[SOLVED] GPU temps: Do I need to re-paste? Or is it the different case?

adsoyo

Distinguished
Mar 26, 2012
68
4
18,565
This is in my custom water cooled living room PC. I just finished upgrading from a 980 Ti to a MSI Gaming X Trio RX 6800 XT. I'm running this exact same card on my main battlestation PC with the exact same Alphacool water block. I'm also running similar loops on both with all EKWB components. The battlestation PC has GPU idle temps of ~32C with hotspot at ~37C. After 10 minutes of Heaven Benchmark it's at 53C and hotspot is 63C.

The living room PC idle's roughly the same at 30/35C but then tops out at 60/90C in Heaven. It initially hit 75/105C in Heaven and 3DMark but then I took the back plate off, tightened the 5 screws holding the water block around the die, then reinstalled the back plate and got the 60/90C temps. I initially had the screws only snug according to the instructions. Not hot enough to thermal throttle but I knew the water cooling should've performed way better. I used Arctic Silver 5 paste instead of the included Alphacool paste and evenly spread it across the die according to the instructions.

The living room PC's case is a Fractal Define 7 so it has a front door and sound dampening mats on the sides. It has 1x front 280mm rad with 2x 140mm intake fans and 1x top 240mm rad with 2x 120mm exhaust fans, 1x 140mm rear exhaust fan and an extra 140mm intake fan on the floor under the reservoir. Radiator fans are spinning just under "too noisy" threshold which is +/- 1500rpm. When I open the door temps drop by 6C.

By contrast the battlestation has a Fractal Meshify S2 with an open grill front and no sound dampening. It has 1x front 280mm rad with 2x 140mm intake fans and 1x top 360mm rad with 3x 120mm exhaust fans, 1x 140mm rear exhaust fan and an extra 140mm intake fan on the floor under the reservoir. Fans are spinning a little slower at +/- 1000rpm. I can't remember if I used the Arctic Silver or Alphacool or Thermal Grizzly paste on this card, it took a couple attempts for me to get it right so I was trying different brands.

Anyway I would really like to avoid the huge PITA ordeal of reapplying thermal paste if I don't have to. I know 60/90C leaves a ton of breathing room until thermal throttling but it still seems high to me.
 
Solution
Hmm...

I take it x/xC is core/hot spot?
Memory junction temperature is ok? GDDR6 is good at 95C or lower: https://www.micron.com/products/ultra-bandwidth-solutions/gddr6

The living room PC's case is a Fractal Define 7 so it has a front door and sound dampening mats on the sides. It has 1x front 280mm rad with 2x 140mm intake fans and 1x top 240mm rad with 2x 120mm exhaust fans, 1x 140mm rear exhaust fan and an extra 140mm intake fan on the floor under the reservoir. Radiator fans are spinning just under "too noisy" threshold which is +/- 1500rpm. When I open the door temps drop by 6C.
That's a difference by just enough to note that 'chassis airflow is ok, and could be a little better', but does not make up for that...

Phaaze88

Titan
Ambassador
Hmm...

I take it x/xC is core/hot spot?
Memory junction temperature is ok? GDDR6 is good at 95C or lower: https://www.micron.com/products/ultra-bandwidth-solutions/gddr6

The living room PC's case is a Fractal Define 7 so it has a front door and sound dampening mats on the sides. It has 1x front 280mm rad with 2x 140mm intake fans and 1x top 240mm rad with 2x 120mm exhaust fans, 1x 140mm rear exhaust fan and an extra 140mm intake fan on the floor under the reservoir. Radiator fans are spinning just under "too noisy" threshold which is +/- 1500rpm. When I open the door temps drop by 6C.
That's a difference by just enough to note that 'chassis airflow is ok, and could be a little better', but does not make up for that ~30C gap between the gpu core and hot spot.
10-20C is a more common range between them, and when the gap is even wider, I would suspect one, or a combo of the following:
-how the user handled mounting the cooler; incorrectly, or missed something.
-the paste application; may not have achieved full coverage.
-the paste used. Some do not work as well on naked die as well as others; the surface is smoother than that of a cpu's IHS, and some pastes get 'pumped out' more easily on such a surface with thermal cycles.

I would suggest using the same paste that you used on the battlestation - you need to try and remember which that was - and use it on the living room PC too, as well as mounting the cooler/block in the same manner.
After that, you should have your answer - again, I really doubt it's airflow with that big a gap between core and hot spot.
 
