Liquid metal on 3090 water block

[Dari69]

I got a corsair Waterblock for my 3090 FE and I was wondering if it is worth it to apply liquid metal and try and run it at a stable 2100mhz+ coating all of the electronics around the die with clear nail polish it’s going into a duel 360 rad custom loop with a 12 900 K. I found a lot of different answers most of them veered towards yes it is worth it but I haven’t seen any concrete evidence or benchmarks. Anyone can give me their advice you don’t have to have done it but if you have liquid Metaled  A waterblock GPU specifically a higher tier one like a 2080_ 3080_ or 3090 could you tell me if it did anything over just regular old good thermal paste (nh1)

[LIGISTX]

3 minutes ago, Dari69 said:

I got a corsair Waterblock for my 3090 FE and I was wondering if it is worth it to apply liquid metal and try and run it at a stable 2100mhz+ coating all of the electronics around the die with clear nail polish it’s going into a duel 360 rad custom loop with a 12 900 K. I found a lot of different answers most of them veered towards yes it is worth it but I haven’t seen any concrete evidence or benchmarks. Anyone can give me their advice you don’t have to have done it but if you have liquid Metaled  A waterblock GPU specifically a higher tier one like a 2080_ 3080_ or 3090 could you tell me if it did anything over just regular old good thermal paste (nh1)

I wouldn’t, no. Normal TIM has plenty of thermal transfer ability for dies as large as GPU’s. Plus liquid metal doesn’t really fill low spots at all, it works ok for CPU die’s since they are small and have IHS’s pushing down very flat against them, but over the size of a GPU die you really need to get good spread across the entire thing. You may end up with crazy hot spot temps and such with LM. It may work, but the befits really don’t make much sense. A 3090 will be a much larger heat load then my 2080, but my 1080 and 2080 both ran at about 50c on the loop in my signature, with fans at ~900-1100 rpm in games depending on ambient temps. 
 

You may gain a few C, but that is at best 1 boost bin. Is ~15 MHz worth the potential for hot spots that end up frying the chip or potential shorting issues? Where LM is good is for poor CPU TIM under the IHS. A Intel CPU trying to dissipate ~160 watts like my old 4770k was trying through horrible TIM under the IHS, it just couldn’t do it. I saw a 20c drop by delidding it and using LM. But that is a very different use case then a GPU. Just run a good TIM, you will be more then fine. 

[Dari69]

28 minutes ago, LIGISTX said:

I wouldn’t, no. Normal TIM has plenty of thermal transfer ability for dies as large as GPU’s. Plus liquid metal doesn’t really fill low spots at all, it works ok for CPU die’s since they are small and have IHS’s pushing down very flat against them, but over the size of a GPU die you really need to get good spread across the entire thing. You may end up with crazy hot spot temps and such with LM. It may work, but the befits really don’t make much sense. A 3090 will be a much larger heat load then my 2080, but my 1080 and 2080 both ran at about 50c on the loop in my signature, with fans at ~900-1100 rpm in games depending on ambient temps. 
 

You may gain a few C, but that is at best 1 boost bin. Is ~15 MHz worth the potential for hot spots that end up frying the chip or potential shorting issues? Where LM is good is for poor CPU TIM under the IHS. A Intel CPU trying to dissipate ~160 watts like my old 4770k was trying through horrible TIM under the IHS, it just couldn’t do it. I saw a 20c drop by delidding it and using LM. But that is a very different use case then a GPU. Just run a good TIM, you will be more then fine. 

OK I saw frame chasers do this to his 3080 and he said that it worked and he showed concrete evidence so I thought it would be worth it even more to do it for a 3090. He applied some sort of Polish to the cold plate and then

coated both die and cold plate with liquid metal and said there were no issues with the hotspot. Do you think that’s because he applied it very well and the issue you’re describing is from user error? My die it’s perfectly smooth and he applied  that polish wouldn’t it make the cold plate smoother? Solving that issue. Running my fans at max on my loop should give me at least 100 extra megahertz from how putrid the air cooler really is. That’s almost 7 bins applying liquid metal should get me at least one or two what should be if anything at least more stable FPS drops when turning a big corner or something. Have you tried it? on a die at all.

[LIGISTX]

48 minutes ago, Dari69 said:

OK I saw frame chasers do this to his 3080 and he said that it worked and he showed concrete evidence so I thought it would be worth it even more to do it for a 3090. He applied some sort of Polish to the cold plate and then

coated both die and cold plate with liquid metal and said there were no issues with the hotspot. Do you think that’s because he applied it very well and the issue you’re describing is from user error? My die it’s perfectly smooth and he applied  that polish wouldn’t it make the cold plate smoother? Solving that issue. Running my fans at max on my loop should give me at least 100 extra megahertz from how putrid the air cooler really is. That’s almost 7 bins applying liquid metal should get me at least one or two what should be if anything at least more stable FPS drops when turning a big corner or something. Have you tried it? on a die at all.

