Thought undervolting was supposed to lower temps on a rtx 3070

[Reaper]

I've always tried to do huge overclocks, but the returns have been getting lower with each generation of cards so I started to try and undervolt to lower my temps, but the lower I drop the voltage (and it's still stable in benchmarks) the higher my temps go. How is that possible?

[Reaper]

Reaper původně napsal:

Went and did some before and after benchmarks with pictures.

Stock results from Time Spy benchmark:
https://gyazo.com/50d8a02a55ba3f6950361ad3eca443d4

900mv:
https://gyazo.com/883d8f69d5bfe414b867702c3fd2a783

875mv, also lowered the clock speed from stock 1965MHz to 1950MHz:
https://gyazo.com/d8dadd1bf25637f95b253d953f521a37

I can try to go lower, but more than one guy from that reddit thread said he had a 3070 that went from 70c stock voltage to maxing 65c at 900mv. I get 70c at stock voltage so based on other people's results, I should theoretically get reduction at least somewhat, not go up.

Do you have a priority switch like what I do. If so put your temp limit to 80 C and put your priority towards the temp limit. and see if that changes anything. What you could try is putting it to .850/.875 mV @ 1600 MHz. If this doesn't work then you might need to put new thermal past on the gpu.

https://ibb.co/M8Pfhwn

[Komarimaru]

Undervolting does lower temperatures, but also keep in mind it depends on many other factors. Generally locking FPS can lower temps, for example I lock my FPS to 120 FPS in my games with my GPU, keeping temps around 50c.

My card is undervolted to 0.875 with a max stable boost of 2025Mhz due to also having an overclock, though I am running a 3080TI FTW3, so it's just a beast of a card anyway.

Undervolting, aka setting a custom curve, is simply trial and error. Each card will be different.

Way I did mine, is stress the GPU with Furmark and other benchmarks, and started from the lowest boost and worked my way up the voltages to the highest.

For example, my curve let's me get to 1400 at 0.700mv. Not all cards will do this.

My max speed at 0.825 is 1995Mhz, max stable speed at 0.850 is 2010Mhz, etc etc.

Nvidia changes clock in 15Mhz increments most times for 30-40 series, believe 20 series as well.

So, it's down to trial and error. A lot of tweaking, and due to how nvida boost scaling works, you want the card to cool down fully before changing settings each time, since it boosts/click based on a temperature scale.

[Jamebonds1]

Komarimaru původně napsal:

Undervolting does lower temperatures, but also keep in mind it depends on many other factors. Generally locking FPS can lower temps, for example I lock my FPS to 120 FPS in my games with my GPU, keeping temps around 50c.

My card is undervolted to 0.875 with a max stable boost of 2025Mhz due to also having an overclock, though I am running a 3080TI FTW3, so it's just a beast of a card anyway.

Undervolting, aka setting a custom curve, is simply trial and error. Each card will be different.

Way I did mine, is stress the GPU with Furmark and other benchmarks, and started from the lowest boost and worked my way up the voltages to the highest.

For example, my curve let's me get to 1400 at 0.700mv. Not all cards will do this.

My max speed at 0.825 is 1995Mhz, max stable speed at 0.850 is 2010Mhz, etc etc.

Nvidia changes clock in 15Mhz increments most times for 30-40 series, believe 20 series as well.

So, it's down to trial and error. A lot of tweaking, and due to how nvida boost scaling works, you want the card to cool down fully before changing settings each time, since it boosts/click based on a temperature scale.

0.875 v are still within the normal operating voltage, so it is not undervolt. Undervolting could cause computer crash and stressed components by high current to keep up with lower voltage.

[Reaper]

Reaper původně napsal:

Reaper původně napsal:

Went and did some before and after benchmarks with pictures.

Stock results from Time Spy benchmark:
https://gyazo.com/50d8a02a55ba3f6950361ad3eca443d4

900mv:
https://gyazo.com/883d8f69d5bfe414b867702c3fd2a783

875mv, also lowered the clock speed from stock 1965MHz to 1950MHz:
https://gyazo.com/d8dadd1bf25637f95b253d953f521a37

I can try to go lower, but more than one guy from that reddit thread said he had a 3070 that went from 70c stock voltage to maxing 65c at 900mv. I get 70c at stock voltage so based on other people's results, I should theoretically get reduction at least somewhat, not go up.

Do you have a priority switch like what I do. If so put your temp limit to 80 C and put your priority towards the temp limit. and see if that changes anything. What you could try is putting it to .850/.875 mV @ 1600 MHz. If this doesn't work then you might need to put new thermal past on the gpu.

https://ibb.co/M8Pfhwn

Already have that set to prioritize temps and it's maxed at 90c so it shouldn't even come into play. However, I don't know why HWMonitor is saying I'm hitting a temp limit when I'm nowhere near 90c. Starting to wonder if Afterburner isn't playing nice with my EVGA card. Maybe when I eventually upgrade in a few years, I'll go with an MSI card

[Reaper]

Reaper původně napsal:

Reaper původně napsal:

Do you have a priority switch like what I do. If so put your temp limit to 80 C and put your priority towards the temp limit. and see if that changes anything. What you could try is putting it to .850/.875 mV @ 1600 MHz. If this doesn't work then you might need to put new thermal past on the gpu.

https://ibb.co/M8Pfhwn

Already have that set to prioritize temps and it's maxed at 90c so it shouldn't even come into play. However, I don't know why HWMonitor is saying I'm hitting a temp limit when I'm nowhere near 90c. Starting to wonder if Afterburner isn't playing nice with my EVGA card. Maybe when I eventually upgrade in a few years, I'll go with an MSI card

What version update of MSI afterburner do you have the current version is v4.6.5

[Reaper]

Reaper původně napsal:

Reaper původně napsal:

Already have that set to prioritize temps and it's maxed at 90c so it shouldn't even come into play. However, I don't know why HWMonitor is saying I'm hitting a temp limit when I'm nowhere near 90c. Starting to wonder if Afterburner isn't playing nice with my EVGA card. Maybe when I eventually upgrade in a few years, I'll go with an MSI card

What version update of MSI afterburner do you have the current version is v4.6.5

4.6.4 When I check updates with the settings, it says that's the most current version

[Komarimaru]

Jamebonds1 původně napsal:

Komarimaru původně napsal:

Undervolting does lower temperatures, but also keep in mind it depends on many other factors. Generally locking FPS can lower temps, for example I lock my FPS to 120 FPS in my games with my GPU, keeping temps around 50c.

My card is undervolted to 0.875 with a max stable boost of 2025Mhz due to also having an overclock, though I am running a 3080TI FTW3, so it's just a beast of a card anyway.

Undervolting, aka setting a custom curve, is simply trial and error. Each card will be different.

Way I did mine, is stress the GPU with Furmark and other benchmarks, and started from the lowest boost and worked my way up the voltages to the highest.

For example, my curve let's me get to 1400 at 0.700mv. Not all cards will do this.

My max speed at 0.825 is 1995Mhz, max stable speed at 0.850 is 2010Mhz, etc etc.

Nvidia changes clock in 15Mhz increments most times for 30-40 series, believe 20 series as well.

So, it's down to trial and error. A lot of tweaking, and due to how nvida boost scaling works, you want the card to cool down fully before changing settings each time, since it boosts/click based on a temperature scale.

0.875 v are still within the normal operating voltage, so it is not undervolt. Undervolting could cause computer crash and stressed components by high current to keep up with lower voltage.

Hate to break it to you, but the 3080ti is rated for a Boost of 1665Mhz@1.1 Volts.
https://www.techpowerup.com/gpu-specs/geforce-rtx-3080-ti.c3735

Not 2025 at 0.875V. Far under specification sheets.

Even my model, the EVGA 3080ti FTW3 Ultra, is rated at 1800Mhz@1.1 Volts.
https://www.techpowerup.com/gpu-specs/evga-rtx-3080-ti-ftw3-ultra.b8959

So compared to a FE, the EVGA is 8% faster@stock speeds/voltage. With my undervolt+overclock, it's 20% difference.

And no, it's not common to get those speeds at all at that voltage. Called winning the silicon lottery, since each chip is different.

[Komarimaru]

Reaper původně napsal:

Reaper původně napsal:

Do you have a priority switch like what I do. If so put your temp limit to 80 C and put your priority towards the temp limit. and see if that changes anything. What you could try is putting it to .850/.875 mV @ 1600 MHz. If this doesn't work then you might need to put new thermal past on the gpu.

https://ibb.co/M8Pfhwn

Already have that set to prioritize temps and it's maxed at 90c so it shouldn't even come into play. However, I don't know why HWMonitor is saying I'm hitting a temp limit when I'm nowhere near 90c. Starting to wonder if Afterburner isn't playing nice with my EVGA card. Maybe when I eventually upgrade in a few years, I'll go with an MSI card

Did you make sure you set a custom fan curve, in MSI afterburner?

[Jamebonds1]

Komarimaru původně napsal:

Jamebonds1 původně napsal:

0.875 v are still within the normal operating voltage, so it is not undervolt. Undervolting could cause computer crash and stressed components by high current to keep up with lower voltage.

Hate to break it to you, but the 3080ti is rated for a Boost of 1665Mhz@1.1 Volts.
https://www.techpowerup.com/gpu-specs/geforce-rtx-3080-ti.c3735

Not 2025 at 0.875V. Far under specification sheets.