Solution

adsoyo

Distinguished
Mar 26, 2012
68
4
18,565
Hmm...

I take it x/xC is core/hot spot?
Memory junction temperature is ok? GDDR6 is good at 95C or lower: https://www.micron.com/products/ultra-bandwidth-solutions/gddr6


That's a difference by just enough to note that 'chassis airflow is ok, and could be a little better', but does not make up for that ~30C gap between the gpu core and hot spot.
10-20C is a more common range between them, and when the gap is even wider, I would suspect one, or a combo of the following:
-how the user handled mounting the cooler; incorrectly, or missed something.
-the paste application; may not have achieved full coverage.
-the paste used. Some do not work as well on naked die as well as others; the surface is smoother than that of a cpu's IHS, and some pastes get 'pumped out' more easily on such a surface with thermal cycles.

I would suggest using the same paste that you used on the battlestation - you need to try and remember which that was - and use it on the living room PC too, as well as mounting the cooler/block in the same manner.
After that, you should have your answer - again, I really doubt it's airflow with that big a gap between core and hot spot.

Yes correct about x/xC. That 30C gap between GPU temp and hotspot temp looked way too wide. Unfortunately all signs are pointing towards a repaste. Temps are well within spec so I won't rush to get it done but I think I'll clean the Arctic Silver off and use the included Alphacool paste. Thanks for your input.
 
  • Like
Reactions: Phaaze88

Karadjgne

Titan
Ambassador
Not a fan of AS5, especially for liquid cooling. It's mediocre for aircooling, but liquid cooling benefits usually from a slightly tighter compression between IHS and cold plate. With aircooling you'll generally get @ 200 heat cycles before it hardens up, which is still good since it's not the wet part of the pastes that transfer the heat energy, but the silver and silicates in the paste itself.

With the higher compression, on a cold plate, and the silicon nature of gpu dies, AS5 tends to kinda suck. Worse than mediocre. I'd use the included AlphaCool paste, it has finer silicate composition than AS5, so transfers the heat energy in a better pattern.

If you look at a cpu die, first that heat radiates through the tim, then the IHS, so a broader gap with larger particulate is fine. A gpu die is direct heat, no radiation, so by necessity every single transistor core (there's thousands of cores) must be covered perfectly. A larger particulate, with bigger gaps between, doesn't cover all the cores with anything except wet, allowing certain cores to see very minimal transfer and consequent higher Hotspot temps.
 
  • Like
Reactions: adsoyo

adsoyo

Distinguished
Mar 26, 2012
68
4
18,565
Not a fan of AS5, especially for liquid cooling. It's mediocre for aircooling, but liquid cooling benefits usually from a slightly tighter compression between IHS and cold plate. With aircooling you'll generally get @ 200 heat cycles before it hardens up, which is still good since it's not the wet part of the pastes that transfer the heat energy, but the silver and silicates in the paste itself.

With the higher compression, on a cold plate, and the silicon nature of gpu dies, AS5 tends to kinda suck. Worse than mediocre. I'd use the included AlphaCool paste, it has finer silicate composition than AS5, so transfers the heat energy in a better pattern.

If you look at a cpu die, first that heat radiates through the tim, then the IHS, so a broader gap with larger particulate is fine. A gpu die is direct heat, no radiation, so by necessity every single transistor core (there's thousands of cores) must be covered perfectly. A larger particulate, with bigger gaps between, doesn't cover all the cores with anything except wet, allowing certain cores to see very minimal transfer and consequent higher Hotspot temps.

Outstanding explanation thank you!
 

Karadjgne

Titan
Ambassador
Oh. Forgot to mention. When you do repaste, irs not a cpu, it's a gpu. That means direct silicon die contact. It's vital to cover every square nanometer of surface, no blob or X technique. And don't worry about using too much, there's no socket, so when the paste squashes out over the edge, that's perfectly fine.

And don't forget to inspect the thermal pads, those should be in good condition, no tears or blank spots on the vram/VRM's.

Good luck.
 

adsoyo

Distinguished
Mar 26, 2012
68
4
18,565
Went ahead and spent 2 hours reapplying the thermal paste today. I used the Alphacool paste this time and the difference between that and the Arctic Silver 5 was easily noticeable. The best way I can describe it is the AS5 was like thick sticky peanut butter and the Alphacool was like pudding. Much easier to spread. Unfortunately the difference in temps isn't as drastic as I would've liked. Now instead of GPU/ Hot spot temps being 60/90C they're 55/80C.
 

