gtx 780 ti shuts down under load

[iceman_303]

Hello everybody. I recently bought a used msi gtx 780 ti gaming, so i could play recent games with good details. I planned to put the card on a exp gdc beast 8.5, which is a mpcie adapter that connects to your laptop and allows you to use an external graphics card. When i received the card, i put it in my slower desktop for a test. The card uses one 6 pin pcie and one 8 pin pcie power connectors. I plugged them in, but i plugged the 8 pin the other way around, without noticing, I had to force it, but I didn´t realize that at the moment. I powered the pc, but the pc did not turn on. I did it a couple of times, and then i realized the 8 pin was not plugged in correctly. When I received the exp adapter in the mail, i installed it, put the graphics card on it, powered the graphics card and the adapter ( cpu power connector and 21 pin power connector ) with the psu, and it booted up, i installed the drivers, and it worked. But when I run benchmarks, like combustor or so, os some more intensive games, on replays and such, it crashes with a black screen, but i can still hear the sound. I then used msi afterburner to lower the power limit to minimum, (57%) and it runs everything ok, even the benchmarks or games. I then unlocked the voltage control. and got to the max voltage of around 1000mv, with power limit to 100% and default clocks. With those settings, the benchmarks run ok. But if I put the voltage on auto it crashes. The power supply is 450w, 2 twelve volt rails of 15 and 16a each, 24 amps on 5, and 30 on 5v. I even bought another power supply, a 600w nox, with 2 twelve volt rails of around 21 amps each, making a total of 450w on the 12v rails, 24a on 5v, and 32a on 3v. With this power supply, i can only use lower max voltages of 900mv or so, with lower power limits. I have measured the voltage on 12v rail when the card is under load, and it´s very low, almost at 11.1v. One psu doesn´t shut down, but the nox one does. I suspect the card is putting a heavier load on the psu then it should. I have some knowledge in electronics, mostly analog, and i have a multimeter and some other tools. I dont know what is wrong with the card, i've looked at the 780ti pcb just to have an ideia of how everything is layed out. I apologize for the long post, but i tried to describe it as well as i could so it should be easier to troubleshoot. I came here, because I had seen some very good tech advice here some time ago, regarding a 780ti, and I thought I'd ask for some help. The card is a msi gtx 780 ti gaming, my pc is a toshiba laptop satellite l 850 1pd with a i7 3630qm and 4 gigs ram.
Any help is appreciated. Many thanks in advance.

cheers,

joao g.

[Nusa]

What I got from all that was "works as advertised if I run it at normal settings". Perhaps that's all you're going to get out of this "used" purchase.

[iceman_303]

Hello nusa. thanks for replying. I dont think it's properly working. On auto voltage ( default setting ), and 100% power limit ( default ), it crashes.
The stock voltage, i forgot to mention, is around 1.15 v, maybe a little bit more. It only runs stable at around 0.95v, to 1.0v, with 100% power limit.
And I dont think it should shut down the power supply, the power should be enough for the card.

cheers,

jg

[Zucca]

Is everything ok on the thermal side? Overheating?

[wraper]

You likely bought a faulty graphics card. Reversing power connector unlikely has anything to do with this. Card should be completely dead if it received a damage from this. You should return it if possible. If card is factory overclocked (EDIT - it is), you might look for compatible bios with lower clocks. Otherwise it's beyond economic repair.

I then unlocked the voltage control. and got to the max voltage of around 1000mv, with power limit to 100% and default clocks.

You are doing it in a completely wrong way  . You should not overvolt it but try reducing clocks of RAM and GPU.

[wraper]

Is everything ok on the thermal side? Overheating?

Overvolting helps, yet it causes even more heat dissipation. So not very likely. Maybe only if some RAM chip lacks a thermal pad.

[wraper]

I even bought another power supply, a 600w nox, with 2 twelve volt rails of around 21 amps each, making a total of 450w on the 12v rails, 24a on 5v, and 32a on 3v. With this power supply, i can only use lower max voltages of 900mv or so, with lower power limits.

You mean this? https://www.pccomponentes.com/nox-urano-600w
If yes, you have some hunch on buying old junk. I was a piece exceptional crap when was new and by now probably is already faulty.

[wraper]

It supposedly has updated bios version which also might have updated memory timings. You could try it. Ensure that you have exactly the same model https://forum-en.msi.com/index.php?topic=176958.0, look at post #18 for latest bios and post #2 for instructions.

