EVGA's Double-Stacked Thermal Pads Tested: RTX 2060 KO Thermals
video description
Date: 2020-05-06
Related videos
Comments and reviews: 10
ajm8127
I want to see numbers without any pads. We use CA glue to attach thermocouples to the part package for development testing. I am sure you could try that here if you removed the pads. There is going to be very little heat transfer through the edge of the fins. The only reasonable interface I see is to the heat pipes. The chips are going to reject a fair amount of heat through the BGA balls and into the board. If moving air from the fans can get down to the board, it might cool more effectively than the 4-5 mm pad touching the very edge of the fins. No air is getting to the chips or between them when the pads are in place. That PCB is going to have multiple plane layers for power and ground. A lot of heat will flow through the board itself. So I am thinking cooling the RAM by cooling the board with forced convection might be more effective than relying on conduction through the part body and those thick pads to move the heat into the heatsink fins through the tiny end cross section of the fins. In the end this is not a product that is meant to be optimized, it is meant to hit a price target. I was in the market for a 2060 and purchased a founder's edition card for 299 specifically because I thought it had better cooling.
reply
I want to see numbers without any pads. We use CA glue to attach thermocouples to the part package for development testing. I am sure you could try that here if you removed the pads. There is going to be very little heat transfer through the edge of the fins. The only reasonable interface I see is to the heat pipes. The chips are going to reject a fair amount of heat through the BGA balls and into the board. If moving air from the fans can get down to the board, it might cool more effectively than the 4-5 mm pad touching the very edge of the fins. No air is getting to the chips or between them when the pads are in place. That PCB is going to have multiple plane layers for power and ground. A lot of heat will flow through the board itself. So I am thinking cooling the RAM by cooling the board with forced convection might be more effective than relying on conduction through the part body and those thick pads to move the heat into the heatsink fins through the tiny end cross section of the fins. In the end this is not a product that is meant to be optimized, it is meant to hit a price target. I was in the market for a 2060 and purchased a founder's edition card for 299 specifically because I thought it had better cooling.
reply
Anonymous
I recently purchased this card and took it apart myself as well. The memory chips on the bottom, as well as the top and bottom ones on the side, only have one 4mm thermal pad which then makes direct contact with the heat pipes. It's the two middle chips on the side that get screwed by having the double stacked pads and contact only with the heat sink fins. To be fair the double stack pad area happens to hit on the heat sink in an area where there is some flat parts (can be seen on the pads in the video at 6: 17, which I think are there to reinforce the heat sink's rigidity. Also, Steve kind of touched on this in the video, but both sets of memory chips also have thermal pads on the back side of the board that make contact with the metal (yes metal) back plate. So I'm guessing some cooling takes place there as well, even though it does have to travel through the board first. I tossed some better thermal compounds on the die. I also put separate small thermal pads (lol) on the VRMs to extend their contact up to the heat sink as well. Then I closed it up. Thermals seem to be okay for me.
reply
I recently purchased this card and took it apart myself as well. The memory chips on the bottom, as well as the top and bottom ones on the side, only have one 4mm thermal pad which then makes direct contact with the heat pipes. It's the two middle chips on the side that get screwed by having the double stacked pads and contact only with the heat sink fins. To be fair the double stack pad area happens to hit on the heat sink in an area where there is some flat parts (can be seen on the pads in the video at 6: 17, which I think are there to reinforce the heat sink's rigidity. Also, Steve kind of touched on this in the video, but both sets of memory chips also have thermal pads on the back side of the board that make contact with the metal (yes metal) back plate. So I'm guessing some cooling takes place there as well, even though it does have to travel through the board first. I tossed some better thermal compounds on the die. I also put separate small thermal pads (lol) on the VRMs to extend their contact up to the heat sink as well. Then I closed it up. Thermals seem to be okay for me.
