I've been surfing on the net (who doesnt) and seeing some weird and scientifically impossible claims about thermal pastes.
Claims like: 5 degree load differences, new load temperature being lesser then de old idle temperature etc.
I've used 3-4 brands of thermal materials and the biggest temperature difference I've ever seen is 0 degrees. (bu the way I am using a water block)
I just dont get it.
Arctic silver claims "<0.0045°C-in2/Watt (0.001 inch layer)" thermal resistance. For a 100watt cpu (which is a overkill assumption). The difference a zero thermal resistance paste to a well applied arctic silver is at most 0.45C.
What are your experiences? Any one ever seen more then 1C difference?
Thanks in return.
What's up with thermal pastes?
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
I've seen a 2-3 degree increase switching from Shinetsu to Zalman's bundled generic goo.
But, it may be just the application.
I don't think I saw more than 1C difference between Servisol, Anabond, and Nanofusion with my Ninja and undervolted A64.
Perhaps the difference will be greater with a hotter processor.
In your example... you believe entirely what Arctic Cooling claims, with all their fan noise measurement and such? And there's a lot of things to say about the example... I just dunno where to start...
Regarding your situation, if you're using a powerful w/c system and a cool processor such that the temperature difference between CPU and atmosphere is like <10 degrees or even <5, no way you can see any significant temperature change. Coz in w/c the water soaks up a lot of heat first then sends it to the radiator, the difference in temperature between waterblock and CPU is going to be small. Combined with the previous factor,
if the difference is less than 2 degrees how to expect a 1C difference by changing thermal paste?
But, it may be just the application.
I don't think I saw more than 1C difference between Servisol, Anabond, and Nanofusion with my Ninja and undervolted A64.
Perhaps the difference will be greater with a hotter processor.
In your example... you believe entirely what Arctic Cooling claims, with all their fan noise measurement and such? And there's a lot of things to say about the example... I just dunno where to start...
Regarding your situation, if you're using a powerful w/c system and a cool processor such that the temperature difference between CPU and atmosphere is like <10 degrees or even <5, no way you can see any significant temperature change. Coz in w/c the water soaks up a lot of heat first then sends it to the radiator, the difference in temperature between waterblock and CPU is going to be small. Combined with the previous factor,
if the difference is less than 2 degrees how to expect a 1C difference by changing thermal paste?
I think this article was on the front page just a while ago.
http://www.madshrimps.be/?action=getart ... rticID=635
Personally I haven't noticed a difference, but then again I haven't actually paid attention either. The differencies on the madhshrimps-test are pretty noticeable, so I do think that atleast with boxed cooler/other aircooler those differencies should always exist.
http://www.madshrimps.be/?action=getart ... rticID=635
Personally I haven't noticed a difference, but then again I haven't actually paid attention either. The differencies on the madhshrimps-test are pretty noticeable, so I do think that atleast with boxed cooler/other aircooler those differencies should always exist.
Re: What's up with thermal pastes?
Customary to refer to Dan's Data whenever thermal interface material is mentioned. Go toothpaste 
Jipa, the article at madshrimps, is one of the articles I was mentioning that popped questions.
I've no idea how they managed to achieve such huge improvements. It does not make any sense. Even the application method caused too much difference.
wwenze, regarding your comment:
Think of a thermal interface as a series resistor. Each component brings its own resistance. Heat flow can be tought as amperage where the temp difference can be tought as a voltage difference. A steady heat flow will always occurs. It's a matter of how much watt you are pouring to the system.
Lets say we have a 100 watt processor. And lets say you have a processor temperature of 55C with air temp at 25C. So you have a 30C temp difference. That creates a total thermal resistance of 30/100=0.3C/W.
Arctic silver claims a thermal resistance of <0.0045°C-in2/Watt that is pretty insignificant regarding the total thermal resistance of 0.3C/W. Nearly 1/66 times of the total resistance.
I've no idea how they managed to achieve such huge improvements. It does not make any sense. Even the application method caused too much difference.
wwenze, regarding your comment:
Think of a thermal interface as a series resistor. Each component brings its own resistance. Heat flow can be tought as amperage where the temp difference can be tought as a voltage difference. A steady heat flow will always occurs. It's a matter of how much watt you are pouring to the system.
Lets say we have a 100 watt processor. And lets say you have a processor temperature of 55C with air temp at 25C. So you have a 30C temp difference. That creates a total thermal resistance of 30/100=0.3C/W.
Arctic silver claims a thermal resistance of <0.0045°C-in2/Watt that is pretty insignificant regarding the total thermal resistance of 0.3C/W. Nearly 1/66 times of the total resistance.
