[SOLVED] GPU throttling @ 70c.

[Nihilistenthusiast]

So, before I start, here is my hardware:

MSI Z390-A Pro
Corsair Vengeance LPX 2400mhz (overclocked to 2800mhz currently)
i5-9600k @ 4.5ghz with a thermaltake tower cooler (forget the model, it fits a 120mm, never exceeds 58c)
Gigabyte 1660 OC Edition
Antec HCG 650w 80+ Gold

Problem:
Trying to dial in an overclock on the gpu. Currently have +90mhz on the core and +75 on the mem. When I hit 65c I drop about 15mhz off the core, and @ 70c, I drop 30mhz. Using MSI afterburner to do the overclock. Fans, Core Voltage, and Power Limit set to 100%. The stock temp limit is 83c by default, but cranked it to 90c. Highest temp on the gpu since the overclock has been 71c(normally doesn't go above 68c). Under load, the core voltage hits 1031mv and stays there, at idle it sits at 731mv (no idea if this is normal)

Could the vrm be heating up and causing the core to down clock? How would I test this?

I'm new to overclocking so any advice is appreciated.

 

[hotaru.hino]

If you think it's temperatures, you can try setting a more aggressive fan curve.

However, I don't consider this a thermal throttling scenario unless the GPU can't even maintain the base speed.

 

[Nihilistenthusiast]

The fans are already set to 100%.

And I don't think it's a problem on the core temp. @ 60c, I get the full +90mhz on the core, but as soon as I get up to 65c, the core starts to down clock, even if it is only slight, I would still like to pin point why. As far as I understand, Afterburner only reads the core temp of the GPU so my theory is something else is either hitting a power/heat limit or both.

 

[Phaaze88]

1)Nvidia's cards use 15mhz increments.

2)Ever since 10 series: the Gpu Boost algorithm has several temperature thresholds, unlike power, which just has one.
Power limits aside, a few degrees CAN make the difference in holding certain OCs.
The cooler these cards run, the better.

3)Core clock settings you make in software are not absolute, but more of a suggestion; the card will still downclock if it's 'not comfortable' with its parameters(power & thermals).
Probably the best things that can be done with Pascal(10), Turing(16 & 20), and now Ampere(30):
A)Max out the power limit slider in software.

B)Keep it as cool as possible.

C)Find the 'optimal' frequency settings; squeeze as much performance out of the card while tripping the power and thermal limits as little as possible.
A heavy OC isn't going to be great if it frequently trips power or thermal limits; it'll throttle back harder if it does, and that'll just hurt average performance.
This, hands down, will require the most trial and error because the card will behave differently across games - benchmarks/stress tests aren't an all in one representation either.

Easy enough to just do the first 2 though.

 

[Nihilistenthusiast]

The 90mhz (15x6) was set by running the OC scan in Afterburner, but after more messing around/testing, something interesting happened in that the card will turbo up to 2055mhz (I didn't do this and never saw it reach that during the OC scan), but I can actively track when the card will throttle down. @ around 52c, it drops to 2040mhz, @ 58c - 2025mhz, @ 60c - 2010mhz, @70c 1995mhz

Im also noticing the core voltage is dropping at every increase in temp noted above, so when the card does hit 2055mhz, It's reading 1.05v and will drop to 1.031v but still run @ 2040mhz until the temp increases to what I suppose is the next tripping point.

Just to be clear then, does this sound like normal operation in a way in your opinion? Should I downclock to just run @ 1995mhz until I find a better cooling solution?

 

[Phaaze88]

That is normal.

Whether to leave it at 1995mhz is up to you. Like I mentioned earlier, the card may not run those clocks in all the games you currently play:
It may sustain higher than 1995mhz in some titles, or even lower than 1995mhz in others.

 

[Nihilistenthusiast]

I guess what I mean, is it safe to let it hit the temps and down throttle on it's own so long as I don't get a full thermal throttle? I don't plan to increase the core frequency anymore than it is now.

 

[Phaaze88]

Oh!
Yeah, that's fine.

 

[Nihilistenthusiast]

Awesome, thanks for all the advice. As mentioned, I'm a noobie to all this so it's very much appreciated.

 

[Karadjgne]

What do you mean for the power limit, voltage etc at 100%? Do you mean you left the sliders alone, and those are stock 100% values, or did you move those sliders to the max? I only ask because generally with a gpu OC, there's some movement required on the power limit slider, just so the card has some room above stock requirements to actually use the increased clocks.

 

[hotaru.hino]

Also in the grand scheme of things as well, 15-30Mhz when you're already floating around 2.0GHz is between 1-2%, which is already within a margin of error. Unless you were trying to aim for leaderscore boards, I wouldn't really worry about it.