Thermal Paste Shootout - Q209 @ [H]

I would submit that a more dense cheese would probably have fared better, afterall american nearly melts at room temp. Perhaps something in an aged gruyere?
 
I'm glad I read this before building my computer! I saw the expensive prices on compounds and thought for sure that some old cheese I had laying around would do just as good. After reading this article it is clear that if I am to overclock I need to invest in something other than cheese
 
Last edited:
Something I'm not quite sure of and I've seen some debate. Perhaps you can enlighten us as to how you folks clean to re-apply paste. Some use alcohol in one form or another. But I've never really seen a definitive answer.
 
Huh, I've been using that MG thermal compound on non critical surfaces for years IE ramsinks n such, and AS5 for critical ones like between cpu and sink and also gpu and sink, with only a 3 C difference between the two I guess that MG does damned well. coulda saved myself 30-40 bucks over the last 5 years or so ;)
 
I have always used WD-40 applied to a q-tip to remove the bulk of the paste, and the strongest isopropyl alcohol I can find on the store shelf to remove the rest of the paste and any residue from the WD-40. Works like a charm every time. :D
 
I use alchohol to clean the CPUs surface prior to applying thermal compound. Kyle often writes in magic marker what CPUs are what when giving them to us for testing. They tend to go back to him blank. :cool:
 
Good info. I'm gonna be getting some AS5 soon, is it ok to use 70% alcohol wipes to clean the old thermal compound from the cpu and heatsink?
 
I would have also liked to see Zalman's paste in here, but it's cool.

As you mentioned, there wasn't a huge variance between the TIMs. Have you guys considered using a version of LINPACK, which stresses the CPU quite a bit more than P95 and thus might show a larger difference between TIMs?

:)

~Ibrahim~
 
I was trying to be funny...but...grap this one...Loved the Article...

Also, we wanted to share our data with the community. You can grap the Excel file from here in a zip. Feel free to use this data for your own graphs or articles, all we ask in return is that you give us a link in your article and please send it to us. We always like to see how others represent the data and what can be mined from it. Thanks!

I had another one, but I've lost it now...

Great job nonetheless.

Thanks for the extra eyes, I have fixed the fat fingering.
 
I use alchohol to clean the CPUs surface prior to applying thermal compound. Kyle often writes in magic marker what CPUs are what when giving them to us for testing. They tend to go back to him blank. :cool:

Yes, and I thank you for that when I don't know what CPU I have anymore. ;) That's why I only ever give you one spec!

Great review! Is there a similar test with the i7 920?

I don't see that happening any time soon and quite frankly I would not expect results to contradict these in any way.
 
Except for the fact that the slice of cheese would probably turn into melted nacho cheese faster than you can say "yo quiero Taco Bell"... :D
 
Great article. Something about using the cheese in this shoot-out and the BMG in the hard drive "shoot-out" that give off the old school [H] feel around here.

 
First benchmark reviews and now you guys all of sudden take notice of thermal pastes in the past month. You have given more concrete proof to your claims and I applaud your dedication to making these reviews as thorough as possible.

Next time maybe try doing a UPS shootout and since you can't use cheese try using two variations on a DIY ghetto fabulous UPS. I would love to see the difference in using DIY lead acid vs DIY lithium ion.

One thing that thoroughly confuses me about this is I wonder if Shinetsu is getting very different results for the X-23 7762 compared to their other products, which vastly outperformed this compound?


Now it's a pity you didn't inlude most [H]ard of all thermal pastes - Coollaboratory liquid pro - which consists of liquid metal (but i don't know if you can buy it in US)
http://www.tweakers4u.de/artikel/coollaboratory_liquid_pro_wlp,,12,2005,,458.htm

Yes it can be purchased in the US and now this article has made me curious about compounds people are mentioing here such as the karat and this one you mentioned.

Oh well t's not a big deal but if I was shopping around I would see no reason anyone would pay for the X-23 7762 if it holds up in other tests having similar results.
 
I would have used Velveeta instead. It's much closer to the consistency of thermal paste designed for higher temperatures.
 
I haven't read every post, so no idea if this has been mentioned yet, but an Australian site did a review similar to this. They even included some food as a thermal paste. Being Australians, they used vegemite though. :) Oh, and if I recall right, they used an Aussie brand of toothpaste too.

Here, if I may take the liberty of linking another site: http://www.dansdata.com/goop.htm

Looks like the toothpaste and vegemite held it's own according to that page I just linked too.
 
Honestly, I don't know many people who use a computer for longer than 4 years. :shrug:

If the system is being used for stuff like simple webbrowsing, paying bills, and etc. You would be surprised how low of specs you can get away with. Especially when your dealing with a aunt and uncle, grand ma, or etc. that simply does the "monkey do" after seeing how to work a computer.

