Question RTX 3070 Ti Ventus fan speed is linked to GPU usage, not temperature

[krzys6e]

Hey!

For few months I'm fighting with my RTX 3070 Ti Fan speed. When GPU usage is high, RPM go for 3500 and PC sounds like jet is starting nearby. I've tried undervolting (which helped a little bit) and tried changing GPU fan curve (in MSI Afterburner, in Fan Control and in GeForce Experience). Although chaning GPU fan curve was working (I was able to set for example for fans to be at 100% on low temp) I hadn't been able to solve my problem. At high GPU usage something seems to override the curve.

Today I noticed, that the spikes in RPM aren't linked with GPU temperature - they are linked with increase of GPU usage. For example:

When GPU is at 10% usage at 35°C the fans are quiet. But when the usage goes up to 50%, and temperature stays the same, the RPM goes crazy. And in gaming - when GPU usage is at around 70% at 65°C, but then goes to 99% usage but temperature stays the same, the RPM goes crazy again.

I have no idea what to do now. I wanna play Starfield but I know that will be miserable experience when I will have starting jet next to my legs. I have Corsair 4000D airflow case, and even taking of side panel didn't change a thing. The temperatures are fine. The crazy RPM isn't.

Does any of you, by any chance, know what can I do to solve this issue?

 

[kerberos_20]

gpu has multiple sensors, if one of them is reaching thermal limits, it may trigger your max fan speed issue, if your GPU isnt from MSI, then MSI afterburner may not show you all temperatures available and their position on card pcb, as each vendor has its own app
open gpu-z and look there at sensors

 

[krzys6e]

The GPU is from MSI.
The sensors at GPU-Z at 80% GPU usage shows at few minutes stress test at avarege:
65-68°C GPU temp
105°C Hotpoint
62°C Memory temperature

 

[Phaaze88]

The GPU is from MSI.
The sensors at GPU-Z at 80% GPU usage shows at few minutes stress test at avarege:
65-68°C GPU temp
105°C Hotpoint
62°C Memory temperature

37~40C gap between the core and hot spot?
I think we've found the problem. ~20C and below is typical for that gap.

Looks like it's time for a repaste. Put some Arctic MX-6, MX-4, or NT-H2 on it.

 

[krzys6e]

I did a little experiment. I left PC on desktop, then turned on Warhammer 3 total war to do benchmark. Immediately the GPU usage went up to around 70 (still in game menu) and hot spot went from around 55° to over 100° - like in a few seconds. Jet sound followed.

 

[kerberos_20]

The sensors at GPU-Z at 80% GPU usage shows at few minutes stress test at avarege:
65-68°C GPU temp
105°C Hotpoint

if its MSI then it should be showing both GPU temperatures aswell, but...
65-68C is on sensor on PCB
105C is on GPU die

replace thermal paste (TIM)