After testing over 100 different thermally conductive pastes, of which almost 80 are already available in my online database, I was able to gather extensive experience that calls into question some common assumptions in this area. Of particular interest was the analysis of contradictions between my generalized measurement data, the frequently stated general values for the thermal conductivity of the pastes and the results actually measured under practical conditions. These discrepancies occurred above all with very thin layer thicknesses of less than 100 µm, whereby the deviations were particularly clear in the range below 30 µm.
The thermal conductivity values given in many data sheets often suggest a linear transferability to practice, but this could not be confirmed in my tests. While certain pastes certainly exhibit high thermal conductivities in standardized laboratory measurements, they do not always achieve these values in real applications, especially with very thin layer thicknesses. This is partly due to the interaction between the surface properties of the components to be cooled and the specific viscosity and filler structure of the paste in question. Especially with layer thicknesses of less than 30 µm, the surface roughness and micro-irregularities of the contact surfaces have a considerable influence on the actual heat transfer, which is often not taken into account in the general data.
In today’s article, I will therefore discuss the most frequently occurring surfaces and their properties. I will show how different structures – from polished heatspreaders to coarser, machined surfaces – affect the effectiveness of the thermal compound. I will then turn to the tested pastes themselves, focusing on how to recognize whether a particular paste is sufficient for the intended use. Not only the nominal thermal conductivity values play a role here, but also aspects such as the processing properties, the flow behavior under pressure and the behavior at minimum layer thicknesses.

I will pay particular attention to the maximum layer thickness actually required. It has been shown that many users either overestimate or underestimate the thickness of the layer, which can lead to suboptimal results. In my measurements, I was able to clearly define for which applications thinner layers are sufficient and where problems could potentially occur – be it due to insufficient wetting, air inclusions or failure of the paste with extremely thin applications. These findings should help users to select the right paste for their specific requirements and at the same time optimize application techniques to achieve the best possible thermal performance. The database will be continuously updated with new findings and further products to provide a sound and practical basis for decision-making and most importantly: I have modified it once again in the evaluation and search criteria.
As a sorting criterion, I use the effective thermal conductivity as the mean value for the selected layer thickness, as it takes into account both the thermal resistance of the paste and the contact resistance of the two contact surfaces and therefore accurately reflects reality. I decided against sorting by temperature because the differences measured at very low layer thicknesses may no longer be clear in some cases due to inaccuracies and measurement tolerances. The determination of the effective thermal resistance and the effective thermal conductivity, on the other hand, includes six independent temperature values and the consideration of a gradient and is therefore many times more accurate.
The necessary changes to the thermal compound database
After extensive tests and detailed measurements, I have redefined and restructured the three search criteria and the evaluation of the thermal pastes. The aim of this adjustment is to better reflect the real performance, especially for very thin layers of less than 30 µm, as this area is the most relevant in practice. In previous databases and manufacturer specifications, this critical area was often not sufficiently taken into account, although it plays a decisive role in determining how efficiently a thermal compound actually works in real application scenarios. The results of my research will later clearly demonstrate why this area is so important.
The first group within my new categorization (“Smooth and even surface”) therefore focuses exclusively on performance with layer thicknesses below 30 µm. Here it is particularly clear which pastes still guarantee high thermal conductivity and reliable wetting even when applied extremely thinly. In this range, there are many contradictions between nominal thermal conductivity values and real measurements, as filler distribution, viscosity and surface adaptation of the paste play a decisive role. Most manufacturers do not provide any specific data for this area, although it is precisely here that the thermal efficiency of a cooling system is significantly influenced.
The second group (“Rough and uneven surface”) refers to rather rough, curved or uneven surfaces, as they occur in particular with certain CPU and GPU variants. These include CPUs in the LGA1700 socket, which place particular demands on the thermal compound due to their characteristic curvature, as well as very large monolithic GPU chips, whose surface structure also requires an adapted assessment. This segment measures how well a paste is able to ensure consistently effective heat transfer despite such unevenness, without unwanted air pockets or excessive thickening.
Finally, the third group (“gap filler”) forms an average performance value of the effective thermal conductivity across all possible layer thicknesses of a paste up to the upper limit of 400 µm. While most applications are in a significantly lower range, this group enables a holistic evaluation of paste performance, especially for scenarios where a thicker layer is required due to greater unevenness or mechanical compression. It serves as a complementary benchmark and allows a well-founded selection to be made even for unusual applications.