Careful with that Ceramique if you use the thin layer method. It gets VERY sticky after awhile. I had it yank out a locked down Athlon 3800x2 out of the socket of an AsRock Socket A board.

I hate to be the BS detector, but the socket A mobo's peaked with the athlonXP's, when the athlon 64's hit 754, then when the dual core ahtlon64's came out they were originally for the 939 (or 940) sockets only.

I would have used Velveeta instead. It's much closer to the consistency of thermal paste designed for higher temperatures.

I am guessing Kyle chose to use american, because it doesn't really spoil or start stinking after being between a cpu and heat sink for a few days due to it's high content of hydrogenated fat.

I didn't say it in my previous post but, this is a great article that obviously took much time and effort to create the scientific data in it.
 
Good review, was hoping to see the Zalman thermal paste that you apply with a brush, kinda curious about the temps on it.

That's the stuff I use, too. After I used up my tube of the AS I bought, I got the Zalman with the brush (and cute square bottle)... At worst, I see maybe another 1 to 2 degrees Celsius increase in temp... but it's effectively the same with the added bonus of being much easier to clean up and reapply. The Zalman paste is pretty resistant to drying out, still maintains a fairly sticky consistency even after a year. Just redid the paste on my AMD64 X2 box about 2 months ago. Temps were the same before cleaning up the paste and after applying the new layer. Not nearly as scientific as the shootout here, but it's definitely worth using.
 
I was pretty sure that AS5 has a cure time of 200hours of high-temperature exposure to reach max performance.

XCPUs.com found the same exceptional performance out of the Shin-Etsu X23 as well. Too bad the stuff costs an ARM+LEG.


EDIT: and what is up w/ the CHEESE?
err.gif
 
Last edited:
I was pretty sure that AS5 has a cure time of 200hours of high-temperature exposure to reach max performance.

XCPUs.com found the same exceptional performance out of the Shin-Etsu X23 as well. Too bad the stuff costs and ARM+LEG.

Would you please point us to documentation on the 200 hour cure time for AS5?
 
http://www.arcticsilver.com/pdf/appinstruct/as5/ins_as5_singlecore_expsd.pdf
http://www.arcticsilver.com/pdf/appinstruct/as5/ins_as5_singlecore_wcap.pdf
http://www.arcticsilver.com/pdf/appinstruct/as5/ins_as5_amd_dual_wcap.pdf
http://www.arcticsilver.com/pdf/appinstruct/as5/Ins_as5_amd_quad_wcap.pdf
http://www.arcticsilver.com/pdf/appinstruct/as5/ins_as5_intel_dual_wcap.pdf
http://www.arcticsilver.com/pdf/appinstruct/as5/ins_as5_intel_quad_wcap.pdf

Section 3:
Due to the unique shape and sizes of the particles in Arctic Silver 5 conductive matrix, it will take a up to 200 hours and several thermal cycles to achieve maximum particle to particle thermal conduction and for the heatsink to CPU interface to reach maximum conductivity. (This period will be longer in a system without a fan on the heatsink or with a low speed fan on the heatsink.) On systems measuring actual internal core temperatures via the CPU's internal diode, the measured temperature will often drop 2C to 5C over this "break-in" period. This break-in will occur during the normal use of the computer as long as the computer is turned off from time to time and the interface is allowed to cool to room temperature. Once the break-in is complete, the computer can be left on if desired.
 
TheMead said:
If the system is being used for stuff like simple webbrowsing, paying bills, and etc. You would be surprised how low of specs you can get away with. Especially when your dealing with a aunt and uncle, grand ma, or etc. that simply does the "monkey do" after seeing how to work a computer.

I've got lots of systems that get "handed down" to grandparrents, aunt-in-laws, etc. However nearly all the ones I've handed down end up there because something went bad and it was cobbled together, i.e. a fresh installation of paste. Although, my wife's old machine has been put together for at least 4 years now I think and hasn't had a new set of paste and it still works fine so... :shrug:
 
I'm having trouble with this article's testing methods

The discrepancies between the programs and hardware monitoring is irrelevant if you accept that CoreTemp, for example, provides as consistent results as any other method. When the temperature goes up, the temperature goes up the same in any different program. You fail to mention that it's the change that matters, not the raw numbers.

By using a probe on the surface of the CPU, you are interfering with the test itself. Thermal paste belongs on the CPU, not a probe. How a thermal paste transfers heat away from the plastic on the probe, I think you would agree, is different from how thermal paste transfers heat away from the IHS.

The way you justify this testing method is by how Intel tests their own processors. But, you're not testing the processor.

