KingEngineRevUp

chrismmcnamara
Pretty digressive on mine but I am stable at 787 mV at 1750 MHz. Was coming from a 1080 GTX too though so even at that level of underclock/undervolt the performance is pretty insane to me still. Power consumption while gaming varies between 150 at 250 W which is a huge improvement for me. I've also got this in a custom loop with a 10920X with only a single 240mm radiator so the power reduction is important. Fans hit 100% while gaming but the temperatures of both the processor and GPU are about 50-54°C.

That's about -10% in performance. Is it really worth that drop?

DiegoFloresEVGA
ohhh okay makes sense. i just have the ftw3 card. 1440p ultra on heaven benchmark i was seeing 82c with no undervolting (not terrible i know) and when i undervolted it was 78c. while gaming most i have seen is 75c. so I think I am okay. I play at 1440p 144hz so this card will able to handle everything anyways. I could deal with stock but man those fans would get very loud or I would get a slight coil whine during menus 

03whitegsr

KingEngineRevUp
But why not undervolt and OC to the best you can and maybe do 300-350W and not lose performance? Find a middle gournd. 

I'm fairly new to PC gaming, I apologize if this is some well known "don't do"...

But if you use a frame limiter setting in game or through the control panel to match your monitor refresh rate, on games that can run below 100% GPU load, the GPU will run at a lower clock speed automatically. This pushes the operating point to a lower voltage level, and there by reduces the power anyway?

Seems like the ideal way here would be to create a FV table that is at the stability limit at each voltage level? This way you are always at the max frequency for a given power level and the game need will then drive the power draw.

I've been trying to do this but have found two things that make this difficult.

1. Well known, but a FV setting that's stable in benchmarks isn't usually stable in games. This appears to be because even if you have a set frequency at a voltage, in game, it seems to go 1-2 bins higher. For example, I can run 825mV@1890MHz in benchmarks. This curve though will end up spiking to 1920MHz in game though and crashing. If set it two bins lower, 825mV@1860MHz though, in game it will spike to 1890MHz but won't crash. The benchmark point seems to be good, but you have to tweak it to deal with it running higher then it should in game.

2. Afterburner fights you on FV curve settings. Your can set your points you know are stable and then smoothly transition the points in between, but when you hit apply, the resulting curve is something very different. Usually moving your set points several bins up or down, depending on seemingly nothing.

demon09yep what doing -200 -280 or anything big like that it just brings your core clocks on the curve low so that when you bring what ever voltage you want up it will go flat after it but it's like overclocking make sure to stress test

demon09
it doesn't really matter once you move it to your end point at the voltage you pick the rest of the curve moves up

demon09

SleepAgreeable

demon09
it doesn't really matter once you move it to your end point at the voltage you pick the rest of the curve moves up

Thanks for the help. Is playing any game that pushes the card to 99% utilization for couple of hours enough of a stress test? 

try unigen super postion at like 4k for a pass or two. Normally I find where I'm at stable at in heavy synthetic bench marks and then pull back a boost bin or two say I'm stable at 1950 but crash at 1960 I just set 1930 as the head ache of crashes is not worth the 1%

03whitegsr
I don't know if it matters or not, but by setting the clock to -200, it shifts the entire curve down. This lowers the frequency even at the lower voltages below the stock level.

As an alternative, if you hold CTRL (or maybe it's shift??? It's one of the two) while you drag the highest voltage setting down, it pivots the whole curve around the lowest setting and the lower voltages move much less. This should keep the card running at higher clocks when the card isn't running at 100% load. (Or run a lower voltage for a given clock speed...)

chrismmcnamara

KingEngineRevUp

chrismmcnamara

KingEngineRevUp

chrismmcnamara
Pretty digressive on mine but I am stable at 787 mV at 1750 MHz. Was coming from a 1080 GTX too though so even at that level of underclock/undervolt the performance is pretty insane to me still. Power consumption while gaming varies between 150 at 250 W which is a huge improvement for me. I've also got this in a custom loop with a 10920X with only a single 240mm radiator so the power reduction is important. Fans hit 100% while gaming but the temperatures of both the processor and GPU are about 50-54°C.

That's about -10% in performance. Is it really worth that drop?

For the performance I need at the moment it is. I don't think I have a choice either with the radiator I have available. Before undervolting and underclocking the full 400 W was pushing my temperatures quite high because my radiator is just undersized for this level of power. I really do need to add a second which is in progress and requires a case swap to fit it (SFF size build).
 
Right now I am mostly playing Warzone, DOOM, etc and those are more than exceeding the 144 FPS my monitors are capable of at 1440p so the impact of the 10% drop is not being felt at the moment.

But why not undervolt and OC to the best you can and maybe do 300-350W and not lose performance? Find a middle gournd. 

