Liquid Metal on a MSI 3070 ventus 2x

[Frostd]

This GPU sits at 80+ C° average, i have a corsair 570x crystal series case with the 3 stock front fans as intake, 2 corsairs LL120 as exhaust mounted on a kraken x53 radiator and one on the back of the case as exhaust also corsair LL120.

 

I still haven't had the time for a long session of gameplay to see how the 3070 behaves but i am assuming is going to reach 90+ which i don't feel comfortable with. As soon as i start a game on 1080p ultra settings it goes to 80+ right away.

 

The question is can i use liquid metal ? I am afraid the heatsink is made of aluminum in this particular model.

[zarbotx]

Hey, I happen to have the same gpu. Just yesterday I undervolted and I can reach the same clocks but at max 85º on heavy gaming, take in mind that i have a NZXT H200I Case, and the gpu has a bit of a struggle cooling properly, so the results may be even better in your case. Here´s the video I used, hope it works

 

[Frostd]

19 minutes ago, zarbotx said:

Hey, I happen to have the same gpu. Just yesterday I undervolted and I can reach the same clocks but at max 85º on heavy gaming, take in mind that i have a NZXT H200I Case, and the gpu has a bit of a struggle cooling properly, so the results may be even better in your case. Here´s the video I used, hope it works

 

Nice! i am definitely going to try this before doing the risky procedure of applying liquid metal. Thanks!

 

Still for anyone out there who knows if this GPUs have aluminum heatsinks or nikel plated copper.

[Guest]

Hi,

I was having some issues with my new ventus x2 3070, games freezing/stuttering etc.....I was advised in another thread to use afterburner to set the the core clock at -25 and power to 90%....this does seem to have had a good impact on things, with the card running generally around 70ish during gaming instead of 80-83, but I'm just wondering if there's any difference/benefit to doing it this way instead?

[Guest]

also just found this guide on reddit, looks pretty similar to what we have in that video.

 

https://www.reddit.com/r/nvidia/comments/jn878f/small_beginners_undervolting_guide_for_rtx_3070_fe/

[Frostd]

4 hours ago, Glenn-Tidbury said:

Hi,

I was having some issues with my new ventus x2 3070, games freezing/stuttering etc.....I was advised in another thread to use afterburner to set the the core clock at -25 and power to 90%....this does seem to have had a good impact on things, with the card running generally around 70ish during gaming instead of 80-83, but I'm just wondering if there's any difference/benefit to doing it this way instead?

I think it is the same benefit cause undervolting is reducing the power in which it works by default if i am correct. If you undervolt way more than needed either it crashes or just don't send video output.

 

You said you reduced the power that way i don't know what was the procedure you took but i guess it is exactly the same.

 

 

[Frostd]

Still anyone knows what material is the heatsink made? and if someone has applied LM to this GPU.

[Guest]

10 minutes ago, Frostd said:

I think it is the same benefit cause undervolting is reducing the power in which it works by default if i am correct. If you undervolt way more than needed either it crashes or just don't send video output.

 

You said you reduced the power that way i don't know what was the procedure you took but i guess it is exactly the same.

 

 

so instead of using the ctrl+f graph and moving certain points, etc....like in the video and the reddit guide.....on the front page of afterburner, there's a slider for core clock and another for power......i move the power from 100 to 90, and core clock from 0 to -25.

 

Maybe I'll try the long/manual/graph way later and see if it makes a difference.

 

Sorry, no idea what the heatsink is made of though, or if LM can be used :s

[Guest]

Hmmmm, if I use the curve method to undervolt, time spy benchmark fails to complete, just giving an error.

 

Using the sliders to just reduce power limit and core clock, it finishes, and as expected, I get slightly lower score than clock speed, but temps are reduced

[Guest]

ok, i managed to get the curve method working....but just not pushing it as far as the guides say.

 

set the straight line at 1800MHz from 0.875mV onwards (after starting the setup with a core clock -220).

 

Also boosted gpu mem clock by +1000.

 

Benchmarks like time spy now barely get to 70 degrees, with average temp of 68, and a lot of the tests sitting in the mid 60's, clock goes actually up to 1815, but never goes beyond that. the card seem much more stable now, and anecdotally, I'm not really getting stutters anymore (only expected ones, like when a game isn't well optimised and you get loading area stutters etc), and temps over-all seem lower. Those two things were my main concern, but doing it this way I seem to be able to keep the majority of the performance, if not slightly better, and actually more stable while the temps are lower.

 

Still not the most incredible bench scores, but it's stable, better than I got on stock, and lower temps.

 

Could still experiment more, but with it being stable as it is, I might just leave it.

 

[Guest]

managed to push the card a little further, putting it at 1830 for 875mV. If i try and go down to 850mV, I need to drop clock to 1785. Both options seem stable and very similar temps, so probably going to stick with the 1830@875.

 

One strang thing I've noticed is that, so matter what clock I set on the graph, the card usually goes up to around 15mHz above that level....i.e. with it set at 1830, I get clock of 1845 (which is only 15 away from the top speed I benchmarked with stock settings, but with lower temps.

 

Anyone with afterburner experience know whther one option would be better than another?

 

The 2 most stable settings I've found are:-

 

1830mHz @ 875mV

1785mHz @ 850mV

 

pretty similar performance and temperatures on each, but is one safer than the other?

[Frostd]

Very Useful information, i will test this and also post my scores with that particular benchmark and stats

 

Have you tried Superposition benchmark? What is the score, temps and frames.

 

But i think it might vary a little cause of my CPU (i9 10900k) don't think it will that much, i have heard that this CPU runs hot.

