GX502GV Memory clock drops for few minutes in games on Silent mode

[Zlapsa]

Dear community, I am facing some kind of bug while playing games in Silent mode. Constantly in game I am getting sudden GPU Memory drop to 810Mhz from 7k for about a minute or two and then it bumps up to 7k Mhz again.
It happens constantly - every ~10 minutes of gaming there is 30s-1m-2m drop and after those few minutes - all is working fine again and frequency bumps to 7000 again.
I know that gaming in silent is not good experience, but I cannot play with more noise in my room only at night I can switch to performance mode. And I am fine with less FPS on Silent mode! And extremely happy with CPU freq kept @3Ghz in this mode, because i7 is hell of a beast.
It seems like some software problem, because in silent mode there is stable temperature on CPU (80deg) and GPU(75deg). For example - while playing in Performance mode - temperatures are more higher, because software not limiting CPU to 3Ghz and CPU hits ~95deg and GPU about 80deg.
And one more thing - if MEM FREQ drops - I can manually turn Performance mode with FN+F5 for breef second and shift back to Silent and MEM FREQ will bump up! Thats why I think it is software problem! There is also second option how to force bump GPU FREQ up - is open in-game menu, but it seems that only few games affacts it. It seems, that if you open in-game menu and in-menu there is no need for performance - MEM FREQ drops lower and after closing in-game menu GPU/driver/software knows that it need to bump up FREQ again and does that. For example - it is not happening in GTA IV, because FPS in in-game menu seems to be stable (80FPS) as in game, so GTA IV renders game even in in-game menu! (I hope you will understand what I wrote here with my bad english 🙂 )
I tried to start "manual" profile with MAX 50% of fan speeds in Armoury crate, but no luck - with Manual profile software is not limiting CPU to 3Ghz so it hits max temperature!
P.S. Feature request if ASUS listening - add more functions to manual mode, for example - limit CPU to 3GHZ like in Silent mode!

Maybe someone had this issue and fixed it?
I tried a lot of fixes - updated Armoury crate, NVIDIA driver, chipset drivers, played with some tweaks on XTU to lower voltage, but no luck - every 10-20minutes I get 2 minutes of 25% FPS and it is annoying. Did not expect this high-grade gaming laptop to throttle/lag like this.
Last games tried - GTA IV and Black mesa. In black mesa there is even no 100% load, GTA IV seems to have more impact, because it is likely to be more hardware intensive game.

If there is no answer - will write to support. Thanks!

Pictures to show more what I mean below! I even did not played, just left Nico Bellic in car and created print-screens for 30minutes!

Silent mode with all OK (MEM Clock 7001):


Silent mode after drop (see MEM Clock 810):


Data with Performance mode (see that temps are higher! AND NO MEMORY FREQ DROPS!!!):


Manual mode with 0-50% CPU and GPU fan speed (EVEN HERE - NO MEMORY FREQ DROPS in these temperatures):


All data was captured constantly in one time, ambient temp 23deg, every screenshot captured after 5min game was running for avaraging temperatures!

Laptop: GX502GV/i7-9750H/16G/2060/512 P660
BIOS: 308
WIN 10 PRO, updated
nVidia drivers 445.87

### GAMES TESTED:
Black Mesa
GTA IV
Metro Exodus

### Data from Armoury Crate app:

Armoury Crate UWP App 2.6.14.0
ROG Live Service 1.0.24.0
AURA Service (Lighting Service) 3.03.95
Armoury Crate service 2.6.17 --->>> I reinstalled it 3 times and downloaded and installed Armoury Crate Service app from asus support page with version 2.6.5, but it shows lower version!
ASUS Aac_NBDT HAL 2.2.18.0
ASUS Keyboard HAL 1.0.33
ASUS MB Peripheral Products 1.0.29
ASUS Mouse HAL 1.1.0.8
AacAudioHal 1.2.1
AacDispaly 1.1.21


Sorry for my bad English! If you cannot understand somethink - just ask, will try to re-write text for better understanding 🙂

[Aldracity]

Necro time! I think I might've found why the problem happens, and one possible solution.

For some reason, the Silent preset has a ~76C thermal limit on the GPU, and if you ever hit that limit it'll chokeslam your GPU until it cools back down.

My solution is to undervolt the GPU. I used this guide: https://linustechtips.com/topic/1085513-laptop-undervolting-guide-intelnvidia/

If you've read that, my undervolt on the GX502GV RTX 2060 was stable with 1800 Mhz at 850 mV (+172). Hitting "apply" might bump that to 1815 Mhz at 850 mV, which is still stable, and if you can't get those tiny-ass dots to line up with 1800, then picking ~1801 will still work properly (selecting ~1799 drops the clock to 1785).

Undervolting past that was stable in Unigine Heaven and Furmark but crashed in Genshin Impact of all things; I'm pretty sure undervolting too far doesn't leave enough power for the VRAM when it tries to load in new stuff on the fly while also running at high frequencies.

I shaved off about 2-3 degrees by doing this, which isn't a ton, but it was enough to completely eliminate the regular semi-freezes on Silent mode. The downside is that you also have to manually re-apply your GPU undervolt every time you switch power profiles in Armoury Crate, but thankfully the undervolt also means Silent now has performance that's roughly comparable to the Balanced preset (at least, in Heaven, I haven't tried other benchmarks yet).

EDIT:

Sat down and torture tested with Heaven and Prime95 running at the same time. It starts throttling at 75 C, then does the chokeslam at 76 C. It seems like the 76 C throttle will absolutely murder the power limit (~650 mV) and will hold it there until the system cools back down to 70 C.

It looks like the undervolt will keep thermals lower in general, prevent temperatures from spiking past 75 C (peak power consumption is like 1100 mV+, capping that to ~850 mV seems to avoid the power spike) and even if you hit 76 C the "crash" only seems to throttle the GPU to ~1200 Mhz without the VRAM throttling.

It's possible that the huge forced power drop from 1100 mV to 650 mV is causing hardware errors, while dropping from 850 mV to 650 mV is within normal load variance.