Running a GPU with 60-70% powerlimit until new PSU arrives

[Gamer Guy]

My current system is 12400F + RTX 3050. RRX 3050 is sold and I am upgrading to Inno3D RTX 3070Ti 3X OC model because I got a good deal. The catch here is my current PSU is Corsair CX550M 2021 (Black label). It cannot handle the RTX 3070Ti and I will buy a 650/750W PSU but it will take 3-4 days. I have the 3070Ti now and planning to use the GPU at 60-70% power limit or even less because I am not going to game on the card till my new PSU arrives (adjusted in MSI afterburner and profile set to start on startup). I need to complete some documentation work in the upcoming 4 days which cannot be avoided. Is this fine way to run?

 

PSUs yet to be decided between (tight in budget and I need to order it urgent so can't wait for saving more):

 

(all prices converted to USD)

CX650F RGB - $72

XPG Core Reactor 650 - $79

XPG Core Reactor 750 - $88

CX750F RGB - $90

 

 

Please pick one for me!

[jaslion]

Get a msi mpg, corsair rmx, enermax revolution d.f. or fractal ion. They have 650w units in your price bracket and are higher quality too

[191x7]

No need to underpower the GPU, the CX550M can handle it, especially since your other parts aren't power hoggs.

[adm0n]

You could also look into undervolting your card a bit. You might end up with a bit more performance or at least way better efficiency. But your current PSU should be fine. In the long run, I'd still upgrade it though.

[Mark Kaine]

26 minutes ago, adm0n said:

You could also look into undervolting your card a bit. You might end up with a bit more performance or at least way better efficiency. But your current PSU should be fine. In the long run, I'd still upgrade it though.

I think undervolting is possible but i wouldn't recommend it in this case because,  i undervolted my card heavily,  -100mv, runs 20c or so cooler... still reaches its power limit with ease. setting a power limit is much simpler and more reliable. 

These cards are designed to clock as high and use as much power as possible,  which is why its really difficult to lower power consumption by undervolting... you'd kinda need to lower frequencies *a lot* at which point i wonder why even use the card, its much simpler to set a power limit and the card will still clock "normally".

 

43 minutes ago, Gamer Guy said:

all prices converted to USD)

CX650F RGB - $72

XPG Core Reactor 650 - $79

XPG Core Reactor 750 - $88

CX750F RGB - $90

see if you can find a rmx 650... should be around 90, maybe a bit more, but the ones you listed aren't ideal to deal with transient power spikes. 

 

ps: as others said, maybe your current psu can handle it, yes (im not so sure) but getting a new psu of the same class basically makes no sense, if you do it should be an actual upgrade,  hence something like a rmx (which has zero issues with transient power spikes)

[adm0n]

5 minutes ago, Mark Kaine said:

I think undervolting is possible but i wouldn't recommend it in this case because,  i undervolted my card heavily,  -100mv, runs 20c or so cooler... still reaches its power limit with ease. setting a power limit is much simpler and more reliable. 

Something can not be right, with what you are saying. If your card runs 20 Degrees Celsius cooler, it has to use less power, since the power it uses and it's temperature are directly linked. Lowering the voltage usually just makes it more efficient in the sense that an increase of power will also return more processing performance.

 

Setting a power limit is of course simpler, which is why I only added it as a possibility. If you want to undervolt with the intent of lowering power consumption, you can most certainly succeed, but you will need to invest some time into it. So I see this as a good learning opportunity. Of course OP has to decide for themselves, if that is something they are after.

[Mark Kaine]

9 minutes ago, adm0n said:

Something can not be right

tested it a million times... you're right its difficult to understand,  but i think a big part of it is how the cooling actually works, ie "heat build-up", maybe my fan curves play a bigger role than i think, but im not sure about that.

the facts are: it uses less voltage,  it uses the same watts (270) , it has higher frequencies (2010 instead of ~1950) temps are overall *much* lower. 

And its true i have tested pretty much everything *except* with the stock fan curve, because hell if i do, its so awful lol.

 

ps: the thing is: if i run with my fancurve but without undervolt the card still gets really hot (thats the thing, how? why? i dont really care, i tested it a million times as said) doesn't mean it wouldn't run even hotter with stock fan curve (i bet it would)

[PDifolco]

9 minutes ago, adm0n said:

Something can not be right, with what you are saying. If your card runs 20 Degrees Celsius cooler, it has to use less power, since the power it uses and it's temperature are directly linked. Lowering the voltage usually just makes it more efficient in the sense that an increase of power will also return more processing performance.

 

Setting a power limit is of course simpler, which is why I only added it as a possibility. If you want to undervolt with the intent of lowering power consumption, you can most certainly succeed, but you will need to invest some time into it. So I see this as a good learning opportunity. Of course OP has to decide for themselves, if that is something they are after.

From experience undervolting don't reduce much power consumption if at all (on CPU and GPU), while it has noticeable effects on temps

Maybe because temps that are monitored are max temps on a part of the chip, not average all over it, then they DON'T translate to power usage...

