3090 zotac Trinity Oc dead or psu weak?

[bug]

try looking into 3.3V monitoring on the Hwinfo64.

I have a combo of Seasonic PSU and Asus MoBo (several years older then yours though) and running in the issue of the 3.3V is running as low as 2.9 during 350W+ GPU load. MoBo detects it and shuts down the system.

The actual reason is quite quircky, since there is NO any high load on 3.3V line. I suppose that the PSU tries to use too clever wire-voltage-drop compensation for 12V that somehow negatively affect the 3.3 levels.

When I got those low 3.3 readings - I reseat the main 24-pin connector several times (both sides, MoBo and PSU) and after this the issue disappeared

Agreed. Monitoring is the only thing that can help pinpoint the real problem.

 

[lexluthermiester]

If the PSU is too weak, the PC should shut down or restart.

Not always. For example, if the PSU the OP has is a dual rail model and the 3090 is only connected to one of those rails, the PSU would not shut down instead the 3090 would be power starved and become unstable. This is a well known problem. If that is the situation then the OP just need to make sure that one power cable is connected to one rail and the other power cable is connected to the second rail.

 

[bug]

Not always. For example, if the PSU the OP has is a dual rail model and the 3090 is only connected to one of those rails, the PSU would not shut down instead the 3090 would be power starved and become unstable. This is a well known problem. If that is the situation then the OP just need to make sure that one power cable is connected to one rail and the other power cable is connected to the second rail.

Tbh, the issues a PSU can cause are so diverse, they should come with built-in displays for monitoring their status/output.

 

[lexluthermiester]

the issues a PSU can cause are so diverse

Very true!

 

[Ruru]

Your PSU should be enough, I have a 3080 and runs fine with more hungry system. Your CPU draws less juice than my AMD 5800X.

Didn't read the whole thread, sorry.

 

[Veseleil]

Judging by your input so far, my guess is that the GPU VRAM is wasted. Advice? Get your money back and don't buy used GDDR6X cards if you can't be sure they've done zero computing work(load). Overheating GDDR6X chips are prone to a faster degradation under heavy loads.

 

[lexluthermiester]

Judging by your input so far, my guess is that the GPU VRAM is wasted.

Except the OP didn't mention any artifacting. VRAM failure would cause that and the OP would have mentioned it. The screenshots they provided in the first post don't show any artifacting either. All signs point to either the GPU die itself having a fault or the PSU. My vote is the PSU.

 

[bug]

Except the OP didn't mention any artifacting. VRAM failure would cause that and the OP would have mentioned it. The screenshots they provided in the first post don't show any artifacting either. All signs point to either the GPU die itself having a fault or the PSU. My vote is the PSU.

Especially since the seller tested the card with a similar PSU and it worked just fine.

 

[dzmdd]

If the PSU is too weak, the PC should shut down or restart.
What is this crash you are speaking of? I've seen crashes refer to shut down, restart, bsod, driver timeout, etc. Try to be more specific.

Indeed Pc restarted many times and posted in Safe Mode after. I'm waiting for a bigger PSU actually a 1000w PSU. I'll see if PSU is the problem. Problems are as followings.
- One time BSOD
- Many other time screen flickering under full charge load or purely crash.
Problems appears when I charge big models (it's load at 99% Core and something like 74 % of Vram, so it's in full load for GPU Core) like Hunyuan video in fp8( a 11gb models on the card), FP16 (bigger models) models make GPU crash the same way too.
I ran two OCCT (first one crash at 1sec) the other one worked without error for each. It was Power supply test, here you be able to see screenshot for sensor information from OCCT. I don't really know if memory chips are dead. I ran this test 2 times. VRAM can be faulty i don't really know for the moment but when we ran OCCT with seller we tested 3D Standard, 3D Adaptive and VRAM with combine test and it worked fine. Today in all cases i'll be sure if GPU is really fried like Bankman or if with was a PSU issue ^^. Thanks a lot in all cases guys you rocks
Edit: I know i should probably not buy second hand hardware but many times it worked nicely and i don't have enough money for buy new 5090. For AI workload the only interresting card are 90 series because they have 24 GB VRAM. 4090 are like BBQ with 12HPWR connector too i wanted to dodge this problem but i met another now. Actually my RX6800XT from second hand run nicely. Nvidia GPU's are too expansive since some years, buying not used GPU is completely ouf of my range sadly.

 

[lexluthermiester]

Especially since the seller tested the card with a similar PSU and it worked just fine.

While that's a fair point, sellers can't always be trusted. The card need to be tested with a second PSU to confirm/rule-out the main PSU before we all doom the card.