 

I have a lenovo laptop with 17 4710 HQ and a GTX 860m temps were at 95+ and i undervolted the cpu overclocked the GPU and removed the dust filter, temps dropped to 70+ and only 3-4 frames higher lol.

 

It started crashing because i overclocked way too much.

[Guest]

Im basically back to the drawing board. Lol.

 

Seemed stable, and even played online for hours last night... gone to do extract same benchmarks this morning and got glitches and crashes, so I think I was trying to undervolt too much to get temps down.

 

Trying some less aggressive undervolting now. Once I find something that works I'll try super position too

[Guest]

Something im noticing often though, is that the curve moves ALOT.

 

Like, I can set a certain voltage at say 1860, but in tests im getting peaks of 1890, go to look at the curve and it's moved itself up to 1875 or something. 

 

Best way ive found to try and combat this..... and something im trialling right now, is a method I saw online where rather than using the sliders in the main after burner window to bring the whole curve down, you hold alt and drag from the last point.... it moves the whole curve still, but your slider in main window stays at zero.... then you drag your chosen mV level single point up to meet your desired frequency and hit apply.... however, what I then do is click on that point again, move it down by 5 points and hit apply.... what you might see is that after you hit apply, although the change value will be different (e.g. from +260 to +255), the frequency has shifted to be the same (e.g 1860 still on 1860). Keep testing that until the curve STOPS being a straight line... then put it back up by 5 points. 

 

Currently testing -60 from the end point, 887mV at +255. Gives me a clock of 1830 to 1860 (1860 was my peak during tests at stock values).

 

All very confusing still though, and finding something that works reliably is so hard.

 

Im also trying to leave the mem clock at stock, as I worry about over clocking things :S

 

my aim isn't to get BETTER performance than stock, I just wasn't the same but lower temps.

[Guest]

what I've go currently is:-

 

ALL SLIDERS on main page left at default.

 

press Ctrl + F to open the voltage curve

 

hold ALT and click and drag the LAST POINT of the curve to -60 (this put it at 1845). Click apply in the main window.

 

click and grad (not holding ALT now) the point at 887mV to +255, which sets THAT to 1845...click apply in the main window...if I go down 5 clicks, the straight line is no longer straight, so I know this is the right spot for the graph to stay at 1845....although in actuality, it's still 15mhz higher, go figure!

 

This gives me a pretty solid and stable 1860Mhz clock rate throughout all the tests.

 

 

 

ok, so here are my result from superposition at 4k optimised, 1080p high and 1440 high, which is the native res of my screen, and what I want to play games at (whole reason for getting 3070).

 

In other games I usually get higher fps than this bench though at 1440....but obviously, the point of the bench is to push things beyond normal gaming?

 

Also attached the 3dmark time spy results, which also seem fine.....so fingers cross this is stable. Not as MUCH of a saving in temps as I'd like, but if I start going lower in mV it loses stability for sure.

I can probably boost the artificial benchmark scores by putting the mem frequency up, you can apparently take them to +1000 on 3070 cards....but I've found it made no difference to temps and really no difference to fps.....so I don't see the point....it's LITERALLY just booting your benchmark score, but giving the exact same performance in games......pointless.

I'll have to keep playing things over the next few days to make sure it stays stable and that the temp drops remain soli in games, not just in benching.....got to go to work now though.....fingers crossed

 

[Guest]

Just to update....I can't find any way of keeping it stable when using the graph/ curve method.

 

So what I'm trying now is just reset everything to default, and just set the max power slider in the main msi window. If I set it to 75% temps generally stay below 70, but I do lose about 5-10fps in benches (5 in time spy, 10 in superposition).

 

To me, this seems like a worthwhile trade, and I'm not getting any more crashes or glitches while running the tests, which is nice. Heaven benchmark still stutters in places, but it does it in the same places regardless of what I try, so I think it's just a bit of a dud test for the 30xx cards? Or a bad download? Still good as a free stress test though..... leave it running for a while and check your pc hasn't crashed or that temps don't go through the roof. The still getting around 120fps (as that's what i artificially limit to anyway, never seen the point going higher, it's impossible to see the difference), and again, so far, temps stay below 70, which is what I was aiming for. 

 

So for me personally, with my exact card, I think power limit to 75% is probably what im going to stick with and just test some games when I eventually get time. Lol.

[Frostd]

First test without changing anything using all stock frequencies.

I noticed my clock frequency is higher but everything else is lower compared to your tests, but running at 77 C°

 

 

 

Ran all benchmarks in a row and seems i stay a 83 degrees, i wonder why at QHD i get less points, at 4k very little difference and at FHD a big difference. I don't understand why, don't think the temperature is affecting the score and probably the CPU is the one giving me higher results, but why at 1440p (QHD) i get lower.

 

Next post will be with the scores after following the undervolting guide.

[Frostd]

I just saw that msi afterburner has a degrees limit which was set to 83.

I did another test unlocking that limit and ramping fan speed to 100% of the GPU, this is the result.

 

4 degrees less, and 2 frames average more, i guess temperature does affect the score.

 

[Guest]

those are some good results.....do you not find that the noise from the card is unbearable at those temps though? I think I got pretty unlucky with my card lol. Anything abvoe 71 for a few minutes and it's soooo loud.

interesting that putting the temp limit UP actually makes the card run cooler?

I wonder if we could achieve the same thing without raising the limit, but by changing the fan curve?

by the way, my current test is using teh voltage again. I set a core clock of -200, the brought the 887mV (I think??? not at home at the moment) point up to 1845Mhz. This gives me a clock in games/test of anywhere between 1830 and 1860, never going above 1860, but crucially, temps barely reach 70 now. I have also changed the airflow of my case, so I now have 2 front intake, 1 top exhaust and have kept the fan/radiator/fan setup at the rear.