reply
DOOTNOOT
hey, 2 thoughts GN. 1. With such thick thermal pads that have such small surface area thats actually contacting the heatsink fins; would it at all be more effective to remove the thermal pads all together and allow the airflow to just hit the VRAM for cooling? 2. since they're so much clearance between the cooler and the VRAM, couldn't you just remove the thermal pads and apply your own small heatsinks with some very thin adhesive pads? even a big flat plate (shim) of copper for the air to come though and hit would do the trick pretty well. with 5-9mm of clearance i don't see why you couldn't squeeze in say, a small RPI heatsink or even a long flat heatpipe with thermal pads on top to get way more surface area connection to the fins.
reply
hey, 2 thoughts GN. 1. With such thick thermal pads that have such small surface area thats actually contacting the heatsink fins; would it at all be more effective to remove the thermal pads all together and allow the airflow to just hit the VRAM for cooling? 2. since they're so much clearance between the cooler and the VRAM, couldn't you just remove the thermal pads and apply your own small heatsinks with some very thin adhesive pads? even a big flat plate (shim) of copper for the air to come though and hit would do the trick pretty well. with 5-9mm of clearance i don't see why you couldn't squeeze in say, a small RPI heatsink or even a long flat heatpipe with thermal pads on top to get way more surface area connection to the fins.
reply
Blahorga
To the best of my knowledge this kind of thermal pad is much worse at transferring heat than say aluminium. It's been many years since I last looked into it but back then it was about ten times difference or more. It's been ages ago so I don't remember the particulars other than it made me decide to really try to avoid thik TIM pads in whatever it was we were making at the time. So what would happen if you were to machine a aluminium or even copper block just slightly thinner than the gap the TIM pads make up for, use paste between the memory and the block and a thin high quality TIM pad between the block and the finstack? It would be a lot of work for results of dubious value other than to satisfy some curiosity.
reply
To the best of my knowledge this kind of thermal pad is much worse at transferring heat than say aluminium. It's been many years since I last looked into it but back then it was about ten times difference or more. It's been ages ago so I don't remember the particulars other than it made me decide to really try to avoid thik TIM pads in whatever it was we were making at the time. So what would happen if you were to machine a aluminium or even copper block just slightly thinner than the gap the TIM pads make up for, use paste between the memory and the block and a thin high quality TIM pad between the block and the finstack? It would be a lot of work for results of dubious value other than to satisfy some curiosity.
reply
Benzo
sub-105 degree specification. So GDDR6 modules are officially safe up to 105 degrees? Will GDDR6 memory modules degrade at lower temperatures, such as in the 90's? I have a Radeon RX 5700 and, according to either HWiNFO64 or MSI Afterburner, have my memory go above 90 when using 4k Super Virtual Resolution. I suppose I should also ask if GDDR6 memory temperature is indeed precisely the number reported as 'GPU Memory Junction Temperature' in HWiNFO64 and reported as 'Memory temperature (MEM)' in MSI Afterburner? I think for AMD cards, this number is the highest temperature among all of the memory modules. I don't know if this is different for Nvidia cards.
reply
sub-105 degree specification. So GDDR6 modules are officially safe up to 105 degrees? Will GDDR6 memory modules degrade at lower temperatures, such as in the 90's? I have a Radeon RX 5700 and, according to either HWiNFO64 or MSI Afterburner, have my memory go above 90 when using 4k Super Virtual Resolution. I suppose I should also ask if GDDR6 memory temperature is indeed precisely the number reported as 'GPU Memory Junction Temperature' in HWiNFO64 and reported as 'Memory temperature (MEM)' in MSI Afterburner? I think for AMD cards, this number is the highest temperature among all of the memory modules. I don't know if this is different for Nvidia cards.
reply
Scott
You guys should do a test (if only just for fun) with an AMD card using the same thermocouple arrangement (without the double stacked thermal pads, of course, or as close as you can to it. Then you may be able to get a close, albeit still not 100% accurate delta between the known readings in software that AMD allows, and the thermocouple reading. Just a thought, could be a good experiment and then you d have a general idea of what the delta is for the Nvidia cards. You could try a few cards to see if the delta stays the same or within a degree or two.