If the cold plate is copper (which it is), you don’t want to get liquid metal on it as liquid metal eats copper. So you definitely don’t want to polish it down and expose the copper under the zink or whatever Corsair uses as a shiny silver coating (it may even eat through that too tho, not sure what they are using and liquid metal eats all sorts of metals), and even if you flatten out the cold plate the die may not be flat…

 

Basically, chasing a few MHz on a 3090 won’t do anything for you. It won’t smooth out FPS dips, you won’t notice any difference. Also, running fans at 100% on water sort of defeats the purpose. Water cooling custom loops are only “worth it” because of the cooling potential vs noise they provide. I run my PC about inaudibly while in game, and I get 50c GPU temp and low 60’s on the CPU.  Cranking the fans to max fives me a few degree reduction, but…. A few degrees is relatively meaningless. Just use EVGA precision or MSI afterburner to dial in a automatic core clock curve and be happy   I have been overclocking for 15 years, my gtx 470 did a 45% core OC from 607 to 850 MHz. These days, I can work for a few hours trying to get the max GPU core clock possible and the EVGA precision automatic curve that takes 5 minutes came within 1 boost bin of what I was able to get…….. things have changed, and became way more simple. 

[Lord Nicoll]

I can speak directly from experience in using LM on my watercooled Matrix Platinum GTX 980 ti (note an older card but still a hard to find card, I look forward to putting in on LN2 one day) and also using graphite pads on the memory. Yes the temps are pretty great, but quite frankly, don't, it's easy to mess up, note all the shiny areas around the package on the PCB, thats clear coat to stop shorts, and the pink nail polish on the passives around the die, it's a lotta risk for a slightly largish gain but it's not earth shattering. Look on those temps and decide for yourself. (that's idle in case it wasn't obvious, load temps are low 40's)

[Dari69]

2 hours ago, Lord Nicoll said:

I can speak directly from experience in using LM on my watercooled Matrix Platinum GTX 980 ti (note an older card but still a hard to find card, I look forward to putting in on LN2 one day) and also using graphite pads on the memory. Yes the temps are pretty great, but quite frankly, don't, it's easy to mess up, note all the shiny areas around the package on the PCB, thats clear coat to stop shorts, and the pink nail polish on the passives around the die, it's a lotta risk for a slightly largish gain but it's not earth shattering. Look on those temps and decide for yourself. (that's idle in case it wasn't obvious, load temps are low 40's)

ok besides the risk tho i don't care about that im surrounding my pc with pipes full of fast moving liquid almost all of the risk has been done taking my card out and putting it into a water block is risky. But after all of it buying all of the loop parts 2 rads 6 $$$ fans just for them the only thing left besides a shunt mod and voltage mod which im not doing but the LM is not that hard and aint cost much 30 dollars for any performance gain is worth it especially for a 2000$ card that can easily be scalped at 3k  5 or 10 percent at 400 fps is 20+ fps.

. MY question is more of a will i get 4,5,6,7+ percent gain added to the water-cooling performance gain or only 1 fps. If is doesn't do any thing but give me 1 or 2 fps then i could care less but if i does actually make my core go from 1800-1900 to ~2100 then ill do it. sorry for confusion i don't mind doing the LM, all you did was paint around the die correct them applied the LM spread it then closed it and that's it? and what was your performance gain and clock speed gain over reg thermal paste on the gpu. 

[Lord Nicoll]

27 minutes ago, Dari69 said:

ok besides the risk tho i don't care about that im surrounding my pc with pipes full of fast moving liquid almost all of the risk has been done taking my card out and putting it into a water block is risky. But after all of it buying all of the loop parts 2 rads 6 $$$ fans just for them the only thing left besides a shunt mod and voltage mod which im not doing but the LM is not that hard and aint cost much 30 dollars for any performance gain is worth it especially for a 2000$ card that can easily be scalped at 3k  5 or 10 percent at 400 fps is 20+ fps.

. MY question is more of a will i get 4,5,6,7+ percent gain added to the water-cooling performance gain or only 1 fps. If is doesn't do any thing but give me 1 or 2 fps then i could care less but if i does actually make my core go from 1800-1900 to ~2100 then ill do it. sorry for confusion i don't mind doing the LM, all you did was paint around the die correct them applied the LM spread it then closed it and that's it? and what was your performance gain and clock speed gain over reg thermal paste on the gpu. 

If LM spills out from the thermal interface it will in most liklihood kill it so I wouldn't write that risk off entirely. 

Actual temp improvments were not spectacular, it was somewhere in the region of 5°C but the reason I did it was because this kinda crappy Bykski block has a die contact area smaller than the die itself, I think they took the regular 980 matrix card and used that as the basis but forgot the GM102 die is bigger than GM104. Now this being a matrix it has full voltage control, it runs in the low to mid 40's at 320-350 watts total board board with 1.25-1.28v on the core, 1520-1530MHz range, 1.7v on the memory around 2097MHz. The loop has two 360mm radiators, D5 pump, the rads are a CE and XE from EK, so a medium and thick rad. The GPU is the last thing in the loop before the rads so it'll be the hottest. Given it's last and still never hits 50°C even overvolted and overclock, yes it works, but again idk if it'll have the same gains on your case.