Even my model, the EVGA 3080ti FTW3 Ultra, is rated at 1800Mhz@1.1 Volts.
https://www.techpowerup.com/gpu-specs/evga-rtx-3080-ti-ftw3-ultra.b8959

So compared to a FE, the EVGA is 8% faster@stock speeds/voltage. With my undervolt+overclock, it's 20% difference.

And no, it's not common to get those speeds at all at that voltage. Called winning the silicon lottery, since each chip is different.

If that is the case. By your logic, the PSU specified 12V and measured 11.70V, then it is considered as an undervolt.

I have read your source and I didn't see any normal voltage range. That said, you're not supposed to read the GPU specifications except one integrated circuit BGA specifications. It will tell specified normal operating voltage range, but I think it is for customers only (the GPU manufacturers).

Also, I do not think MSI afterburner allow you go to below 600 mV or whatever it limited you from lower voltage.

[Komarimaru]

Jamebonds1 původně napsal:

Komarimaru původně napsal:

Hate to break it to you, but the 3080ti is rated for a Boost of 1665Mhz@1.1 Volts.
https://www.techpowerup.com/gpu-specs/geforce-rtx-3080-ti.c3735

Not 2025 at 0.875V. Far under specification sheets.

Even my model, the EVGA 3080ti FTW3 Ultra, is rated at 1800Mhz@1.1 Volts.
https://www.techpowerup.com/gpu-specs/evga-rtx-3080-ti-ftw3-ultra.b8959

So compared to a FE, the EVGA is 8% faster@stock speeds/voltage. With my undervolt+overclock, it's 20% difference.

And no, it's not common to get those speeds at all at that voltage. Called winning the silicon lottery, since each chip is different.

If that is the case. By your logic, the PSU specified 12V and measured 11.70V, then it is considered as an undervolt.

I have read your source and I didn't see any normal voltage range.

Also, I do not think MSI afterburner allow you go to below 600 mV or whatever it limited you from lower voltage.

MSI afterburner let's you go as low as 0.500v. It's designed to support many cards so has many settings that advanced users can take advantage of. e.i. Setting exact voltage to what core frequency you want. You already admitted you can't drop your own card below 0.900v, probably due to again, silicon lottery or old hardware since not all cards can undervolt well due to different technologies.

Also due to custom PCB with EVGA, and better components that can handle more strenuous settings and tweaks, allows such a thing, plus a custom Bios from EVGA.

And if a PSU is at 11.7v, that's a bad PSU since quickly approaching the 5% voltage difference of ATX standard.

Either way, your input is unwarranted. They are asking for advice on undervolting, which you don't understand, since you have your own personal opinion, even with it being wrong, on what undervolting is.

[Jamebonds1]

Komarimaru původně napsal:

Jamebonds1 původně napsal:

If that is the case. By your logic, the PSU specified 12V and measured 11.70V, then it is considered as an undervolt.

I have read your source and I didn't see any normal voltage range.

Also, I do not think MSI afterburner allow you go to below 600 mV or whatever it limited you from lower voltage.

MSI afterburner let's you go as low as 0.500v. It's designed to support many cards so has many settings that advanced users can take advantage of. e.i. Setting exact voltage to what core frequency you want. You already admitted you can't drop your own card below 0.900v, probably due to again, silicon lottery or old hardware since not all cards can undervolt well due to different technologies.

Also due to custom PCB with EVGA, and better components that can handle more strenuous settings and tweaks, allows such a thing, plus a custom Bios from EVGA.

And if a PSU is at 11.7v, that's a bad PSU since quickly approaching the 5% voltage difference of ATX standard.

That is normal for PSU to go from 11.97 to 11.70 when it is loaded At 50% to 75%. Another reason why I disliked cheap PSU tester ~$10.

12V × 5% = 11.4 to 12.6 voltage

Anything below 500 mV will be considered as an undervolt in engineering term words. But they may designed prototype then tested to make sure the GPU is work in undervolt event.

As I said, GPU manufacturers may have specifications manual for the graphic processor integrated circuit BGA with normal operating voltage range, but I doubt they will share that.

PS: Yes I understand what undervolting mean, and it is outside of the normal operating voltage range. And you just said 11.97 V is an undervolt for the PSU. This is not my personal opinion, I don't see any of your GPU's processor's normal operating voltage range on your source. So we don't know any normal operating voltage range, except depending on the software to prevented you from undervolting.

[Komarimaru]

Jamebonds1 původně napsal:

Komarimaru původně napsal:

MSI afterburner let's you go as low as 0.500v. It's designed to support many cards so has many settings that advanced users can take advantage of. e.i. Setting exact voltage to what core frequency you want. You already admitted you can't drop your own card below 0.900v, probably due to again, silicon lottery or old hardware since not all cards can undervolt well due to different technologies.

