point to an inadequate PSU.
Is this the same as "power tuning"? I can only move the slider to -10%. Where can I set it to 60%? I was getting crashes at -10, but it's worth trying other settings.

Or is that undervolting with 60%?
 

blckgrffn

Diamond Member
May 1, 2003
9,110
3,028
136
www.teamjuchems.com
Is this the same as "power tuning"? I can only move the slider to -10%. Where can I set it to 60%? I was getting crashes at -10, but it's worth trying other settings.

Or is that undervolting with 60%?

You can use Radeon Chill and set the cap FPS to like 30 fps. Use the Radeon Overlay (ctrl-shift-o, I believe) and that will show you the power usage. The memory is likely to go full speed but you'll see the power usage of the card.

This is a very effective way of limiting the amount of power the core uses.

The 7900XT likely pulls up to 2x the power of the 1070, and maybe even more under some circumstances - heaven isn't even using all the hardware the 7900XT has. If you have a multi rail PSU, it might just be a weak one. No matter how you play with it, I am guessing we won't really know until you have a different 7900 or a different PSU to test with. The rest of this is likely just moving the deck chairs around.

What is the model of your PSU, specifically?
 

Cableman

Member
Dec 6, 2017
78
73
91
OK, I set up everything quite low. Undervolted to 64% (lowest setting) , set max frequency to 90%, and set the power limit to -10% (lowest setting). See attached. Power never exceeded 308 W. Almost got to the end of Heaven, but crashed again. I think it might actually be a faulty GPU...20230207_093654.jpg
 

In2Photos

Golden Member
Mar 21, 2007
1,599
1,637
136
Is this the same as "power tuning"? I can only move the slider to -10%. Where can I set it to 60%? I was getting crashes at -10, but it's worth trying other settings.

Or is that undervolting with 60%?
You would likely have to limit the frequency to keep the power down since the power slider is limited. You could also try undervolting to reduce power but I would only do this if the system were stable to start with. Reducing voltage on an unstable system could cause it to be more unstable. So try dropping the max frequency 100MHz at a time and see if you can get it to run without crashing.
 

In2Photos

Golden Member
Mar 21, 2007
1,599
1,637
136
OK, I set up everything quite low. Undervolted to 64% (lowest setting) , set max frequency to 90%, and set the power limit to -10% (lowest setting). See attached. Power never exceeded 308 W. Almost got to the end of Heaven, but crashed again. I think it might actually be a faulty GPU...View attachment 76065
Put the voltage back to 100. Enable Advanced Control just below the voltage reading. This will give you values instead of % to tune with. Set max frequency to 2600 and benchmark it again.
 

blckgrffn

Diamond Member
May 1, 2003
9,110
3,028
136
www.teamjuchems.com
OK, I set up everything quite low. Undervolted to 64% (lowest setting) , set max frequency to 90%, and set the power limit to -10% (lowest setting). See attached. Power never exceeded 308 W. Almost got to the end of Heaven, but crashed again. I think it might actually be a faulty GPU...View attachment 76065

Undervolting blindly might make it less stable at anything near boost clocks.

Yeah, it might be faulty but I would expect a lot of cards to fail if you dialed them back like that.
 

In2Photos

Golden Member
Mar 21, 2007
1,599
1,637
136
I just went on Seasonic's website and ran their wattage calcualtor. I'm not sure what CPU you have or the number of fans, drives, or other components, but the minimum size PSU they showed for an R7 7700X and 7900XT is an 850W.

 
  • Like
Reactions: Cableman

Leeea

Diamond Member
Apr 3, 2020
3,599
5,340
106
He said it might already be out of the return period. Apparently, his work bought it and he got it sometime later. This probably makes returns or rmas more complicated.

Testing a PSU, even if it’s 50/50, is low risk and straight forward, you really don’t even need to take your old PSU for a quick test. The only easier thing to test would be if he had a solid second PC or workstation to at least try it in.
oh!

That changes things.

Makes RMA much uglier.


It will be up to him, swap GPU or swap PSU.
 
  • Like
Reactions: Cableman

Cableman

Member
Dec 6, 2017
78
73
91
Put the voltage back to 100. Enable Advanced Control just below the voltage reading. This will give you values instead of % to tune with. Set max frequency to 2600 and benchmark it again.