[iceman_303]

hi again guys. thanks for the replies. zucca, the temperatures are ok, around 50, 60ºc. Sometimes it's even colder and it gives the blank screen.
wraper, i think you maybe correct, the card may have a problem, i dont know if it came damaged ( the seller even sold it to me with paypal warranty, but i let the
card sit for a lot of time, before i tried it ), maybe it was. The nox supply as you say, it´s kind of junk, i know, but it's what i could get my hands on at the time.
As it not being economic to repair it, i'm not sure, maybe it's a semi shorted mosfet, or something alike on the supply part of the card. I dont think it is normal for a card to lower the voltage on the 12v rail to lower than 11v, ( with around 450w available on the 12v rail ). Not being rude, but i would at least try to fix it. At least try to figure out what is happening. If nothing should work, i will indeed have to run the card at lower than stock voltages.
thanks again.

[Rasz]

start by testing on good power supply, borrow one, test card in friends computer etc

[wraper]

As it not being economic to repair it, i'm not sure, maybe it's a semi shorted mosfet

There is no such thing as semi shorted power mosfet. Either it works or gets completely shorted and likely blows up a hole in the middle. Possibly burning PCB as well.

[janoc]

Crashes with black screen while under load (doesn't need to be overheating) => dead card.

I had a GTX 970 die on my with these exact symptoms after some 18 months of normal use (mostly programming, occasional gaming, stock settings). It would run when severely underclocked but as soon as the machine warmed up (card at 60C or so tops), the screen would go black and the driver restart. Fortunately I still could get it replaced under warranty. Plenty of unscrupulous people would try to sell it on eBay instead as "working".

This is a fairly common problem - if you google "GeForce black screen", you will find plenty of people complaining about this. There is most likely something on the board that dies with the thermal cycling - maybe some of the voltage regulator capacitors, maybe the GPU itself fails.

BTW, starting to mess with the voltage settings when you don't know what is actually wrong is a terrible idea. That's like troubleshooting a warning light on a car dashboard by randomly reprogramming the ECU and wondering why the entire thing is a mess now.

You have mentioned a low voltage on the 12V rail and that it runs if you increase the chip voltage - that would point to a poor power quality/underpowered power supply. 780TIs are notorious power hogs that run hot, the card alone can take 260W or so. So if you are running the supply close to its limits it will likely be overloaded, with sagging rails and likely large output ripple. Especially many cheap supplies are unable to deliver their sticker ratings (that Nox supply certainly qualifies here!). This GPU + an i7 on a 600W supply is asking for trouble, IMO, especially with a cheap junker like the one you have bought - it likely can barely deliver 450-500W and you are overloading it.

I am not sure I would even bother trying to repair this. It is a long obsolete card, 4 generations back. You have bought it used, who knows how was it used and abused (e.g. if it is from a crypto mining rig, oof) and how many thermal cycles did it have to go through. There could be hidden problems that could be very difficult to find. Plus there is zero documentation for this kind of product and no spare parts - and even if you got them, many components are very difficult to replace without special equipment.

If I was you I would rather send this one back as faulty and buy a cheap GTX1060. Those are way less power hungry (less than half, e.g. 1060 uses only 120W vs the 260W of 780ti) and faster than a 780ti. And it costs around $200, so you won't break a bank neither. Trying to save money by buying used/half-broken and buying cheap junk really doesn't work when it comes to computers. The time you spend fixing/debugging this and the extra stuff you will have to buy (such as a proper brand name power supply) will cost you more than buying a new graphic card.

[iceman_303]

thanks guys. I appreciate all the info from all of you. i will now do as rasz said, i will try it on some other computer. Maybe the psu's i am using for this card are insufficient. I will also try another psu, if i have the chance. Ok janoc, i understand your idea, its a old card, you're right, maybe i'll keep it for a while and then i´ll get a new one.
As soon i try the card on another computer i will let you guys know. many thanks.

regards,

jg

[Rasz]

It is a long obsolete card, 4 generations back.
buy a cheap GTX1060. Those are way less power hungry (less than half, e.g. 1060 uses only 120W vs the 260W of 780ti) and faster than a 780ti. And it costs around $200

-two
-its in the same speed ballpark, 1080p 60fps with good quality in modern games
-$300

[wraper]

This is a fairly common problem - if you google "GeForce black screen", you will find plenty of people complaining about this. There is most likely something on the board that dies with the thermal cycling - maybe some of the voltage regulator capacitors, maybe the GPU itself fails.

From description of the issue it sounds like GPU or RAM no longer can work on factory overclocked speed.

[janoc]

From description of the issue it sounds like GPU or RAM no longer can work on factory overclocked speed.

This is often a power problem - either the main PSU in the machine or the voltage regulators on the board conking out when under load. Could be dead electrolytic caps due to the many thermal cycles and the card systematically running hot but could be something more serious too.

[janoc]

It is a long obsolete card, 4 generations back.
buy a cheap GTX1060. Those are way less power hungry (less than half, e.g. 1060 uses only 120W vs the 260W of 780ti) and faster than a 780ti. And it costs around $200

-two

Kepler, Maxwell, Pascal are at least three, plus Kepler had 3 revisions ...

-$300

https://www.newegg.com/Product/Product.aspx?Item=N82E16814487263&cm_re=gtx_1060-_-14-487-263-_-Product

[wraper]

From description of the issue it sounds like GPU or RAM no longer can work on factory overclocked speed.

This is often a power problem - either the main PSU in the machine or the voltage regulators on the board conking out when under load. Could be dead electrolytic caps due to the many thermal cycles and the card systematically running hot but could be something more serious too.

Do you know that electrolytic caps are very rarely used in modern graphics cards, especially high end? Not a single electrolytic cap on the photo.

This is often a power problem

Often it's just a chip degradation.

[elgonzo]

It is a long obsolete card, 4 generations back.
buy a cheap GTX1060. Those are way less power hungry (less than half, e.g. 1060 uses only 120W vs the 260W of 780ti) and faster than a 780ti. And it costs around $200

-two

Kepler, Maxwell, Pascal are at least three, plus Kepler had 3 revisions ...

Yeah, that makes exactly two. The current microarch for Geforce is still Pascal, so you cannot count Pascal as being one generation back. Wait some months until Turing comes out, then we can continue arguing

[Rasz]

-two

Kepler, Maxwell, Pascal are at least three, plus Kepler had 3 revisions ...

so you agree with me its two

-$300

https://www.newegg.com/Product/Product.aspx?Item=N82E16814487263&cm_re=gtx_1060-_-14-487-263-_-Product

https://www.newegg.com/global/pt/Product/Product.aspx?Item=N82E16814487263

197 euro + 26 euro shipping + ~20 euro customs agent + IVA 23%

[Nusa]

From description of the issue it sounds like GPU or RAM no longer can work on factory overclocked speed.

This is often a power problem - either the main PSU in the machine or the voltage regulators on the board conking out when under load. Could be dead electrolytic caps due to the many thermal cycles and the card systematically running hot but could be something more serious too.

Do you know that electrolytic caps are very rarely used in modern graphics cards, especially high end? Not a single electrolytic cap on the photo.

Actually I see 8 aluminum electrolytic SMD caps in the picture. You're right that there are no through-hole caps in the picture.

[wraper]

Actually I see 8 aluminum electrolytic SMD caps in the picture. You're right that there are no through-hole caps in the picture.

They are not aluminium electrolytic but polymer capacitors. And 6 of them ARE though hole.

[Nusa]

Actually I see 8 aluminum electrolytic SMD caps in the picture. You're right that there are no through-hole caps in the picture.

They are not aluminium electrolytic but polymer capacitors. And 6 of them ARE though hole.

Polymer capacitor is a shortened version of Polymer electrolytic capacitor. In this case, polymer aluminum electrolytic capacitor. One of the many subsets of "electrolytic capacitor". https://en.wikipedia.org/wiki/Electrolytic_capacitor#Electrolytic_capacitors_family_tree

I stand corrected on the through-hole aspect.

[wraper]

Polymer capacitor is a shortened version of Polymer electrolytic capacitor. In this case, polymer aluminum electrolytic capacitor. One of the many subsets of "electrolytic capacitor". https://en.wikipedia.org/wiki/Electrolytic_capacitor#Electrolytic_capacitors_family_tree

I stand corrected on the through-hole aspect.

You can call them conductive polymer electrolytic caps (i knew it many years ago without your wiki link) but they have nothing common with normal electrolytic capacitors and reraly mentioned as polymer electrolytic. They have completely different specs, endurance and failure modes. You called them simply electrolytic caps, so surely you did not mean polymer caps.

[Nusa]

So the problem here is you use a generic term to mean a specific thing or aspect (from context, your "normal" presumably means anything with a wet electrolyte), while I recognize it as a generic term that covers a whole slew of variations, including what's on the board. So in that sense, nobody is wrong, it's just a communication failure.