Hey Kevin! I read one of your replies explaining power = voltage x current or something. I did try my hand at undervolting my 3090 a bit, but it may not seem like such a good idea anymore with the increased amperage. I've read numerous posts about how undervolting extends the life of your GPU. Is this not true because of the increased amps?

glocked89
Hey Kevin! I read one of your replies explaining power = voltage x current or something. I did try my hand at undervolting my 3090 a bit, but it may not seem like such a good idea anymore with the increased amperage. I've read numerous posts about how undervolting extends the life of your GPU. Is this not true because of the increased amps?

It actually doesn't extend the life of the GPU.
 
Running it at stock is what it was designed to do and they will last longer than MOST people would keep or use them anyway.
 

It's pretty simple, Kevin. Boost sucks. The end. 
 
But seriously, I'll give you an example. When I play Baldur's Gate 3, my clocks start out (at all stock settings) at around 1950mhz. After about 2 minutes of load, I am at around 1800-1850mhz just letting the card do it's thing with the boost algorithm. It will remain at these speeds for the rest of the game. After testing my card as any true overclocker would, I figured out that I can run 1950mhz stable at 0.862v. This is significantly lower voltage than the boost algorithm uses to run 1950mhz.. you can't just "reduce the powerlimit" to get those settings because it won't work. So I undervolted my card to run 1950mhz at 0.862v, which looks like this:
 

 
What happens when I run Baldur's Gate 3 using these settings? My clocks start at 1950mhz and never leave. The boost algorithm never feels the need to reduce the clocks because it is no longer hitting the power limit or concerned with thermals. The result is increased performance in every metric. 
 
The same thing applies when you are "maxing out" your clocks for benchmarks and such. Sure, maybe you can reach 2200mhz by moving the sliders all the way up... but what good is it if the boost algorithm reduces everything 30 seconds into the benchmark? You'd get better performance undervolting and stabilizing the clocks at say... 2150mhz instead. That is why people undervolt.

Where exactly are these fictional people who are "undervolting for the sake of undervolting"? They do it for the reasons and benefits outlined in my post, not just for giggles.

Celeras
It's pretty simple, Kevin. Boost sucks. The end. 
 
But seriously, I'll give you an example. When I play Baldur's Gate 3, my clocks start out (at all stock settings) at around 1950mhz. After about 2 minutes of load, I am at around 1800-1850mhz just letting the card do it's thing with the boost algorithm. It will remain at these speeds for the rest of the game. After testing my card as any true overclocker would, I figured out that I can run 1950mhz stable at 0.862v. This is significantly lower voltage than the boost algorithm uses to run 1950mhz.. you can't just "reduce the powerlimit" to get those settings because it won't work. So I undervolted my card to run 1950mhz at 0.862v, which looks like this:
 

 
What happens when I run Baldur's Gate 3 using these settings? My clocks start at 1950mhz and never leave. The boost algorithm never feels the need to reduce the clocks because it is no longer hitting the power limit or concerned with thermals. The result is increased performance in every metric. 
 
The same thing applies when you are "maxing out" your clocks for benchmarks and such. Sure, maybe you can reach 2200mhz by moving the sliders all the way up... but what good is it if the boost algorithm reduces everything 30 seconds into the benchmark? You'd get better performance undervolting and stabilizing the clocks at say... 2150mhz instead. That is why people undervolt.

I had a curve like that too as well, but the minor issue I had was that the curve kept resetting to something similar and not exact after a reboot. Afterburner starts with windows and I have the little "Windows" box checked as well. I have to manually click Profile 1 each time.

Celeras
I don't want to hijack his thread with your tech support but are you sure? The curve looks a little bit different when you click the "Profile 1" and when you click "Apply".

Hijack away, any and all undervolting / voltage locking, tips / insights are welcome.

glocked89

Celeras
I don't want to hijack his thread with your tech support but are you sure? The curve looks a little bit different when you click the "Profile 1" and when you click "Apply".

Oh I just wanted to know if your curve stayed the same after a reboot. Mine will change slightly and I have to manually apply it each time.

Yeah it stays the same, I'm just saying that it LOOKS different when you press "Profile 1" compared to when you press "Apply"... so maybe you thought it was changing when it really wasn't. 

Celeras
It's pretty simple, Kevin. Boost sucks. The end. 
 