-
Bluefront
- *Lifetime Patron*
- Posts: 5316
- Joined: Sat Jan 18, 2003 2:19 pm
- Location: St Louis (county) Missouri USA
I've never seen any measurable difference with any kind of paste, but I usually lap the base of the heatsink, and the CPU if it has an IHS.
The absolute ideal heat transfer is between two flat metal surfaces touching each other. This TIM is only necessary to fill in any minute air gaps between the two surfaces.
Some of the TIM materials dry up quicker than others....that's about it.
The absolute ideal heat transfer is between two flat metal surfaces touching each other. This TIM is only necessary to fill in any minute air gaps between the two surfaces.
Some of the TIM materials dry up quicker than others....that's about it.
No Sendorm, you can't look at it like that, temperature difference comes about as a result of thermal resistance, not the other way round.
And Arctic Siler's claim of <0.0045°C-in2/Watt (0.001 inch layer) is not insignificant, well not as insignificant as you may think. That value, you see, has the unit in2/Watt (0.001 inch layer). (yea I know that's a lot of factors to put in, it just shows how complicated this heat issue is)
For one, actual useful surface area is definitely less than an inch^2, because the die underneath the IHS is small, and the IHS was proven to be a non-perfect conductor (removal yields better temperatures), so heat travelling sideways across the IHS has to overcome further resistance, making those parts not as important (and probably the reason why "bird-dropping" TIM application works, if you've been reading the other thread).
Secondly, look at the bracket words. How thin can a human actually spread the paste?
If considering the above, it wouldn't be surprising if the actual resistance ends up higher than 0.01C/W. That'd be 1 degree for 100W lol. Still too low for us to see and smaller effect than a bad application.
So for a 2 degree difference (with a 100W processor), that'd be 0.02C/W and so on.
So now the question is, why did I say that with water cooling, the effect of thermal paste thermal resistance is reduced.
100W here doesn't really mean the output power of the CPU, to be precise it means the rate of energy transfer from the CPU to the heatsink/waterblock.
For HSF, the transfer is CPU -> heatsink -> air
Air blows at heatsink constantly to keep temperature down. Temperature difference causes energy to flow from hotter to colder. You're right thinking it's p.d., and p.d. induces current.
For w/c, the transfer is CPU -> waterblock -> water
and basically stops there, because the next step involves water flowing through pipework, which is relatively slow. And w/c works differently from air in that the water acts as a buffer for the heat, since it has good thermal capacity.
So water flows in with a temperature not too far from the CPU and waterblock.
Same thermal resistance at CPU -> waterblock. But because now, heat has been constantly soaked from CPU to the water, temperature of CPU is low, less difference results in less heat flow, hence smaller effect of the resistance of the contact.
Reverse applies also - if temperature difference between CPU and the heatsink is BIG, then a good TIM really plays a role. E.g. phase-change cooling, where temperature of CPU can reach >0 degrees in extreme overclocking cases while on the other side, it's freezing.
Think of something like, temperature increases because heat generated < heat conducted away, but as temperature increases heat conducted away increases, until a point at which heat generated = heat conducted away and temperature stabilizes. That's how thermal resistance can be measured @ change in temperature per power dissipation. And this is the reason why a stock HSF can dissipate as much heat as a Ultra-120, albeit at different temperatures. (same CPU, same power dissipation, isn't it?)
Or you can think in a simpler way, if the temperature on the other side is high and there's higher thermal resistance, the CPU needs to be at a higher temperature to wack the heat through.
This is the little bit that I know. To be honest I haven't quite mastered how it all works yea, so any correction is welcomed.
And Arctic Siler's claim of <0.0045°C-in2/Watt (0.001 inch layer) is not insignificant, well not as insignificant as you may think. That value, you see, has the unit in2/Watt (0.001 inch layer). (yea I know that's a lot of factors to put in, it just shows how complicated this heat issue is)
For one, actual useful surface area is definitely less than an inch^2, because the die underneath the IHS is small, and the IHS was proven to be a non-perfect conductor (removal yields better temperatures), so heat travelling sideways across the IHS has to overcome further resistance, making those parts not as important (and probably the reason why "bird-dropping" TIM application works, if you've been reading the other thread).
Secondly, look at the bracket words. How thin can a human actually spread the paste?
If considering the above, it wouldn't be surprising if the actual resistance ends up higher than 0.01C/W. That'd be 1 degree for 100W lol. Still too low for us to see and smaller effect than a bad application.
So for a 2 degree difference (with a 100W processor), that'd be 0.02C/W and so on.
So now the question is, why did I say that with water cooling, the effect of thermal paste thermal resistance is reduced.
100W here doesn't really mean the output power of the CPU, to be precise it means the rate of energy transfer from the CPU to the heatsink/waterblock.