 

In the OP case I suppose his TXM 550W PSU can hold the 3070Ti , at worse PSU will shutdown, but not a big risk anyway

Lowering powerlimit to 80% should imo make it safe

 

[Mark Kaine]

This is definitely a case of "the math says this, so it must be right"

 

Even though the calculation misses crucial parts, such as how heat build up works and therefore is *wrong* : D

 

6 minutes ago, PDifolco said:

From experience undervolting don't reduce much power consumption if at all (on CPU and GPU)

yeap... my gpu actually uses more when undervolted, 273 (my record!) vs ~269 at stock (yeah, anecdotal,  but its definitely not using less, at max at least)

my old r5 3600 was even "worse" like over 100w with negative voltage offset,  without that it would pretty much stick to 65w... ¯\_(ツ)_/¯ 

[--SID--]

Nothing wrong with a XPG Core Reactor. Quality is on par with the Corsair RMx, no issues with transient spikes and €88 for the 750w is a steal. That's the one I should get.

[Mark Kaine]

5 minutes ago, --SID-- said:

Nothing wrong with a XPG Core Reactor. Quality is on par with the Corsair RMx, no issues with transient spikes and €88 for the 750w is a steal. That's the one I should get.

ah good to know then...

[adm0n]

57 minutes ago, Mark Kaine said:

tested it a million times... you're right its difficult to understand,  but i think a big part of it is how the cooling actually works, ie "heat build-up", maybe my fan curves play a bigger role than i think, but im not sure about that.

the facts are: it uses less voltage,  it uses the same watts (270) , it has higher frequencies (2010 instead of ~1950) temps are overall *much* lower. 

And its true i have tested pretty much everything *except* with the stock fan curve, because hell if i do, its so awful lol.

 

ps: the thing is: if i run with my fancurve but without undervolt the card still gets really hot (thats the thing, how? why? i dont really care, i tested it a million times as said) doesn't mean it wouldn't run even hotter with stock fan curve (i bet it would)

I can't really tell you what is going on, the theory of @PDifolcocould be true. By undervolting you reduce the temperature in specific hot spots of the card, but the average energy the card uses stays the same. It could also be, that the power readings don't get adjusted properly when applying the undervolt.

But I guess the link between the power consumption and temperature readings of a GPU doesn't have to be as straight forward as I assumed.

 

You are however running your card harder than before, so you are getting that increased efficiency, but you are using it to squeeze more than stock power out of your card. Undervolting saves energy, if you aren't doing that. Here is an article from Igors lab -> https://www.igorslab.de/en/3070-ti-undervolting-can-we-also-deny-the-most-unpopular-of-all-tis-binge-drinking/5/ (it's machine translated to english, but you can see, that they save around 50W at the same frequency, just by lowering the voltage).

 

Also here is a research paper about finding the efficiency sweet spot for GPUs used for matrix multiplications (and errors this might bring):

https://dl.acm.org/doi/pdf/10.1145/3330345.3330373

 

 

54 minutes ago, PDifolco said:

From experience undervolting don't reduce much power consumption if at all (on CPU and GPU), while it has noticeable effects on temps

Maybe because temps that are monitored are max temps on a part of the chip, not average all over it, then they DON'T translate to power usage...

Fair point. If you compare the power consumption while leaving the frequency alone, you will generally see a decline. So fixing the frequency at 1000Mhz, running it at 1000mV will draw more power than running it at 800mV.

[Mark Kaine]

3 minutes ago, adm0n said:

Fair point. If you compare the power consumption while leaving the frequency alone, you will generally see a decline. So fixing the frequency at 1000Mhz, running it at 1000mV will draw more power than running it at 800mV.

yeap,  but you specifically have to do that... mines clocking higher (which from my pov is the point, besides temps)

and another factor is also nvidia boost... i don't have evidence for this, but it seems pretty obvious the card actually *wants* to use more power whenever possible,  the only thing holding it back is the vbios power limit... so that's another reason why it reaches the more or less artificial power "limit" with ease (mind you i really need to push it, typically my gpu is at ~50% usage and 120w or so...)

[adm0n]

Just now, Mark Kaine said:

yeap,  but you specifically have to do that... mines clocking higher (which from my pov is the point, besides temps)

and another factor is also nvidia boost... i don't have evidence for this, but it seems pretty obvious the card actually *wants* to use more power whenever possible,  the only thing holding it back is the vbios power limit... so that's another reason why it reaches the more or less artificial power "limit" with ease (mind you i really need to push it, typically my gpu is at ~50% usage and 120w or so...)

I've only really undervolted by using the curve tool in MSI Afterburner. And it more or less does exactly what I'm setting it to, within one step in either direction of accuracy. I don't really know how the core voltage offset works, but if you are using that, maybe the cards normal boosting behavior will still run into whatever target it has set. And I think normally that is Temperature->Power.