But seriously, I'll give you an example. When I play Baldur's Gate 3, my clocks start out (at all stock settings) at around 1950mhz. After about 2 minutes of load, I am at around 1800-1850mhz just letting the card do it's thing with the boost algorithm. It will remain at these speeds for the rest of the game. After testing my card as any true overclocker would, I figured out that I can run 1950mhz stable at 0.862v. This is significantly lower voltage than the boost algorithm uses to run 1950mhz.. you can't just "reduce the powerlimit" to get those settings because it won't work. So I undervolted my card to run 1950mhz at 0.862v, which looks like this:
 

 
What happens when I run Baldur's Gate 3 using these settings? My clocks start at 1950mhz and never leave. The boost algorithm never feels the need to reduce the clocks because it is no longer hitting the power limit or concerned with thermals. The result is increased performance in every metric. 
 
The same thing applies when you are "maxing out" your clocks for benchmarks and such. Sure, maybe you can reach 2200mhz by moving the sliders all the way up... but what good is it if the boost algorithm reduces everything 30 seconds into the benchmark? You'd get better performance undervolting and stabilizing the clocks at say... 2150mhz instead. That is why people undervolt.

I went ahead and tested undervolt using a somewhat heavy load (for Heaven) static scene in Unigine Heaven on my 2080 Ti Hybrid.
 
- Please disregard my statements regarding current and voltage vs. power, they are too much of an oversimplification for how a GPU loads up and have little practical bearing.  With 1875mhz @ 0.875v I'd get 220w draw, with 1875 @ 1.050v I'd get about 300w. The power scales somewhat in accordance with the ohms law P = V^2 / R, or the square of the percent increase in voltage, with a static scene as a gross oversimplification the gpu can be taken as a resistive load, thus with a higher voltage across it has to draw (much) more power and current.
 
- I can apply my normal gaming overclock of +120 and restrict power limit to 221w (average GPU-Z), card will sit right around 0.887 to 0.906v and 1890 to 1920 mhz.  Going back to locked voltage, I can dial in .875v at 1905mhz at 221w average.  Both very stable at 45-46C and FPS around 205 fps or so. 
 
This is far from an exhaustive test, but I'm seeing no performance or thermal difference between capping power with the power limit slider vs. locking, with a static scene. 
 
Now when you let Heaven run, that's a different story.  When power limited, the card will clock up and volt up with a light load, with the 221w limit I went as high as 2100mhz a few times.  When locked at the comparable 1905mhz at 0.875v, the card could draw up to about 230w, but averaged 210w since power draw was dropping under light load, instead of the clocks spiking and it continuing to ride the power limit.
 
So as Celeras put it, "Boost Sucks" if you're going to let your FPS free run and have no need for those extra frames and would rather have the GPU power down vs. boosting up against the power limit.  Maybe you'll get a couple degrees difference vs. just using an equivalent power limit, but it's really noise level stuff IMHO.  Maybe there's games with insane swings in load where this would make sense and you would see a bigger difference.  Maybe others where it prevents downclocking due to poor cooling, but using the power limit more aggressively would also help with thermal downclocking in a limited cooling scenario.  But it's applicable when locking voltage anywhere, you can do it at 1.050v or whatever and see the same behavior vs. letting the card do it's thing with a similar conventional setting.
 
If you use a low lag vsync like I do (4k60) and are always at refresh rate, it's not going to make a difference since the gpu will power down with light loads anyway.  The only way I can sustain 2,100mhz is if I lock in at 1.093v, otherwise it's not stable.
 

how can you do it in x1 precision?
i cant get it to do that without dragging every single point?!

jankerson
It actually doesn't extend the life of the GPU.
 
Running it at stock is what it was designed to do and they will last longer than MOST people would keep or use them anyway.

Undervolting actually does extend the life of a GPU. Anything to reduce heat does extend the life of a GPU.
 
Especially as features get smaller, diffusion in silicon gets to be a larger factor in longevity. Diffusion is also exponentially related to temperature.
 
The question is when the degradation becomes enough that chips fail. This was discussed by some with the debut of the 2000 series chips but didn’t seem to turn out to be an issue. The 3000 series have smaller features and appear to be being run even hotter than the 2000 series. Smaller features means diffusion will have a greater effect. Higher temperatures means more diffusion.
 
The jury is still out on high temps with these new chips. It’s going to take time and people running these cards hot to ultimately know. Personally, I’m not worried about it but I’ll also be undervolting and went for the FTW3 to get the extra cooling. Depending on temperatures, I may even water cool.
 
And it isn’t just the GPU and diffusion. Temperature cycles are hard on components in general. Thermal expansion and contraction flexes all sorts of things that can crack, go intermittent, or fail completely. The higher the temperatures in those cycles, the larger the expansion and contractions. It’s part of the issue between the poly and ceramic caps under the GPUs in the 30 series. The ceramic caps are better able to deal with the high frequency noise but are more likely to crack and fail due to thermal stresses.
 
High temperatures are hard on electronics. Repeated high temperature cycling is even worse. I agree people *should* be fine but we won’t know for sure until this generation is all out of service.

notarjy
Those who are using afterburner to undervolt, are you not running PX1 at all or are you able to use PX1 for fan profiles while using afterburner to undervolt?

I just use afterburner. It has a fan curve editor in the settings. I start it at 40% until the temp reaches 35% and then it goes in a straight line upwards until I get to 100% fan speed at 80c. I don't get above 65c really though in normal use.

kevinc313

macktkau

notarjy
Those who are using afterburner to undervolt, are you not running PX1 at all or are you able to use PX1 for fan profiles while using afterburner to undervolt?

I just use afterburner. It has a fan curve editor in the settings. I start it at 40% until the temp reaches 35% and then it goes in a straight line upwards until I get to 100% fan speed at 80c. I don't get above 65c really though in normal use.

IIRC you can't properly control FTW3 fans in Afterburner, but the XC3 should be fine.

I've heard that, but I found in my testing that when using the MSI Afterburner fan control settings:
 
The GPU fans never drop below 50% of the fan speed (as shown in the afterburner monitor). This is equal to about 30% in the afterburner settings fan curve area. It is 900 RPM.
 
30% in the afterburner fan settings area is equal to 50% in the monitoring, and 900 RPM.
40% in the afterburner fan settings area is equal to 58% in the monitoring, and 1200 RPM.
50% in the afterburner fan settings area is equal to 65% in the monitoring, and 1500 RPM.
60% in the afterburner fan settings area is equal to 72% in the monitoring, and 1850 RPM.
70% in the afterburner fan settings area is equal to 79% in the monitoring, and 2100 RPM.
80% in the afterburner fan settings area is equal to 86% in the monitoring, and 2400 RPM.
I assume 90% and 100% are 2700 RPM and 3000 RPM but I didn't test it.
 
I have the settings at 40% fan speed until the temps reach above 40c, at then it goes up until it reaches 80% fan speed at 70c (this is the practical limit for my fan speed because my card hasn't gone over 65c since undervolting).
 
So at the very least two of the fans are working in afterburner, although the settings & report monitoring look to have difference definitions of percentages, and there's possibly a minimum fan speed set still.

Afterburner, although I checked with Open Hardware Monitor and it seems to match up, but that only reports a single fan RPM & percentage
 
I don't know about the 3rd fan at all. For all I know, it might not be going at all.

macktkau

kevinc313

macktkau

notarjy
Those who are using afterburner to undervolt, are you not running PX1 at all or are you able to use PX1 for fan profiles while using afterburner to undervolt?

I just use afterburner. It has a fan curve editor in the settings. I start it at 40% until the temp reaches 35% and then it goes in a straight line upwards until I get to 100% fan speed at 80c. I don't get above 65c really though in normal use.

IIRC you can't properly control FTW3 fans in Afterburner, but the XC3 should be fine.

I've heard that, but I found in my testing that when using the MSI Afterburner fan control settings:
 
The GPU fans never drop below 50% of the fan speed (as shown in the afterburner monitor). This is equal to about 30% in the afterburner settings fan curve area. It is 900 RPM.
 
30% in the afterburner fan settings area is equal to 50% in the monitoring, and 900 RPM.
40% in the afterburner fan settings area is equal to 58% in the monitoring, and 1200 RPM.
50% in the afterburner fan settings area is equal to 65% in the monitoring, and 1500 RPM.
60% in the afterburner fan settings area is equal to 72% in the monitoring, and 1850 RPM.
70% in the afterburner fan settings area is equal to 79% in the monitoring, and 2100 RPM.
80% in the afterburner fan settings area is equal to 86% in the monitoring, and 2400 RPM.
I assume 90% and 100% are 2700 RPM and 3000 RPM but I didn't test it.
 
I have the settings at 40% fan speed until the temps reach above 40c, at then it goes up until it reaches 80% fan speed at 70c (this is the practical limit for my fan speed because my card hasn't gone over 65c since undervolting).
 
So at the very least two of the fans are working in afterburner, although the settings & report monitoring look to have difference definitions of percentages, and there's possibly a minimum fan speed set still.

Whenever you have a chance can you ss your fan curve?

Thanks brother