OK, something new to report! I did the above, set the max frequency to 2600 and ran Heaven again. Max power usage never went above 289 W and I managed to complete the benchmark! I'll try it a few more times and will try games later in the day (when I have more time).
 

Cableman

Member
Dec 6, 2017
78
73
91
I just went on Seasonic's website and ran their wattage calcualtor. I'm not sure what CPU you have or the number of fans, drives, or other components, but the minimum size PSU they showed for an R7 7700X and 7900XT is an 850W.

Just checked that too (did that before getting the GPU). I'm with an i7-8700 (not K). 2 SSDs and a HDD. It recommended 750w minimum, but maybe I'm right at the limit there?
 

Leeea

Diamond Member
Apr 3, 2020
3,599
5,340
106
OK, something new to report! I did the above, set the max frequency to 2600 and ran Heaven again. Max power usage never went above 289 W and I managed to complete the benchmark! I'll try it a few more times and will try games later in the day (when I have more time).
-blinks-
Interesting.

-waits for further updates-

note:
capping the GPU speed that way does not rule the GPU out as the faulty component. You are effectively under clocking the GPU to reduce the wattage. But this also increases stability on the GPU by reducing clocks.

If it fails like this, it is probably a GPU failure.
If it works fine, could be either GPU or PSU.
 

In2Photos

Golden Member
Mar 21, 2007
1,599
1,637
136
Just checked that too (did that before getting the GPU). I'm with an i7-8700 (not K). 2 SSDs and a HDD. It recommended 750w minimum, but maybe I'm right at the limit there?
So both AMD and Seasonic have you at the minimum recommended PSU size. I think this is your issue. If you can drop power slightly and the card runs fine this validates that as well, but the only way to truly know for sure would be to test with a larger PSU and see if the card can run at full power.

For reference, my daughter's new PC has an R5 7600X and 6800XT. I went with a 750W PSU. My new build has an R7 7700 and 6800XT and I went with an 850W.

It is possible you could tune your card to reduce the power usage, but still retain most of the performance and it be stable with your 750W PSU, but it would take some fiddling for sure.
 
  • Like
Reactions: Cableman

blckgrffn

Diamond Member
May 1, 2003
9,110
3,028
136
www.teamjuchems.com
@Leeea is not wrong, and you could embrace Radeon Chill and capped clocks but...

If either the card is bad or your PSU is borderline things could get weird. You might be fine, but temperature of your computing room suddenly matters. Ambient air increases lower the efficiency of PSU and raise the power needed to run given clock speed, not to mention it might dirty your power a bit with some AC or fans and you'll get more weirdness.

I'd still test a PSU from a local store to see if you can run full clocks. If you can, great, you figured it out! If not, then you can likely return that PSU and then take your time figuring out the RMA. Depending on your card, you might need to RMA the card through your company as the warranty might not be transferable - but if you have a receipt, now would be an excellent time to register it to your name. If you don't have the receipt, now is a great time to fire off that email to see if you can get it.
 
Last edited:
  • Like
Reactions: Leeea and Cableman

thecoolnessrune

Diamond Member
Jun 8, 2005
9,670
571
126
Personally, I would also reinstall Windows before messing too much with throwing parts at the problem. You mentioned you did a DDU switching from an NVIDIA GPU to an AMD GPU. Especially if this is an original install from back in the 1070 days, I would start with a clean install of Windows (and making sure your BIOS / UEFI is up to date). At the very minimum it should bring you a faster, refreshed experience.
 
  • Like
Reactions: Makaveli

Stuka87

Diamond Member
Dec 10, 2010
6,240
2,559
136
If the PSU is having issues (and sorry if this was already tried), I would do two things.

1: Set the power limit to minimum (I think I saw -10%)
2: Enable undervolting. There is an "auto undervolt" setting. Or, you can manually drop the voltage down.
3: Disconnect anything else that is powered up that is not needed in the system.
4: Try a different power plug in your house. You may have an issue where that plug doesn't have enough current, or it has dirty power, which will become a bigger issue as current draw increases.

Find a friend with a machine to test the GPU in.
 
  • Like
Reactions: Cableman