Is increasing the Power Limit good or bad for stability for Nvidia cards?

[Hans Power]

Just trying to get everything 100% stable. Right now it works with the vast majority of all games with +200 GPU & +500 mem clock on my MSI RTX 2070 Armor but I still get the occasional driver crash on a few games - notably ones which use Raytracing like Mechwarrior 5 and Control. Right now I got the Power limit cranked up to the max (114%), which definitely increases heat as well as performance slightly but does this increase or decrease stability? Also, does the voltage slider increase stability at the cost of heat like it would on a CPU?

[tishous]

22 minutes ago, Hans Power said:

which definitely increases heat as well as performance slightly but does this increase or decrease stability?

Higher temperatures means lower stability. The extra power limit just basically allows the card to boost higher (as long as you have temperature headroom as well) since it can use more power. I wouldn't say it directly affects stability, although stability may be affected because of it.

24 minutes ago, Hans Power said:

Also, does the voltage slider increase stability at the cost of heat like it would on a CPU?

It's not like a CPU, but if you are voltage limited then changing the slider may allow more performance/better stability. As far as I know, and through doing my own tests, the only way the voltage slider would cause more heat is if you're limited by voltage and your card isn't using as much power as it should be able to, which would mean it uses more power after changing the slider.

[Hans Power]

2 minutes ago, tishous said:

Higher temperatures means lower stability. The extra power limit just basically allows the card to boost higher (as long as you have temperature headroom as well) since it can use more power. I wouldn't say it directly affects stability, although stability may be affected because of it.

It's not like a CPU, but if you are voltage limited then changing the slider may allow more performance/better stability. As far as I know, and through doing my own tests, the only way the voltage slider would cause more heat is if you're limited by voltage and your card isn't using as much power as it should be able to, which would mean it uses more power after changing the slider.

I see, well during stress testing the GPU it maxes out at around 78°C with the power limit cranked up. It definitely does get power limited at maximum load according to GPU-Z and it gets significantly hotter but it also clocks higher. So, can I assume that letting the GPU clock higher at full load by increasing the Power Limit may reduce stability?

[tishous]

1 minute ago, Hans Power said:

I see, well during stress testing the GPU it maxes out at around 78°C with the power limit cranked up. It definitely does get power limited at maximum load according to GPU-Z and it gets significantly hotter but it also clocks higher. So, can I assume that letting the GPU clock higher at full load by increasing the Power Limit may reduce stability?

Yes, kind of, but it's not so black and white. Higher power may lead to higher boost clocks, which may hurt stability. But, at the stock power limit, you probably won't be able to overclock too much because it's so power limited and can't get to the higher clocks so you typically always want to increase power the most you can.

[Hans Power]

2 minutes ago, tishous said:

Yes, kind of, but it's not so black and white. Higher power may lead to higher boost clocks, which may hurt stability. But, at the stock power limit, you probably won't be able to overclock too much because it's so power limited and can't get to the higher clocks so you typically always want to increase power the most you can.

Yeah, but won't letting it boost higher due to a higher power limit decrease stability by simply allowing the higher maximum clock speeds? Or would the higher power limit actually increase stability because the card IS able to draw more power? It seems to run longer till I get a driver crash with MW5 with stock power limits - however that might also indicate that my GPU clock is just too high so when I let the card "stretch it's legs" it starts to stumble with the increased power limit. I'm just not quite sure how things interact with GPU Boost 2.0.

[tishous]

6 minutes ago, Hans Power said:

Yeah, but won't letting it boost higher due to a higher power limit decrease stability by simply allowing the higher maximum clock speeds?

Theoretically, yes. Because you can't increase voltage, it means the higher frequencies acheived by being able to draw more power happen at the same voltages which would decrease stability.

 

In my experience, just having a higher power limit increases performance quite a bit whilst not sacrificing any stability, so I wouldn't say it's a direct cause of instability. It's only when you manually add clocks on top of that is when you start seeing it become unstable, so I think lowering the clocks by 10mhz or so until you don't crash and leaving the power limit at max would be your best bet in being fully stable.

[Hans Power]

1 minute ago, tishous said:

Theoretically, yes. Because you can't increase voltage, it means the higher frequencies acheived by being able to draw more power happen at the same voltages which would decrease stability.

 

In my experience, just having a higher power limit increases performance quite a bit whilst not sacrificing any stability, so I wouldn't say it's a direct cause of instability. It's only when you manually add clocks on top of that is when you start seeing it become unstable, so I think lowering the clocks by 10mhz or so until you don't crash and leaving the power limit at max would be your best bet in being fully stable.

