Question GPU Drivers crashing

[hunterczech]

Hello,
I want to undervolt my RTX3060 to run cooler at same performance but I'm currently facing weird issue. When I lower core clock (even by 80mhz), then start Heaven benchmark and quit it the GPU drivers crash every time (I have 2 GPUs so I just reset setting on my other monitor and GPU jumps back up. Drivers are 512.59 from 20.4.2022. Is it because setting lower GPU clock actually lowers it below level it requires to even render monitor causing it to crash? As the GPU is sitting at 210Mhz (405Mhz Boost) when idle so giving it -80 puts it at 130mhz or something?

EDIT: Crash only happens when quitting Unigine Heaven. So far did not happen anywhere else.

PC:
APU: Ryzen 5600G with enabled Vega 8
GPU: RTX3060
RAM: 32gb 3200mhz
SSD: Samsung 860 EVO
PSU: Seasonic Focus Gold Plus 650W
System is about 4 months old (GPU about 2 weeks).
Software for underclocking/undervolting: MSI Afterburner

 

[Lutfij]

Just an FYI, your APU's iGPU is disabled when you have a discrete GPU in a build. As for your build, you might want to use DDU to rid all GPU drivers from your platform, disconnect from the www, reboot and then manually install your GPU driver in an elevated command, i.e, Right click installer>Run as Administrator then try lowering the clocks without tampering with the voltage/power slider.

AIB maker of your RTX3060?

 

[hunterczech]

Just an FYI, your APU's iGPU is disabled when you have a discrete GPU in a build. As for your build, you might want to use DDU to rid all GPU drivers from your platform, disconnect from the www, reboot and then manually install your GPU driver in an elevated command, i.e, Right click installer>Run as Administrator then try lowering the clocks without tampering with the voltage/power slider.

AIB maker of your RTX3060?

I have had my iGPU up and running by default (even reserved 4GB of my RAM for it). Just had to toy with the drivers a bit to have Radeon drivers along Nvidia ones. Since then I like having two GPUs as each of these can focus on different tasks. I was told many times it can run side by side with no problems. I have already got rid of previous nVidia drivers before (when I changed 960 for 3060).

I've also tried to set custom mV/MHz curve without touching the starting points (up until 950mV) and now it doesn't crash. So is it possible that the theory of mine is true?

EDIT: AIB is PALiT

 

[tennis2]

I have had my iGPU up and running by default (even reserved 4GB of my RAM for it). Just had to toy with the drivers a bit to have Radeon drivers along Nvidia ones. Since then I like having two GPUs as each of these can focus on different tasks.

Explain how you're achieving this?

I was told many times it can run side by side with no problems.

I believe you were misinformed.

Are you doing this [presumably] via MSI afterburner? FYI, setting a positive frequency offset is how you undervolt. Basically tells the card, if you ran 2000MHz @ 1.0V before, run 2000 + x MHz @ 1.0V now. Or explained the other way, if 2000MHz = 1.0V before, you'll hit 2000MHz at <1.0V now.

 

[hunterczech]

Explain how you're achieving this?


I believe you were misinformed.

When I first started my PC I have noticed that I only have 28GB of RAM. iGPU was enabled. It didn't show any display at the time due to Microsoft installing faulty drivers for it. I disabled internet, uninstalled faulty AMD Driver, installed official Radeon driver and since then I have two GPUs. Had only one BSOD yet (months ago back when I've had GTX960 and that BSOD was because I tried to install Emulator with SVM off lol)

And how do you know that two GPUs can't work independently on one PC? I didn't have any issues with it yet.

EDIT: I'm even playing games at both GPUs at same time without single crash

Are you doing this [presumably] via MSI afterburner? FYI, setting a positive frequency offset is how you undervolt. Basically tells the card, if you ran 2000MHz @ 1.0V before, run 2000 + x MHz @ 1.0V now. Or explained the other way, if 2000MHz = 1.0V before, you'll hit 2000MHz at <1.0V now.

Yes I'm doing this with Afterburner. I'm currently setting negative frequency to achieve 850mV @ 1800MHz which decreased temps from 67C (81C Hotspot) to 52C (65C hotspot) with score difference of just 7%. This one was however unstable and crashed my AMD driver lol. But 900mV @ 1800MHz is working fine.

 

[tennis2]

Please explain how many monitors you have and where the HDMI/DP cable from each monitor plugs into your PC.

 

[hunterczech]

Please explain how many monitors you have and where the HDMI/DP cable from each monitor plugs into your PC.

Main Monitor 1080p HDMI - RTX 3060
Secondary Monitor 1080p DisplayPort>HDMI reduction - MOBO (Vega 8)
Third Monitor 720p DVI-D - MOBO (Vega 8)

Also the temps are at much lower fan speed, so PC is overall like twice as quiet now lol (before 85% fanspeed, now 65%)

 

[tennis2]

I would encourage you to just connect all the monitors to the 3060. Running a display doesn't stress a GPU at all TBH. The other benefit that has is to eliminate variables in our troubleshooting efforts here by taking the IGP out of the equation.

 

[hunterczech]

It started crashing even when no clocks were changed or manipulated. Problem was MSI Afterburner - it was cause of these weird issues. I uninstalled it and did not have it ever since. This however means I cannot undervolt my GPU, which is unfortunate - I always want my GPUs to last as long as they can and with this terrible overheating it won't last long (now without my fan curves the temp is 81C (hotspot 96C) in under one minute of stress test).. is there any other way to do GPU undervolt?

EDIT: I used thundermaster till this day, uninstalled it and installed MSI afterburner as it allows undervolting. Problems started today ofc, im using this setup for 2 weeks and with old GPU about 4 months.

 

[hunterczech]

OK so I investigated further and crashes are happening even without MSI Afterburner. It however only happens when quitting Heaven benchmark, so far it didn't happen anywhere else.. i will check further what could be the problem.

EDIT: Forgot to state that I'm on Windows 11 Beta Channel (22610.1 ni_release)

 

[tennis2]

Again, I'd recommend just getting everything on the 3060 and disabling the IGP.

 

[hunterczech]

If I disable iGPU and something happens to 3060 I will be doomed as I won't have any display at all. I have bought 5600G for this exact reason. I could have gone with 5600X but I wanted that secondary GPU which came in handy already multiple times.

And I've already done multiple things like running cyberpunk 2077 on ultra + ray tracing ultra for hour, Furmark for 20 mins and all that with undervolted GPU and it's pretty stable. I've even reduced power limit to 80% so it draws 135W tops while retaining 95% of the performance.

Only situation where driver crash happens is on exiting Unigine Heaven 4.0 - it crashes either nvidia driver or AMD driver. Maybe its not made to work with multiple GPU setups or it just doesnt like my rig?

EDIT: I can even leave the Unigine Heaven loop forever and its fine but once i quit it crashes driver.

 

[tennis2]

If I disable iGPU and something happens to 3060 I will be doomed as I won't have any display at all. I have bought 5600G for this exact reason.

You just plug all 3 monitors into the 3060 and leave the BIOS setting for default graphics adapter on auto. If the 3060 dies, just plug monitors into the mobo display outputs. It's really a non-issue.

I also bought a 5600G to have a backup IGP, but my 2 monitors are plugged into my 3060Ti