I present this revised tripartite division with a corresponding overview at the beginning of the following pages of my analysis so that I can refer to it directly in the subsequent investigations into surface properties. This restructuring not only ensures a more differentiated evaluation, but also makes it possible to work out contradictions between theoretical manufacturer specifications and practical measurement results. Particularly in the relevant area of thin layer thicknesses, this makes it clear which pastes actually deliver what they promise and which merely impress with high nominal thermal conductivity values on paper but do not deliver the expected performance in practice.
I must also mention a mistake that we took the opportunity to correct. Very viscous pastes, such as the Thermalright TFX, whose minimum BLT is 35 µm, are now displayed correctly in the diagrams and no longer appear in the 25 µm charts once the allocation problems have been resolved. This means that the classification and evaluation of this paste is correct again. I apologize for this.
Why I look at the curvature separately
CPUs and GPUs usually have less curvature when installed than when unmounted due to several mechanical and thermal factors. The most important reason is the contact pressure exerted on the processor by the CPU cooler and the bracket. Modern socket mechanisms, such as the LGA1851 for Intel or the AM5 socket for AMD, are designed in such a way that they ensure an even distribution of force on the CPU’s heat spreader. This contact pressure means that slight deformations, which are visible in the unmounted state, are compensated for by the mechanical tension. Intel’s LGA1700 socket was an inglorious exception, but there are workarounds and remedies for this.
Another factor is the stress distribution in the material of the heatspreader and the substrate. Internal stresses can arise during the manufacturing process, which are partially relieved after installation due to the pressure of the cooler and the fixation in the socket. The cooler not only exerts a vertical force, but can also cause a certain “planarization” through its installation by maximizing the contact surface between the heatspreader and the cooler. This leads to a reduction in curvature and better thermal coupling.
Heat development during operation also plays a role. As the CPU heats up, the various materials (such as copper, nickel and the silicon substrate) expand differently. This thermal expansion can help the heatspreader to adhere better to the cooler, especially if the materials are selected so that their expansion coefficients are matched. The operating temperature range of modern CPUs has generally already been taken into account during the development phase so that thermal deformation is minimized when installed.
Finally, the distribution of the thermal paste between the CPU and the cooler also influences the curvature. The paste fills microscopic irregularities and its viscosity can help to compensate for minor deformations. This not only improves heat transfer, but also contributes to a more even contact surface. The combination of mechanical contact pressure, stress distribution, thermal expansion and the use of thermal pastes therefore results in a reduction in curvature when installed, which is crucial for optimizing heat transfer between the CPU and cooler. If you consider the curvature factor to be important for your specific design, you should use the search and selection criterion “Rough and uneven surface” to be on the safe side. The rest simply ignore this factor.
Test setup, measurement methods and basics
Our database is based on real laboratory values that we have elaborately determined according to industry standards. However, many of these results contradict the manufacturers’ marketing claims and ruthlessly expose contradictions and lies, but they are all well-founded, reproducible and legally sound. These measurements not only reflect the general performance values of the pastes, but also enable an assessment of the suitability for a specific area of application (layer thicknesses, surfaces) as well as the suitability taking into account the individual capabilities of the respective user. In addition, the material analysis including digital microscopy is suitable for making your own assessment of the possible durability of a paste, even if I do not want to and cannot assume any guarantee for this. Unfortunately, measuring more than 3000 cycles per paste is not feasible. Statements about the matrix and the particles used, including their size, are also important. Please refer to my other articles and all individual tests on pastes that have shown certain abnormalities.
You are welcome to leave suggestions and comments in the forum, via PN or e-mail. If you would also like to contribute to the project and send me samples of thermal paste that have not yet been entered in the database, please contact me by e-mail. The e-mail address can be found in the imprint. Of course, this also applies to the manufacturers whose products we would like to test, regardless of which continent the product comes from. The scope of the database is de facto unlimited and as the methods and equipment are always the same, it can be expanded over the years and still remain comparable.
35 Antworten
Kommentar
Lade neue Kommentare
Urgestein
1
1
Urgestein
1
Urgestein
Mitglied
1
Urgestein
Urgestein
Mitglied
Urgestein
1
Mitglied
Mitglied
1
Mitglied
1
Mitglied
Alle Kommentare lesen unter igor´sLAB Community →