Karadjgne

Titan
Ambassador
That's a massive difference. If you look at any paste comparisons, generally the difference between crap and top shelf is @ 5°C (not including liquid metals) given the same setup and methodology. So for temps given a normal working gpu with AS5 and the alphacool, I'd generally expect @ 2°C difference under exact same bench, at best. AS5 is pretty firmly smack dab in the middle overall.

And you got 5°C gpu, 10°C Hotspot ? That's huge. And a Heaven bench. Those temps are a very long way from the 75/105 initial.

I like a lot of EK's stuff, but not a fan of their rads, nor their gpu blocks. They used to be pretty top notch, but lately that's slipped with the commercialization. Smaller half blocks or full cover blocks that have nothing but a cold plate surrounded by clear acrylic for rgb usage, yeah looks decent but performance is bleh. Full coverage, solid copper Heatkiller4 for me.

With loops, have to do the homework, with a plan in mind. I wanted performance, and silence, so needed a rad to suit lower rpm fans. Most EK require higher speed fans, they don't perform well at lower sp/cfm. It's all got to match up. You can loose 10° on a loop just by using the wrong fans, at the wrong part of the curve, easily.

And that goes for the blocks too. Got to line up resistances to flow, in both blocks and tubing and rads and match that up with appropriate head pressure / volume on the pump. Too much flow and the fans don't get to work on the heat much before the coolant exits the rad, not enough flow and the first fan is doing the lions share, the next 2 fans are desperately trying to dissipate heat from coolant that's already been dissipated, so do next to nothing. That reduces a 360mm rad to not all that much better than a 120mm.

A well planned, thought out and executed loop is amazing, but can easily look hodgepodge. A fancy, rgb lighted showpiece with all the chalky paste coolants might look fantastic, but be worse for temps than a stock air gpu and an Aio on the cpu.

These guys do some amazing work on water cooling equipment, how-to's, dont-do's, reviews, bench tests etc
 
Last edited:

adsoyo

Distinguished
Mar 26, 2012
68
4
18,565
That's a massive difference. If you look at any paste comparisons, generally the difference between crap and top shelf is @ 5°C (not including liquid metals) given the same setup and methodology. So for temps given a normal working gpu with AS5 and the alphacool, I'd generally expect @ 2°C difference under exact same bench, at best. AS5 is pretty firmly smack dab in the middle overall.

And you got 5°C gpu, 10°C Hotspot ? That's huge. And a Heaven bench. Those temps are a very long way from the 75/105 initial.

I like a lot of EK's stuff, but not a fan of their rads, nor their gpu blocks. They used to be pretty top notch, but lately that's slipped with the commercialization. Smaller half blocks or full cover blocks that have nothing but a cold plate surrounded by clear acrylic for rgb usage, yeah looks decent but performance is bleh. Full coverage, solid copper Heatkiller4 for me.

With loops, have to do the homework, with a plan in mind. I wanted performance, and silence, so needed a rad to suit lower rpm fans. Most EK require higher speed fans, they don't perform well at lower sp/cfm. It's all got to match up. You can loose 10° on a loop just by using the wrong fans, at the wrong part of the curve, easily.

And that goes for the blocks too. Got to line up resistances to flow, in both blocks and tubing and rads and match that up with appropriate head pressure / volume on the pump. Too much flow and the fans don't get to work on the heat much before the coolant exits the rad, not enough flow and the first fan is doing the lions share, the next 2 fans are desperately trying to dissipate heat from coolant that's already been dissipated, so do next to nothing. That reduces a 360mm rad to not all that much better than a 120mm.

A well planned, thought out and executed loop is amazing, but can easily look hodgepodge. A fancy, rgb lighted showpiece with all the chalky paste coolants might look fantastic, but be worse for temps than a stock air gpu and an Aio on the cpu.

These guys do some amazing work on water cooling equipment, how-to's, dont-do's, reviews, bench tests etc

True they're significantly lower than 75/105 initial, I was just expecting to be closer to the battlestation's temps. I'm now realizing the GPU is actually right on the money at 55C vs 53C on the battlestation but the hot spot is 80C vs battlestation's 63C. Maybe the waterblock isn't perfectly flat or something. Either way 80C is still well below the thermal throttling temp of 110C so I'm just gonna send it. Thanks for the help.

Edit: I should also mention that game and boost clocks for this card are 2045 and 2285 MHz and on both computers Afterburner is showing ~2450 MHz for the duration of Heaven and 3D Mark so can't complain at all.
 
Last edited: