Question Overvolt and Power Percentage Increase Causing Shut Downs? or Faulty Component?

[EvilScotsman]

I upgraded to an EVGA RTX 3070 XC3. Temps are amazing - 60c max.

However, I have experienced random shut downs while gaming (3D Mark not enough to cause shut downs but happened in 2 games so far). The PC shuts off abruptly, PSU needs turned off and on to reboot. (some lights still active, such as USB)

I had the power limit at 113% and core voltage at +100 in MSI Afterburner.

Since disabling these I haven't noticed shut downs. (Still got more testing to do though). Assuming it doesn't shut down with further testing, is this normal? Or a sign of a faulty card? I don't have another mobo or PSU to test these components. But PSU draws 380w max while gaming, even when stressing CPU and GPU with stress tests simultaneously, it draws 450w max.

I had a Palit RTX 3070 before the EVGA (power limit and voltage increased on that as well), but something was wrong with its thermals so I returned it. I never experienced shut downs with the Palit card, however, I have noticed the EVGA card draws 280w max, compared to the 250w max of the Palit card.

Power supply is Seasonic SS-650KM3 X Series. Using 2 separate 8 pin cables to power the card.

Would really appreciate some help with this.

 

[hotaru.hino]

You're asking the card to run over its limits specified at the firmware. So I would expect issues to happen because you're asking it to run in a condition above its spec that it wasn't tested to run at.

So no, it's not a sign of a faulty GPU, especially if it runs fine at default settings.

 

[Phaaze88]

I had the power limit at 113% and core voltage at +100 in MSI Afterburner.

The only way to make changes to voltage is through Curve Editor. The voltage slider doesn't do anything, as the boost algorithm ignores it. I can't tell you why they don't just remove it though...

But PSU draws 380w max while gaming, even when stressing CPU and GPU with stress tests simultaneously, it draws 450w max.

Software doesn't pick up transient spikes, which Ampere is notorious for.
The higher power draw is going to lead to worse spikes, which the psu doesn't seem to be keeping up with.
Well, that's my guess anyway.

 

[vicioussss]

Wow eventually i found another with the same problem like me

I use seasonic x650 gold (ss650km) its 5 years old PSU

Cpu amd 5700x
Gpu asus tuff RX6800 non XT
Power draw max 450watt during stress test,
Gaming 400 watt

I experience the same problem when i OC the GPU, its causing the pc to shutoff then i need to unplug the power cord.

Using 3D mark time spy test i can get the score, strest test also pass 20 cycle.
Cinebench, Aida its all pass without shutoff.

But when i play BF2042 ( yeah <Mod Edit> game i know) my pc would off in random time, sometimes i can play for 2-3 hours, on another time only can play in very short like 1-2 rounds

I use y splitter 8 pint to 2x 6+2 pin to power up ( right now i was thinking this is the problem and order another 8 pin connector), but looking from your case i just loosing hope and start to think maybe its the PSU itself.

 

[EvilScotsman]

The only way to make changes to voltage is through Curve Editor. The voltage slider doesn't do anything, as the boost algorithm ignores it. I can't tell you why they don't just remove it though...


Software doesn't pick up transient spikes, which Ampere is notorious for.
The higher power draw is going to lead to worse spikes, which the psu doesn't seem to be keeping up with.
Well, that's my guess anyway.

I'd totally forgotten about transient spikes because, it wasn't relevant to me at the time I heard about it. Came across gamersnexus video again and, yeah, I'm pretty much certain that is the cause at this point.

Do you happen to have a good source that discusses the voltage slider not doing anything with RTX GPU's? I'd like to learn more about this.

 

[EvilScotsman]

Wow eventually i found another with the same problem like me

I use seasonic x650 gold (ss650km) its 5 years old PSU

Cpu amd 5700x
Gpu asus tuff RX6800 non XT
Power draw max 450watt during stress test,
Gaming 400 watt

I experience the same problem when i OC the GPU, its causing the pc to shutoff then i need to unplug the power cord.