The best way to do this test would be to create your own heating surface and thermal monitoring, beneath a genuine Intel IHS. Then you can create your own loads, and monitor temperatures, without introducing variations that you will invariably receive by having a whole computer there and interfering with the testing surface.

(One last edit, now that I've actually read the full article, rather than just testing methods, as per request of "vengence" and "brentpresley": how can you regulate applying a consistent layer of thermal paste? Did you measure the amount which was applied by drop so that drop sizes were the same? There's a lot of issues with human error introduced here, which is considered by many (as you note) to be one of the most important aspects of thermal compound.)
 
Last edited:
I'm having trouble with this article's testing methods

The discrepancies between the programs and hardware monitoring is irrelevant if you accept that CoreTemp, for example, provides as consistent results as any other method. When the temperature goes up, the temperature goes up the same in any different program. You fail to mention that it's the change that matters, not the raw numbers.

By using a probe on the surface of the CPU, you are interfering with the test itself. Thermal paste belongs on the CPU, not a probe.

The way you justify this testing method is by how Intel tests their own processors. But, you're not testing the processor.

The best way to do this test would be to create your own heating surface and thermal monitoring, beneath a genuine Intel IHS. Then you can create your own loads, and monitor temperatures, without introducing variations that you will invariably receive by having a whole computer there and interfering with the testing surface.

When you have a mechancial engineering degree, feel free to come back and explain to us why you are wrong. It's not worth my time.
 
:DDamn thought the cheese would win. Guess I'll just stick with melting it into my scrambled eggs.Seiously what was the point of the cheese. Its got fat in it which gets very hot when heated. You should have done a test without paste and with the intel stock thermal paste/pad. See if its worth buying the performance stuff over what comes with your CPU.
 
(One last edit, now that I've actually read the full article, rather than just testing methods, as per request of "vengence" and "brentpresley": how can you regulate applying a consistent layer of thermal paste? Did you measure the amount which was applied by drop so that drop sizes were the same? There's a lot of issues with human error introduced here, which is considered by many (as you note) to be one of the most important aspects of thermal compound.)

Because of the style of clips used, there is a constant total force, and thus pressure applied, this will insure a repeatable application if using a drop, regardless of total drop size as the resulting thickness is a function of pressure and viscosity both of which are now constants.

If you simply wanted to know the thermal conductivity of the objects you should just go run a ASTM D5470. Thermal conductivity is not what is being measured by this testing procedure.
 
I'm having trouble with this article's testing methods

The discrepancies between the programs and hardware monitoring is irrelevant if you accept that CoreTemp, for example, provides as consistent results as any other method. When the temperature goes up, the temperature goes up the same in any different program. You fail to mention that it's the change that matters, not the raw numbers.

By using a probe on the surface of the CPU, you are interfering with the test itself. Thermal paste belongs on the CPU, not a probe. How a thermal paste transfers heat away from the plastic on the probe, I think you would agree, is different from how thermal paste transfers heat away from the IHS.

The way you justify this testing method is by how Intel tests their own processors. But, you're not testing the processor.

The best way to do this test would be to create your own heating surface and thermal monitoring, beneath a genuine Intel IHS. Then you can create your own loads, and monitor temperatures, without introducing variations that you will invariably receive by having a whole computer there and interfering with the testing surface.

(One last edit, now that I've actually read the full article, rather than just testing methods, as per request of "vengence" and "brentpresley": how can you regulate applying a consistent layer of thermal paste? Did you measure the amount which was applied by drop so that drop sizes were the same? There's a lot of issues with human error introduced here, which is considered by many (as you note) to be one of the most important aspects of thermal compound.)

This article was a valid real world testing situation.
It did not require the stringent testing criteria of a university laboratory.
If you take it on it's face.........the variables were met as squarely as they could be under a well defined situation.
Within the boundaries of the real world, where we all live and work, this was very well done.
 
I'm having trouble with this article's testing methods

The discrepancies between the programs and hardware monitoring is irrelevant if you accept that CoreTemp, for example, provides as consistent results as any other method. When the temperature goes up, the temperature goes up the same in any different program. You fail to mention that it's the change that matters, not the raw numbers.

The temperature goes up but core temp and software DO NOT accurately measure temperature changes. The method used to extract temperatures from CPU's is a kludge and a prayer even if Core Temp bows at the alter of the DTS.

Intel defines a certain Tjunction temperature for the processor. In the case of Yonah it is 85C° or 100C°. First of all the program reads from a Model Specific Register (or MSR), and detects the Tjunction temperature. A different MSR contains the temperature data, this data is represented as Delta in C° between current temperature and Tjunction.

