Undervolting an awkward beast

[Dragsham]

The dreaded gigabyte "eagle" 3070ti.

I'd recently fitted a new z390 board, 4k ballastix ddr4 ram, i9 9900k.
Net result....bloody hot! Ambient 50°.
3 front intakes 2 upper n back exaust fans seemed to make little difference.
Luckily the cpu has it's aio so its running at a cool 45°.

Every time I fired up elite it crashed within a few minutes. My external sensor panel showed the culprit.
Gpu thermally throttling at about 86° upwards.!
Whats it doing that hot?

It took me 100 reboots (including reruns of bench software) and alot of afterburner tinkering to get it right.

Undervolt:

Power at just 80%
Core clock -200
Memory left alone 3070ti don't need it.
Temp set to 88° TEMP as priority not power...very important..they need to be unchained too so powers 80% temps 88°.
Enable voltage adjustment.
Go into curve editor.
Having set those settings above now I selected (finally after hours!) 912mv pushed the slider upto 1770 stock mhz.
Boom! 90 fps average whilst temp ingame just 75°
Sheesh!!
Now that's saved try the over clock.

Core +200
Power 80%
Temp 88°
Set them, and now into curve editor.
975mv slider upto 2010mhz
Save that.
Nice fps boost. Runs a bit hotter at 81°.

My psu is a 850w seasonic titanium I highly recommend it.
But my question is...
If I had a more powerful psu would it have made a difference ?

Ohhh yes course elites running perfectly now in either mode.

 

[STRONTIUM DOG]

whenever you lower the voltage in an ac circuit you up the current so it will run slightly hotter on psu primary transformer coil
doubt having a more powerfull psu would make any difference as once the load is at max demand (gpu its hottest) it wont need any more power
and as long as max demand is lower than the 850w it can output the performance wont change

 

[Dragsham]

Good. 850w is enough...glad that's settled thx

 

[Morbad]

Every time I fired up elite it crashed within a few minutes. My external sensor panel showed the culprit.
Gpu thermally throttling at about 86° upwards.!
Whats it doing that hot?

For either hotspot or GDDR6X temps, this would be entirely expected under load, at stock. It also should not be crashing, even at higher temperatures.

Undervolting is certainly a way to reduce temps and increase performance per watt, perhaps even absolute performance, but it should not be needed for stability.

If I had a more powerful psu would it have made a difference ?

No.

 

[AkenBosch]

These RTX 3000 cards are power hogs. Efficient power hogs (frames per Watt), but still power hogs. My current cramped case can barely hold in check a 200W 3060 Ti, no throttling but I'm not fond of fans trying to keep it in the 80 °C range under load. Can't be bothered with excessive tweaking anymore either (I left Morbad's enthusiasm for the matter way behind me in my distant past ), so I just resorted to cut max power at around 80% of capacity, card doesn't draw more than 160-170 W, stays easily below 70 °C, runs silent and still has plenty of juice for 1080p performance. If I really need those few frames more, I'll just get a better case, or remove the panel from this one. None of this undervolting and fan curves shenanigans ...

(And as Morbad mentioned, hotspot temps are much different, and higher, from GPU ones, a card should never crash no matter the temp, just throttling down or at the very exceptional worst go into thermal shutdown, and the PSU has no business on the GPU thermal management)

 

[Dragsham]

Yeah looks like a waterblock jobby....thermal pads etc

 

[Dragsham]

Damn... its still crashing...or rather just locking up in the headset.
None of this happened with the Omen board I replaced with this z390 aorus elite.
So I'm at a loss...
It's thermals I know it is...if l krank the game down in vr to say, 40% it ticks along nicely... mind you I haven't played it like that for more than a few minutes it's so low res.
With the omen board the 3070ti ticked along nicely even up at over 100% res on steam slider. Never any heat issues.
So I'm wondering if this aorus board has a problem. I know it consumes about 160w thereabouts under load. Which is probably alot more than the omen board.
Anyways I'll plug along. I'm at 80% res steam slider atm 75% power 88 temp 818mv UV @ -200mhz which gives an almost stock core clock. Just a tad over.
Also I re applied thermal paste to the gpu chip.. cleared the sinks of crap. Got about 3 or 4° cooler overall.

 

[Dragsham]

So finally after a marathon I think I've cracked it.
818mv was too unstable so upped to 825mv undervolt. -200mhz core clock.
Power at 100%
Temp at 90% prioritising temp.
Ran unigine bench 3k per lense 100% steam slider.
Averaging a chugging 45fps average. Mmm well its playable !
Seems this mainboard made a big thermal footprint hehe.
Temp never went over 83°. Past that it'd start throttling back grrr.

Seems OK @!
If it ain't I'll post later why hehe...

 

[GroG79]

I got a new GPU too and after that everything was crashing.. Had to do my underclock on the CPU again. When I am happy with temps I will see what power usage I can save on the GPU side of things.
Doing things step by step really helps achieve results but it is a slow process.

 

[Dragsham]

right i feel i can report back sucessfully.
ss .65
hmd 2.0
vr medium with a few tweeks. but fx quality is medium. ambient occlusion is off for now. bloom blur off.
hp g2 reverb @90hz so 11.1 in vrfps optimally.
steamVR set to 48% in general "render resolution". in video, "per application video settings" i doubled it to 96%.
mixed reality portal is set to high quality, both.
only change i made in nvidia panel was power managemant mode to performance.
now the important bit.
afterburner set to 975mv 2050mhz core. power @ 110% (think my psu 850w copes but idk?) temp 83 degrees with link off, temp as priority.
only time im out of the green band (11.1 90fps) in vrfps is wen im at a base. pink n red etc. BUT it dont crash! spent 2 hours solid tonight mission running so it works!