Using 3D mark time spy test i can get the score, strest test also pass 20 cycle.
Cinebench, Aida its all pass without shutoff.

But when i play BF2042 ( yeah <Mod Edit> game i know) my pc would off in random time, sometimes i can play for 2-3 hours, on another time only can play in very short like 1-2 rounds

I use y splitter 8 pint to 2x 6+2 pin to power up ( right now i was thinking this is the problem and order another 8 pin connector), but looking from your case i just loosing hope and start to think maybe its the PSU itself.

Make sure you are not using splitters/daisy chains. Separate cables for each power input with newer cards. AMD cards aren't as bad for transient power spikes as Nvidia, so I'm confident that if you use 2 different cables to power your GPU, you won't get the shutdowns anymore. If your PSU is multi rail, it's also probably beneficial to have each cable coming from a different rail. However, I think multi rail PSU's are generally only at the higher end. My seasonic is higher end, though quite old now. And I'm sure it only has 1 rail.

If that doesn't work, you should still be able to OC the card, just leave voltage and power limit at default. This SEEMS to have fixed my issue for now.

When I got my Palit RTX 3070 which I had to return, initially I used one cable to power it and was getting shut downs. After changing to two cables, never had that issue again even with power limits and voltage increased. As I said in my post though, the EVGA card seems to draw about 30w more under load which might be what's making the transients worse, either that or the EVGA is just worse for transients in general.

 

[joan16v]

It should be your PSU, it's not capable of mantain the power of the components. You will need a most powerful one (and remember to use only quality brands like Corsair, Seasonic, Bitfenix and 80+ Gold ones).

 

[faalin]

The 3070's have power spikes, not as bad as the 3080 and 3090's but its still enough to trip the 12v rail on your PSU. I would recommend moving up to a 750 or 850w PSU to give it more headroom for the spikes.


My 2080Ti recommends a 600w PSU, was using a 750W PSU and was still having issues with power spikes tripping the PSU. Luckily i had a 1600W from a previous build i put in and havent had issues since....yeah its overkill but i had it so i used it.

 

[Phaaze88]

Do you happen to have a good source that discusses the voltage slider not doing anything with RTX GPU's? I'd like to learn more about this.

Personal experience.
Use of the slider acts as more of a 'suggestion' to the gpu that it can use more voltage to pursue higher boost clocks, but the gpu boost algorithm pays it no mind. It won't ignore Curve Editor though.
[The slider worked on GTX 700 series and lower. I can't remember if it worked on GTX 900 or not.]
Besides, more voltage = more heat(temperature) and power, two parameters the boost algorithm is constantly monitoring, and it dynamically adjusts gpu boost clocks based on that.

 

[hotaru.hino]

Do you happen to have a good source that discusses the voltage slider not doing anything with RTX GPU's? I'd like to learn more about this.

The GPU will boost until it hits one of the following reasons:

• Hit the power limit
• Hit the reliable voltage limit (based on the V-F curve, I believe, but it could be based on actual hardware monitoring)
• Hit the absolute voltage limit
• Hit the temperature limit
• There's not enough stuff to do

Most monitoring tools that give video card stats will pull this from NVIDIA GPUs. But in any case, if the GPU hits any of these limits, it doesn't matter if the GPU isn't honoring some other point.

Honestly though, when you give this much control over things that could blow up the GPU, I'd rather have the GPU think of what the user wants as suggestions.

 

[EvilScotsman]

Personal experience.
Use of the slider acts as more of a 'suggestion' to the gpu that it can use more voltage to pursue higher boost clocks, but the gpu boost algorithm pays it no mind. It won't ignore Curve Editor though.
[The slider worked on GTX 700 series and lower. I can't remember if it worked on GTX 900 or not.]
Besides, more voltage = more heat(temperature) and power, two parameters the boost algorithm is constantly monitoring, and it dynamically adjusts gpu boost clocks based on that.

JayzTwoCents did a video where he mentioned it apparently does something...sometimes. lol. I think the idea is that it allows higher voltages sustained for longer, without actually increasing the voltage. (I've yet to test it myself).