I see what you're saying. I definitely have not tried to optimize the clock speed at all and see if I can push that up at the same voltage and remain stable. For the gaming I am doing however, I really don't need the extra 10% performance I have lost. And as I mentioned I don't think I can support probably over 250 W from the 3080Ti until I add a second radiator to the system.

chrismmcnamara

KingEngineRevUp

chrismmcnamara

KingEngineRevUp

chrismmcnamara

KingEngineRevUp

chrismmcnamara
Pretty digressive on mine but I am stable at 787 mV at 1750 MHz. Was coming from a 1080 GTX too though so even at that level of underclock/undervolt the performance is pretty insane to me still. Power consumption while gaming varies between 150 at 250 W which is a huge improvement for me. I've also got this in a custom loop with a 10920X with only a single 240mm radiator so the power reduction is important. Fans hit 100% while gaming but the temperatures of both the processor and GPU are about 50-54°C.

That's about -10% in performance. Is it really worth that drop?

For the performance I need at the moment it is. I don't think I have a choice either with the radiator I have available. Before undervolting and underclocking the full 400 W was pushing my temperatures quite high because my radiator is just undersized for this level of power. I really do need to add a second which is in progress and requires a case swap to fit it (SFF size build).

Right now I am mostly playing Warzone, DOOM, etc and those are more than exceeding the 144 FPS my monitors are capable of at 1440p so the impact of the 10% drop is not being felt at the moment.

But why not undervolt and OC to the best you can and maybe do 300-350W and not lose performance? Find a middle gournd. 

I see what you're saying. I definitely have not tried to optimize the clock speed at all and see if I can push that up at the same voltage and remain stable. For the gaming I am doing however, I really don't need the extra 10% performance I have lost. And as I mentioned I don't think I can support probably over 250 W from the 3080Ti until I add a second radiator to the system.

But is heat really that bad of an issue? Running a warmer GPU and CPU isn't going to kill it contrary to belief. AMD CPU are okay at 85C+ for example.

Temperature wise for the components themselves, no. I'd be willing to let those go a bit higher in temperature even now so that the system would be a little quieter at load. But The issue is fluid temperature. Pump cannot allow the fluid temperature to be more than 60°C. Right now my max fluid temperature is about 44°C under load. I do have some head room here but I am doubtful that under the full load of the 3080Ti that it would not exceed that, and I usually like to carry at least 5°C of margin for being on the safe side. I could definitely do some more tuning and see how far I could really take it but I just don't see the need at the performance I am getting now. Plus the problem will solve itself when I throw another radiator in when I get a new case.
 
Honestly, I am just impressed with the performance of this card despite me reducing its performance at this level. I was worried when I began underclocking that I would not be able to keep everything maxed but almost all the titles I am currently playing, its able to get at least 144 FPS, usually much higher than that if I don't enable frame limiting. It's pretty impressive so I am really looking forward to putting it through its paces with 2042 coming out soon.

snoopy3525

03whitegsr

KingEngineRevUp
But why not undervolt and OC to the best you can and maybe do 300-350W and not lose performance? Find a middle gournd. 

I'm fairly new to PC gaming, I apologize if this is some well known "don't do"...

But if you use a frame limiter setting in game or through the control panel to match your monitor refresh rate, on games that can run below 100% GPU load, the GPU will run at a lower clock speed automatically. This pushes the operating point to a lower voltage level, and there by reduces the power anyway?

Seems like the ideal way here would be to create a FV table that is at the stability limit at each voltage level? This way you are always at the max frequency for a given power level and the game need will then drive the power draw.

I've been trying to do this but have found two things that make this difficult.

1. Well known, but a FV setting that's stable in benchmarks isn't usually stable in games. This appears to be because even if you have a set frequency at a voltage, in game, it seems to go 1-2 bins higher. For example, I can run 825mV@1890MHz in benchmarks. This curve though will end up spiking to 1920MHz in game though and crashing. If set it two bins lower, 825mV@1860MHz though, in game it will spike to 1890MHz but won't crash. The benchmark point seems to be good, but you have to tweak it to deal with it running higher then it should in game.

2. Afterburner fights you on FV curve settings. Your can set your points you know are stable and then smoothly transition the points in between, but when you hit apply, the resulting curve is something very different. Usually moving your set points several bins up or down, depending on seemingly nothing.

I fine tuned my 1080 Ti using Afterburner curves for undervolting.  Not sure if the 3000 series is much different, but the curve moving and changing is also related to temperature.  Your best bet is to set your curve and hit apply at the higher temperature, not at idle.  At lower temp you will see the core speed raise a little, but once it hits higher temps, the core clock will drop to what you set it to.  Not sure what the temperature increments are, maybe 10 degrees?

Attached Image(s)