Yeah, that would make the most sense. I gonna try with +190Mhz next. Can't be that much since it runs like 95% stable already, but those 5% start to get on my nerves.

[tishous]

1 minute ago, Hans Power said:

Yeah, that would make the most sense. I gonna try with +190Mhz next. Can't be that much since it runs like 95% stable already, but those 5% start to get on my nerves.

Yeah they always do annoy me. It's always very specific scenes in games or something like that. Have you tried increasing the clocks on your memory any more than it's at right now?

[Hans Power]

2 minutes ago, tishous said:

Yeah they always do annoy me. It's always very specific scenes in games or something like that. Have you tried increasing the clocks on your memory any more than it's at right now?

Right now it sits at +500. Decreased it to +300 but it still crashed the driver, so I don't think it has an impact on stability. I think I even had the mem clocks at +600 initially and it seemed to work fine. Don't need to push it, though - 100mhz more or less on the memory doesn't seem to have a huge performance impact anyway.

But that begs the question if increasing MEM clocks eats at the power limit and if it should be decreased so that the GPU can clock higher instead (if the silicone is able to, that is).

[tishous]

1 hour ago, Hans Power said:

Right now it sits at +500. Decreased it to +300 but it still crashed the driver, so I don't think it has an impact on stability. I think I even had the mem clocks at +600 initially and it seemed to work fine. Don't need to push it, though - 100mhz more or less on the memory doesn't seem to have a huge performance impact anyway.

But that begs the question if increasing MEM clocks eats at the power limit and if it should be decreased so that the GPU can clock higher instead (if the silicone is able to, that is).

Well I would say the memory has as much impact on performance as the core, possibly even more so in memory intensive situations like gaming at 1440p ultra.

[Falkentyne]

Since you don't have direct voltage control outside of OC LAB or Kingpin cards, or hard mods like Elmor's device, stability is going to be based on how cool you keep the card.  Throw a card on a loop and watch your benchmark scores rise and your boost clocks increase.  At that point, the power limit actually starts hurting you, and then it's time to look into stuff like the EVC2 device or shunt mods.

 

The voltage slider in MSI Afterburner no longer increases voltage since Turing.  It just 'unlocks' a higher boost table on the V/F curve (meaning while the max voltage allowed may increase a tier, the clock speeds will also be higher).  For example if the absolute max at 0% slider was 1.062v, then 100% slider would unlock 1.069v-1.1v (depending on SKU), but the clocks would also rise by 13-15 mhz (depending on tick, all voltage increases are in certain steps, example on Ampere, it's +15 mhz steps, so setting +15 mhz and +28 mhz is the exact same thing (you would need +30 mhz to reach that next tier).

 

If you're on air and stock cooling without shunt mods, don't even bother with that slider.

 

I've also seen some people with a 2080 Ti who posted here, lose stability without an overclock, just by increasing the power limit.  This shouldn't happen and I have no idea why.  

[Hans Power]

6 hours ago, Falkentyne said:

Since you don't have direct voltage control outside of OC LAB or Kingpin cards, or hard mods like Elmor's device, stability is going to be based on how cool you keep the card.  Throw a card on a loop and watch your benchmark scores rise and your boost clocks increase.  At that point, the power limit actually starts hurting you, and then it's time to look into stuff like the EVC2 device or shunt mods.

 

The voltage slider in MSI Afterburner no longer increases voltage since Turing.  It just 'unlocks' a higher boost table on the V/F curve (meaning while the max voltage allowed may increase a tier, the clock speeds will also be higher).  For example if the absolute max at 0% slider was 1.062v, then 100% slider would unlock 1.069v-1.1v (depending on SKU), but the clocks would also rise by 13-15 mhz (depending on tick, all voltage increases are in certain steps, example on Ampere, it's +15 mhz steps, so setting +15 mhz and +28 mhz is the exact same thing (you would need +30 mhz to reach that next tier).

 

If you're on air and stock cooling without shunt mods, don't even bother with that slider.

 

I've also seen some people with a 2080 Ti who posted here, lose stability without an overclock, just by increasing the power limit.  This shouldn't happen and I have no idea why.  

Well, the whole thing might also be just a driver or engine issue. I know with Control there is a registry fix which prevents driver crashes when overclocked, which actually works, so I'm not quite sure if I put it correctly when I call it "stability" issues. But whatever the cause the driver crashes - they are related to overclocks and they gotta go.

I think some games are extremely sensitive when it comes to overclocks while others don't care. I mean, I can stress test for hours without any error found, after all with my original overclocks. Right now I'm trying to find an overclock which works with everything, no exceptions while going as high as possible even with this one last stupid game.