As for applying it, yes cover any exposed solder joins around the die, seal under the die package so blobs of LM can't get to the BGA array (if there is no underfill) and apply it to both sides to be mated, not a lot is needed, you need need the surfaces covered with a couple of blobs, it doesn't take much. Capillary action will cause the two sides to join together. 

[Dari69]

10 minutes ago, Lord Nicoll said:

If LM spills out from the thermal interface it will in most liklihood kill it so I wouldn't write that risk off entirely. 

Actual temp improvments were not spectacular, it was somewhere in the region of 5°C but the reason I did it was because this kinda crappy Bykski block has a die contact area smaller than the die itself, I think they took the regular 980 matrix card and used that as the basis but forgot the GM102 die is bigger than GM104. Now this being a matrix it has full voltage control, it runs in the low to mid 40's at 320-350 watts total board board with 1.25-1.28v on the core, 1520-1530MHz range, 1.7v on the memory around 2097MHz. The loop has two 360mm radiators, D5 pump, the rads are a CE and XE from EK, so a medium and thick rad. The GPU is the last thing in the loop before the rads so it'll be the hottest. Given it's last and still never hits 50°C even overvolted and overclock, yes it works, but again idk if it'll have the same gains on your case.

As for applying it, yes cover any exposed solder joins around the die, seal under the die package so blobs of LM can't get to the BGA array (if there is no underfill) and apply it to both sides to be mated, not a lot is needed, you need need the surfaces covered with a couple of blobs, it doesn't take much. Capillary action will cause the two sides to join together. 

so you would say not worth it then. just water cooling alone will give me what i am asking.

[Lord Nicoll]

12 minutes ago, Dari69 said:

so you would say not worth it then. just water cooling alone will give me what i am asking.

Yes a high quality paste should do enough to reduce the temps, along with a lotta rad space. 2100MHz is a tall ask for any RTX 3090 unless it can already do good clocks, you might need to bin cards to get one that can do 2100MHz on even water if it does 1800-1900 currently. Whats the max stable clock you get right now at what temperature. 

[Dari69]

1 hour ago, Lord Nicoll said:

Yes a high quality paste should do enough to reduce the temps, along with a lotta rad space. 2100MHz is a tall ask for any RTX 3090 unless it can already do good clocks, you might need to bin cards to get one that can do 2100MHz on even water if it does 1800-1900 currently. Whats the max stable clock you get right now at what temperature. 

I easily sit at 2000mhz - 1950 in n terms of "bin" but don't because the cooling aint Adequate altho the core stays about 70 c which aint bad for that clock nothing more. But the memory gets a little hot so i lower it cool it even thought its not throttling it gets about 80-90c so  i can put it that high but it will crash once temps rise after doing some thing big like walking through a door. if i were not to use LM i would be using noctua nh-1, and 12.8w/mk thermal pads the amazon ones there in blue packaging i forget what there called. so in terms of stability with 70 c on core (could get hotter but then not stable) and 85-90c on mem the core can sit at 1875 and 10000 on memory when mining i can push the mem to 1350+mhz with out trying so i guess i haven't found my max yet that's with -500 on core and a power limit. All temps and clocks were at 100 percent fan speed in a 62 degree room on a case with no side panel and a smaller 2070 under it.  the 2070 is idle during all of my tests.  sorry for typos im out on my phone.

Edited December 24, 2021 by Dari69

[thrasher_565]

ied putt it this way you no the risks and its up to you. if you cant afford to replace it then dont if you can then up to you. people do do it so....

 

its just like deliding the cpu has hi risk but could have lower temps.

[Dari69]

1 hour ago, Lord Nicoll said:

Yes a high quality paste should do enough to reduce the temps, along with a lotta rad space. 2100MHz is a tall ask for any RTX 3090 unless it can already do good clocks, you might need to bin cards to get one that can do 2100MHz on even water if it does 1800-1900 currently. Whats the max stable clock you get right now at what temperature. 

i got home and edited what i said sorry lmao 

[AnonymousGuy]

I have LM on my 3090.  I just put kapton tape around the die instead of liquid electrical tape or nail polish or whatever.  I haven't had problems with LM going off the die anyways in the past because I always paint it on very thin as intended.  

 

But it's been working fine for the last year of mining.  Max die temperature is 45C with 27C water running at about 650W to the GPU.  Sooooo yeah it's pretty good.

[Dari69]

1 hour ago, AnonymousGuy said:

I have LM on my 3090.  I just put kapton tape around the die instead of liquid electrical tape or nail polish or whatever.  I haven't had problems with LM going off the die anyways in the past because I always paint it on very thin as intended.  

 

But it's been working fine for the last year of mining.  Max die temperature is 45C with 27C water running at about 650W to the GPU.  Sooooo yeah it's pretty good.

what were the numbers before on reg thermal paste, temps and clocks?

[HM-2]

On 12/24/2021 at 12:53 AM, Dari69 said:

so you would say not worth it then. just water cooling alone will give me what i am asking.

Should do. I reckon with a decent water block you'll run into power limitations before you get temperature related throttling. That's certainly the case on my 2080.