So the actual temperature is calculated like this 'Core Temp = Tjunction - Delta'

The size of the data field is 7 bits. This means a Delta of 0 - 127C° can be reported in theory. But from preliminary tests, the reported temperature doesn't go below 0C°, no matter what kind of cooling was used.

AMD chips report the temperature by a special register in the CPU's NB. Core Temp reads that register and uses a formula provided by AMD to calculate the current temperature.
The formula for the K8 is: 'Core Temp = Value - 49'.
The formula for the K10* is: 'CPU Temp** = Value / 8'.

The sensor in AMD CPUs can report temperatures between -49C and 206C.

Those built in sensors are simply not that accurate (again no matter what Core Temp claims about DTS) and the data is done by extrapolation not by a real measurement.
 
Because of the style of clips used, there is a constant total force, and thus pressure applied, this will insure a repeatable application if using a drop, regardless of total drop size as the resulting thickness is a function of pressure and viscosity both of which are now constants.

If you simply wanted to know the thermal conductivity of the objects you should just go run a ASTM D5470. Thermal conductivity is not what is being measured by this testing procedure.

I don't know why you're taking this so personally, I mean, clearly I don't have a Mechanical Engineering degree, it's not interesting to me. But, that doesn't mean we can't discuss this.

I understand the free-body-diagram/Hooke's "Law" idea you've got here, but let's take something more practical: Imagine you put only a nanogram of paste on in a drop, or imagine you put on the whole tube. You don't mean to tell me that the thickness of these applications is similar, do you? In a perfectly viscous situation, would not the thickness depend only on the amount of paste in the drop?

This might be more rigorous to satisfy you: What does this say of a "thin layer"? There's not regulation here. How much paste used, as is even "shown" in the flash video referenced, has a huge impact on thermal performance. In a scenario where the paste fills only irregularities up to the surface, then regardless of viscosity, the spring will extend until the IHS. Adding more thermal paste here will thicken the layer to a point because the resistance offered by the viscosity is greater than the force offered by the spring. Do you agree? It is at the equilibrium point that we hit the thickness of a drop, and any more paste is squeezed out the sides, as you correctly said.

This article was a valid real world testing situation.
It did not require the stringent testing criteria of a university laboratory.
If you take it on it's face.........the variables were met as squarely as they could be under a well defined situation.
Within the boundaries of the real world, where we all live and work, this was very well done.

The article does attempt this kind of strict testing. It may, indeed, be the purpose of this article that its testing is more technically correct than any other article. In this light, should not every measure have been taken to ensure greatest accuracy?

The temperature goes up but core temp and software DO NOT accurately measure temperature changes. The method used to extract temperatures from CPU's is a kludge and a prayer even if Core Temp bows at the alter of the DTS.
Would you be so kind next time as to provide data proving this point? You provided data on simply load temperatures, that doesn't back up the conclusion that hardware monitoring is better. I'm just saying that you should show me that the load changes in the programs are inconsistent with your hardware testing because that's how you draw your conclusion that hardware is better, you know? So, if CoreTemp goes up by 5C one time, and 6.3C the next, but your hardware goes up by 4C both times, then you've got something.

But, if CoreTemp changes consistently with your hardware, even if not precisely (that is to say, proportional changes in temperature should be acceptable if just to determine the order of performance not caring about how much one is better than the other), then it is a valid measurement, because you are giving us performance of products relative to each other, so the specific number doesn't matter. Surely you agree?
 
Last edited:
Surely you agree?

Surely I don't. The method used to report temperatures by Core Temp and other software is neither accurate or precise which negates your argument as you can't know what the value actually is or how much it really has changed and what was done to the value to come to a final number by the formula. That is why you ALWAYS measure physical and electrical properties with actual instruments that are appropriate for the property being quantified.
 
Surely I don't. The method used to report temperatures by Core Temp and other software is neither accurate or precise which negates your argument as you can't know what the value actually is or how much it really has changed and what was done to the value to come to a final number by the formula. That is why you ALWAYS measure physical and electrical properties with actual instruments that are appropriate for the property being quantified.

I didn't claim that they were precise at all, for sure. My claims deal specifically with a total loss of precision. But, you do have to PROVE that it is inaccurate.

Mainly why this part of the review bothers me is that you use this table of one-test load numbers and somehow magically conclude that hardware testing must be better. You don't prove that it is though, which is misleading. The specific table I'm referring to is this:

1236043439L0bdV188X7_1_7.gif

(from this, they are just different. this provides no evidence that sperry is superior to any other method of measurement)

You're telling me that you don't agree with the following statement: If CoreTemp provided proportionally congruent results to your hardware monitoring, then your results would have been in the same order.

From the article, I can tell you're a smart person, and that you know that they would have.
 
Back
Top