Also due to custom PCB with EVGA, and better components that can handle more strenuous settings and tweaks, allows such a thing, plus a custom Bios from EVGA.

And if a PSU is at 11.7v, that's a bad PSU since quickly approaching the 5% voltage difference of ATX standard.

That is normal for PSU to go from 11.97 to 11.70 when it is loaded At 50% to 75%. Another reason why I disliked cheap PSU tester ~$10.

12V × 5% = 11.4 to 12.6 voltage

Anything below 500 mV will be considered as an undervolt in engineering term words. But they may designed prototype then tested to make sure the GPU is work in undervolt event.

As I said, GPU manufacturers may have specifications manual for the graphic processor integrated circuit BGA with normal operating voltage range, but I doubt they will share that.

PS: Yes I understand what undervolting mean, and it is outside of the normal operating voltage range. And you just said 11.97 V is an undervolt for the PSU. This is not my personal opinion, I don't see any of your GPU's processor's normal operating voltage range on your source. So we don't know any normal operating voltage range, except depending on the software to prevented you from undervolting.

1. It's not normal for a PSU to do that unless it's cheap. Mine goes up in voltage under load, not drop. Use a meter to test, not software.

2. You're off topic and don't understand said topic so why are you even here?

[Jamebonds1]

Komarimaru původně napsal:

Jamebonds1 původně napsal:

That is normal for PSU to go from 11.97 to 11.70 when it is loaded At 50% to 75%. Another reason why I disliked cheap PSU tester ~$10.

12V × 5% = 11.4 to 12.6 voltage

Anything below 500 mV will be considered as an undervolt in engineering term words. But they may designed prototype then tested to make sure the GPU is work in undervolt event.

As I said, GPU manufacturers may have specifications manual for the graphic processor integrated circuit BGA with normal operating voltage range, but I doubt they will share that.

PS: Yes I understand what undervolting mean, and it is outside of the normal operating voltage range. And you just said 11.97 V is an undervolt for the PSU. This is not my personal opinion, I don't see any of your GPU's processor's normal operating voltage range on your source. So we don't know any normal operating voltage range, except depending on the software to prevented you from undervolting.

1. It's not normal for a PSU to do that unless it's cheap. Mine goes up in voltage under load, not drop. Use a meter to test, not software.

2. You're off topic and don't understand said topic so why are you even here?

1. Yes, it is normal for PSU to change voltage down and up based on load while within 5 %.

2. I didn't go off topic, I just proved example why there is no such as components stay at 12V all times or normal operating voltage range without % range. Your source doesn't have normal operating voltage range. Which is why I said by your logic, 11.97 V out of 12 v will be considered as an undervolt.

[r.linder]

Komarimaru původně napsal:

Jamebonds1 původně napsal:

That is normal for PSU to go from 11.97 to 11.70 when it is loaded At 50% to 75%. Another reason why I disliked cheap PSU tester ~$10.

12V × 5% = 11.4 to 12.6 voltage

Anything below 500 mV will be considered as an undervolt in engineering term words. But they may designed prototype then tested to make sure the GPU is work in undervolt event.

As I said, GPU manufacturers may have specifications manual for the graphic processor integrated circuit BGA with normal operating voltage range, but I doubt they will share that.

PS: Yes I understand what undervolting mean, and it is outside of the normal operating voltage range. And you just said 11.97 V is an undervolt for the PSU. This is not my personal opinion, I don't see any of your GPU's processor's normal operating voltage range on your source. So we don't know any normal operating voltage range, except depending on the software to prevented you from undervolting.

1. It's not normal for a PSU to do that unless it's cheap. Mine goes up in voltage under load, not drop. Use a meter to test, not software.

2. You're off topic and don't understand said topic so why are you even here?

You’re wasting your time. Once he starts going off, he’ll never shut up and he’ll still fight you even if you give solid proof, just ignore it until it goes away. If you piss it off enough, it’ll go onto your profile and leave hateful messages, or try to add you so it can spam you in DMs.

[Jamebonds1]

尺．し工几句ヨ尺 původně napsal:

Komarimaru původně napsal:

1. It's not normal for a PSU to do that unless it's cheap. Mine goes up in voltage under load, not drop. Use a meter to test, not software.

2. You're off topic and don't understand said topic so why are you even here?

You’re wasting your time. Once he starts going off, he’ll never shut up and he’ll still fight you even if you give solid proof, just ignore it until it goes away.

Where is a specifications manual of a graphic processor BGA that claimed 850 mV is an undervolt while being outside of normal operating voltage range? 1800MHz @ 1.1 v is not a normal operating voltage range.