reply
You guys should do a test (if only just for fun) with an AMD card using the same thermocouple arrangement (without the double stacked thermal pads, of course, or as close as you can to it. Then you may be able to get a close, albeit still not 100% accurate delta between the known readings in software that AMD allows, and the thermocouple reading. Just a thought, could be a good experiment and then you d have a general idea of what the delta is for the Nvidia cards. You could try a few cards to see if the delta stays the same or within a degree or two.
reply
Siana
Given such ample space below the cooler, why aren't there just tiny little chip heatsinks stuck down with thermal tape? There seems to be enough airflow down there that it should work better than one of those ridiculous pads, and i'm not even sure that it would be more expensive. Would you like to test little heatsinks instead? Or some other solution? Or even whether the chewing gum is better than letting the chips cool bare naked on the component side of the PCB?
reply
Given such ample space below the cooler, why aren't there just tiny little chip heatsinks stuck down with thermal tape? There seems to be enough airflow down there that it should work better than one of those ridiculous pads, and i'm not even sure that it would be more expensive. Would you like to test little heatsinks instead? Or some other solution? Or even whether the chewing gum is better than letting the chips cool bare naked on the component side of the PCB?
reply
Cezar
Why don't they fix the SLI/NVlink stuff once and for all? You have two cards you use one RTX 2080ti for ray tracing, physX, hairFX and AA and the main RTX 2080ti to run the game usual thing textures, shaders, shadows blah blah. How hard can it be for a good team of chip architects to create this stuff? Their chips are powerfull enough, nvlink has enough bandwidth, PCIe 4. 0 has enough bandwidth, Nvidia please get to bussin and stop using that voodoofx 1999 method
reply
Why don't they fix the SLI/NVlink stuff once and for all? You have two cards you use one RTX 2080ti for ray tracing, physX, hairFX and AA and the main RTX 2080ti to run the game usual thing textures, shaders, shadows blah blah. How hard can it be for a good team of chip architects to create this stuff? Their chips are powerfull enough, nvlink has enough bandwidth, PCIe 4. 0 has enough bandwidth, Nvidia please get to bussin and stop using that voodoofx 1999 method
reply
Kellin
I just picked up one of these for a new build I am doing for a friend. I used the Phanteks shift air case and while over thermals look good, when this card ramps up around 65c it gets loud and stays loud. When I got to cool it down it never ramps back down even at around 32c. This makes the whole system loud and unenjoyable to use. I contacted evga and may end up RMAing the card. It's a bummer for sure as the card seems fine otherwise.
reply
I just picked up one of these for a new build I am doing for a friend. I used the Phanteks shift air case and while over thermals look good, when this card ramps up around 65c it gets loud and stays loud. When I got to cool it down it never ramps back down even at around 32c. This makes the whole system loud and unenjoyable to use. I contacted evga and may end up RMAing the card. It's a bummer for sure as the card seems fine otherwise.
reply
mojpismonosa
Could you please compare this GPU vs plain 2060 in deep learning applications? Say darknet/PyTorch/TensorFlow/Keras FPS in double/single/half precision training and inference? You could just do classification using some familiar backbones (darknet53 from AlexeyAB that supports half precision, some ResNets and maybe some NLP-suited models, LSTMs etc. The RTX GPUs are huge in this professional space.
reply
Could you please compare this GPU vs plain 2060 in deep learning applications? Say darknet/PyTorch/TensorFlow/Keras FPS in double/single/half precision training and inference? You could just do classification using some familiar backbones (darknet53 from AlexeyAB that supports half precision, some ResNets and maybe some NLP-suited models, LSTMs etc. The RTX GPUs are huge in this professional space.
reply
Add a review, comment
Other channel videos