[mariushm]

2 hours ago, Gamer Guy said:

My current system is 12400F + RTX 3050. RRX 3050 is sold and I am upgrading to Inno3D RTX 3070Ti 3X OC model because I got a good deal. The catch here is my current PSU is Corsair CX550M 2021 (Black label). It cannot handle the RTX 3070Ti and I will buy a 650/750W PSU but it will take 3-4 days. 

 

PSUs yet to be decided between (tight in budget and I need to order it urgent so can't wait for saving more):

 

(all prices converted to USD)

CX650F RGB - $72

XPG Core Reactor 650 - $79

XPG Core Reactor 750 - $88

CX750F RGB - $90

 

Please pick one for me!

The 3070ti can peak up to 300 watts from 12v at 100% load.

The 12400F peaks at around 120 watts from 12v at 100% load

The rest in your computer consumes maybe 20-30 watts from 12v (fans, hard drive motors, 12v RGB leds) 

 

So your power consumption on 12v will be at most around 450 watts. 

 

The CX550M is rated for  45.8A on 12v, that's 550w ... so you shouldn't need to lower the gpu power budget, the power supply should handle it. 

 

Go with the CX750F if you have to. 

[Mark Kaine]

11 minutes ago, mariushm said:

The 3070ti can peak up to 300 watts from 12v at 100% load.

The 12400F peaks at around 120 watts from 12v at 100% load

The rest in your computer consumes maybe 20-30 watts from 12v (fans, hard drive motors, 12v RGB leds) 

 

So your power consumption on 12v will be at most around 450 watts. 

my 3070 + 3600 used about 500w , according to what my rmi tells me and which seems to be realistic too (mind you i had to run superposition + p95 at the same time to get there ) but playing games like Tomb Raider wasnt much lower... 450-470w maybe... 

 

i agree their psu might work but its cutting it hella close for sure... better get that xpg core imo.

[Mark Kaine]

1 hour ago, adm0n said:

I don't really know how the core voltage offset works, but if you are using that, maybe the cards normal boosting behavior will still run into whatever target it has set. And I think normally that is Temperature->Power.

whats funny with my undervolt is the only limit i ever get is "idle" (according to GPUZ)

 

and nah, i used the curve... 0.930v @ 2010mhz straight line...

 

the thing is afterburner doesn't always adhere to this, sometimes it'll go to 0.981v and / or 2040mhz... so again it's pushing for more power / high frequencies (probably due to low temps)

 

but 2040mhz is max, everything above crashes (it never goes over with my undervolt though) 

So afterburner does a good job overall,  still it would be nice if it adhered to my settings exactly (if i set it to 0.981v it still overwrites that and gives it even more voltage sometimes,  even though the card really just needs 0.930v)

[adm0n]

14 minutes ago, Mark Kaine said:

whats funny with my undervolt is the only limit i ever get is "idle" (according to GPUZ)

 

and nah, i used the curve... 0.930v @ 2010mhz straight line...

 

the thing is afterburner doesn't always adhere to this, sometimes it'll go to 0.981v and / or 2040mhz... so again it's pushing for more power / high frequencies (probably due to low temps)

 

but 2040mhz is max, everything above crashes (it never goes over with my undervolt though) 

So afterburner does a good job overall,  still it would be nice if it adhered to my settings exactly (if i set it to 0.981v it still overwrites that and gives it even more voltage sometimes,  even though the card really just needs 0.930v)

The difference between 0.981v and 0.930v is quite big. If I remember correctly, afterburner can only suggest settings to Nvidia's driver and they are affected by the current temperature of the card? So applying the curves at different temperatures will yield slightly different results. But I doubt that is the reason.

 

I actually think, that this is an artifact of undervolting your card. You are always running into the limit set by the curve, which should be a voltage limit? But it isn't reported as one. So other than the power limit, you can't really trigger any other limits anymore. Which is kinda a good thing, because it means, that your card isn't throttling itself.

[Gamer Guy]

5 hours ago, 191x7 said:

No need to underpower the GPU, the CX550M can handle it, especially since your other parts aren't power hoggs.

What, the 3070Ti + 12400F setup in outervision calculator shows more than 550W. How CX550 can withstand?
Intel Core i5-12400F NVIDIA GeForce RTX 3070 Ti - PSU Calculator - Build hmSCpr (outervision.com)

Edited December 7, 2022 by Gamer Guy
included outer vision calculator link

[Gamer Guy]

4 hours ago, Mark Kaine said:

but the ones you listed aren't ideal to deal with transient power spikes.

XPG core reactor are also Tier A units. RMx series is $24 more here for same wattage.

[Gamer Guy]

4 hours ago, --SID-- said:

Nothing wrong with a XPG Core Reactor. Quality is on par with the Corsair RMx, no issues with transient spikes and €88 for the 750w is a steal. That's the one I should get.

Thanks, I saw comparisons and XPG is on par..