Gigabyte’s so-called “Server-Grade Thermal Conductive Gel” and a degrading thermal paste on the Radeon RX 9070 Gaming OC

Measurement according to ASTM 5470-17

This paste uses a formulation in which the filler content is already close to the maximum possible packing density, so that only a very small amount of carrier matrix (usually silicone oil) remains to ensure a homogeneous, processable paste. However, precisely this ratio appears to be problematic for the material analyzed. The linear regression curve of the ASTM-compliant measurement shows a very high thermal conductivity of 6.825 ± 0.050 W/mK with a simultaneously low thermal interface resistance of 3.2 ± 0.3 mm²K/W. This indicates an extremely tight particle bond, which is dominated by an almost optimal percolation of thermally conductive particles. The paste thus forms a quasi-continuous network of solid contacts, whereby the heat is largely transferred through conductive bridges and no longer via the poorly conductive binder.

However, this high performance can only be achieved because the formulation operates at the upper limit of what is still rheologically acceptable as a workable paste. This was exactly the problem with the rapidly degrading paste on the Asus cards (e.g. RTX 4080, see previous link). The microscopic images on the previous page confirmed an extremely dense particle distribution with few gaps, the silicone matrix is barely visible. Such a condition is associated with a number of typical problems that become particularly relevant in practical use.

Firstly, the mechanical integrity of the paste is significantly reduced under cyclic loading, as the remaining polymer network is too weak to permanently absorb the relative movement between the particles. This manifests itself in typical micro-cracks, as observed in vertically mounted heat sinks after just a few months in operation. The paste tends to form fine cracks and shrinkage zones in which the thermal coupling is greatly reduced locally. In addition, the low viscosity of the matrix components can lead to a slow loss of oil through migration or bleeding, which further impairs homogeneity.

Another problem with overfilled pastes is their pronounced thixotropy. While a certain degree of thixotropy is desirable to prevent “creep” under static load, excessive filling levels lead to limited flowability under pressure. This can lead to the paste no longer fully penetrating microscopic irregularities when applied or under minimal component warping. In combination with the often observed irregular particle sizes (4-17 µm), this results in local air pockets or particle-poor zones – with correspondingly high thermal resistance. At the same time, the risk of abrasive stress on sensitive structures increases, for example during repeated disassembly or through setting processes during operation.

At the end of the long laboratory day, it can therefore be concluded that the excellent thermal properties of this extremely highly filled thermal compound are the result of a very narrow particle-to-matrix ratio. Ideally, this ratio enables an almost percolation-like conductive structure, but is extremely sensitive to mechanical stress, thermal ageing and application errors. In practice, the high performance demonstrated by the measurement curve can only be used in the long term if the material is processed under strictly controlled conditions and, in particular, if the compressive load, surface planarity and operating temperatures are precisely coordinated. Otherwise, the unstable cohesive forces within the matrix and the low elasticity of the material quickly lead to degradation, pump-out and ultimately to a deterioration of the thermal coupling, as was unfortunately already evident in the first images.

Matrix analysis

The following image again shows my LIBS-supported material analysis (laser-induced breakdown spectroscopy) of the thermal paste used in cross-section with an optical image on a scale of around 100 µm. The quantitative element distribution reveals a composition that is characteristic of metal oxide-based high-performance pastes in the overfill range.

At 45.5 percent by weight, aluminum is the dominant filler. Due to the lack of specific oxide separation, the combination with 24.0% oxygen indicates a predominant presence of aluminum oxide (Al₂O₃). This compound is one of the most effective ceramic heat-conducting fillers, as it is chemically inert, electrically insulating and thermally conductive (around 30 W/mK in the crystalline alpha phase). The amount of aluminum in the element analysis also shows that a significant proportion of the particles are in the form of Al₂O₃ granules or platelets, not metallic aluminum.

Zinc is also clearly represented with 10.4 %. It is contained here in the form of zinc oxide (ZnO), which is supported by the simultaneous presence of oxygen. ZnO is often used as a supplementary filler to round off the particle spectrum in the smaller micrometre and nanoscale range and thus optimize the packing density. It acts as an intermediate filler between larger aluminum oxide particles. The high overall filling thus results from a multimodal distribution of the particle sizes, which was also evident in the microscopy images previously.

The 9.4 % carbon and 8.5 % silicon indicate a polysiloxane-based matrix, in which at least some of the carbon carriers may also come from additives or possibly amorphous filler modifications. Silicon is most likely bound to the binder phase as a component of a classic silicone polymer (PDMS or modified variants), which serves as a carrier fluid for the fillers. The only 2.2 % hydrogen is typical for polymeric silicone compounds and again confirms that the paste has a comparatively low matrix density.

The quantitative composition corresponds exactly to the previously made assumptions about the highly filled, almost percolating structure of the paste. The ratio of 45.5 % Al and 10.4 % Zn in combination with around one third oxygen suggests a structure in which the entire thermal conductivity is realized almost exclusively via solid contacts. This assumption is supported by the thermal conductivity of 6.825 W/mK and the linear progression of the ASTM curve.

The high filling, in particular with hard ceramic particles, also explains the mechanical anomalies in operation: the paste is abrasive towards soft surfaces (e.g. copper), it tends to become brittle under temperature changes and shows signs of drying at the (lighter) pressure points due to migration of the thin matrix phase. In conjunction with the previous microscopy images, it is clear that this is a “functionally optimized” but material-sensitive balanced paste, which may be suitable for industrial applications with controlled parameters, but which appears to have only limited long-term stability in the variable environment of a GPU. The fact that this is all at the expense of structural elasticity and application robustness has already been comprehensively demonstrated in the previous tests and images.