For HSF, the transfer is CPU -> heatsink -> air
Air blows at heatsink constantly to keep temperature down. Temperature difference causes energy to flow from hotter to colder. You're right thinking it's p.d., and p.d. induces current.
For w/c, the transfer is CPU -> waterblock -> water
and basically stops there, because the next step involves water flowing through pipework, which is relatively slow. And w/c works differently from air in that the water acts as a buffer for the heat, since it has good thermal capacity.
So water flows in with a temperature not too far from the CPU and waterblock.
Same thermal resistance at CPU -> waterblock. But because now, heat has been constantly soaked from CPU to the water, temperature of CPU is low, less difference results in less heat flow, hence smaller effect of the resistance of the contact.
Reverse applies also - if temperature difference between CPU and the heatsink is BIG, then a good TIM really plays a role. E.g. phase-change cooling, where temperature of CPU can reach >0 degrees in extreme overclocking cases while on the other side, it's freezing.
Think of something like, temperature increases because heat generated < heat conducted away, but as temperature increases heat conducted away increases, until a point at which heat generated = heat conducted away and temperature stabilizes. That's how thermal resistance can be measured @ change in temperature per power dissipation. And this is the reason why a stock HSF can dissipate as much heat as a Ultra-120, albeit at different temperatures. (same CPU, same power dissipation, isn't it?)
Or you can think in a simpler way, if the temperature on the other side is high and there's higher thermal resistance, the CPU needs to be at a higher temperature to wack the heat through.
This is the little bit that I know. To be honest I haven't quite mastered how it all works yea, so any correction is welcomed.
@wwenze, Maybe we are both saying the same thing, but I always look at the matter as heat flowing through interface and the resistance initiates the temperature difference.
Heat transfer from any matter to air is directly proportional to the square of the temperature difference between to interfaces. A 100watt cpu must give 100watt exactly to the surrounding medium (ıf this wasnt true the cpu would have a constant increase in temp), the heat flow (amperes) will always be same. In a stable system, no matter what type of cooling method you are using the heat flow will always be same. So the thermal paste induced temp difference will always be same.
As you've mentioned, even with 0.01C/W thermal interface resistivity, a 100 watt cpu will only generate a 1C difference because of the thermal material used.
I have a e4300 cpu which uses about 30-40 watt so the 0 C(or lets say less then 1C) difference I see between using different thermal materials is logical.
Thus comes the question : how come the reviewer guys are getting 5-10C differences? There is definitely something wrong there.
Heat transfer from any matter to air is directly proportional to the square of the temperature difference between to interfaces. A 100watt cpu must give 100watt exactly to the surrounding medium (ıf this wasnt true the cpu would have a constant increase in temp), the heat flow (amperes) will always be same. In a stable system, no matter what type of cooling method you are using the heat flow will always be same. So the thermal paste induced temp difference will always be same.
As you've mentioned, even with 0.01C/W thermal interface resistivity, a 100 watt cpu will only generate a 1C difference because of the thermal material used.
I have a e4300 cpu which uses about 30-40 watt so the 0 C(or lets say less then 1C) difference I see between using different thermal materials is logical.
Thus comes the question : how come the reviewer guys are getting 5-10C differences? There is definitely something wrong there.
Regarding the resistance of TIM, continuing from my previous post,
the 0.01 value, of course, is just an optimistic guess. Lets guess how bad the variables need to be for e.g. 0.05C/W which will result in 5 degrees difference.
The die of a Brisbane is ~0.19 square inch. If we use that as the useful heat transfer area (and it really helps to assume an IHS-less scenerio here...), then it's <0.024°C/Watt (0.001 inch layer).
0.001 inch... that's 25.4 micron... no human can apply so thin.
But if we have 0.002 inch, then the resistance is already <0.048°C/W.
Now, what happens if the layer is thicker?
If the layer actually ends up 0.005 inch/0.13mm (sounds possible to me), a whopping 0.12°C/W, enough to result in 12°C difference with the 100W CPU.
the 0.01 value, of course, is just an optimistic guess. Lets guess how bad the variables need to be for e.g. 0.05C/W which will result in 5 degrees difference.
The die of a Brisbane is ~0.19 square inch. If we use that as the useful heat transfer area (and it really helps to assume an IHS-less scenerio here...), then it's <0.024°C/Watt (0.001 inch layer).
0.001 inch... that's 25.4 micron... no human can apply so thin.
But if we have 0.002 inch, then the resistance is already <0.048°C/W.
Now, what happens if the layer is thicker?
If the layer actually ends up 0.005 inch/0.13mm (sounds possible to me), a whopping 0.12°C/W, enough to result in 12°C difference with the 100W CPU.