However, I decided to try upping my power limit again, but left voltage. So far, no shut downs, and it allows me to run my GPU clock a little bit higher.

So, assuming no more shutdowns. Seems the voltage slider might do something. Though what I have no idea, and it doesn't help with stability for me anyway, power limit does though.

 

[EvilScotsman]

The 3070's have power spikes, not as bad as the 3080 and 3090's but its still enough to trip the 12v rail on your PSU. I would recommend moving up to a 750 or 850w PSU to give it more headroom for the spikes.


My 2080Ti recommends a 600w PSU, was using a 750W PSU and was still having issues with power spikes tripping the PSU. Luckily i had a 1600W from a previous build i put in and havent had issues since....yeah its overkill but i had it so i used it.

Damn, that is overkill haha.

Turning off my core voltage and power limit seems to have solved the issue. I've actually had the power limit back at max and still had no shut downs. So fingers crossed it was just the combo of power limit and voltage.

However, my next PSU will likely be made overkill as well just to be safe.

I've also read that certain motherboards handle transient spikes better than others. However according to gamersnexus, ensuring you have a beefy enough PSU should mitigate the issue, even if it is technically the motherboard that's struggling.

 

[Phaaze88]

JayzTwoCents did a video where he mentioned it apparently does something...sometimes. lol. I think the idea is that it allows higher voltages sustained for longer, without actually increasing the voltage. (I've yet to test it myself).

However, I decided to try upping my power limit again, but left voltage. So far, no shut downs, and it allows me to run my GPU clock a little bit higher.

So, assuming no more shutdowns. Seems the voltage slider might do something. Though what I have no idea, and it doesn't help with stability for me anyway, power limit does though.

That's a real short sample size though, considering you had under 12 hours to play around with it(as of the last post) and 3D Mark isn't an end-all-be-all test.

Everyone's gpu has a boost target, but that number varies. They don't dynamically boost to infinity - not that they can to begin with. 80/100/120% power target, that number doesn't change. My 1080Ti's target is 1949mhz.
Boost will allow that target to be sustained for longer, if the card doesn't run into the power limit(or less frequently) and operating temperature is as cool as possible - these cards function like Ryzen cpus with their several thermal thresholds, where one can see higher/lower boost clocks on warmer/cooler days.
GTX 10 - RTX 30 are power limited; board power limits are on the conservative side. RTX 40 is actually voltage limited.

That extra voltage needs stronger cooling to sustain... running a few degrees warmer can see Gpu Boost dial back 15mhz or so on occasion - that's including the card not hitting the board power limit.

 

[EvilScotsman]

That's a real short sample size though, considering you had under 12 hours to play around with it(as of the last post) and 3D Mark isn't an end-all-be-all test.

Everyone's gpu has a boost target, but that number varies. They don't dynamically boost to infinity - not that they can to begin with. 80/100/120% power target, that number doesn't change. My 1080Ti's target is 1949mhz.
Boost will allow that target to be sustained for longer, if the card doesn't run into the power limit(or less frequently) and operating temperature is as cool as possible - these cards function like Ryzen cpus with their several thermal thresholds, where one can see higher/lower boost clocks on warmer/cooler days.
GTX 10 - RTX 30 are power limited; board power limits are on the conservative side. RTX 40 is actually voltage limited.

That extra voltage needs stronger cooling to sustain... running a few degrees warmer can see Gpu Boost dial back 15mhz or so on occasion - that's including the card not hitting the board power limit.

I'm aware how GPU boost works. Temps have nothing to do with what was happening to my GPU.

Not really sure what any of that has to do with what I was saying tbh.

12 hours doing the same things that made it shut down in a couple hours is strong evidence that something's improved/fixed. I get what you mean by saying sample size, but in this scenario it's not really applicable, there's no sample size, just more 'data' which indicates whether the problem is solved or just improved, but, I think that's what you meant.

No shutdowns as of yet. Higher voltage draws more power, so voltage + power limit increased in tandem was obviously pushing the transient spikes too far for my PSU.