Paste tear-off
Now there is the video where I start up the test head again after the test. Pay particular attention to the “plop” after approx. 4 seconds, where the slightly caked surfaces audibly tear apart and off and the mechanics gloriously triumph over the holding forces of the paste.
The effective thermal conductivity
I have already written about the role of interface resistance in the basics. I am still leaving it in the calculation, as it is clearly more relevant in practice. And the same applies this time as in all my other reviews: if you already have Rth, you don’t actually need λeff, i.e. the effective thermal conductivity. But anyway, I’ll make the graphics anyway. Of course, we can now see again how the values change over the BLT, although you can no longer expect a linear curve here because of the included area and BLT.
Of course, the whole thing is also shown again as a bar chart for the most important layer thicknesses for a better comparison of the BLTs of interest:
Apart from the fact that I also have the temperatures of the heater and the water, which are of no use to us because they either adapt to the resistors or always remain constant, I have my measurement setup with the temperature sensors 1 to 6 (see diagram in the linked basic article). These values can also be used to make some very interesting considerations.
GPU simulation
Let us first take the values of T3 and T4, which show us the two temperatures at the respective contact surfaces between which the paste is located. These curves are no longer completely linear, as the interface resistance also changes slightly. And we no longer calculate with 6 points, but only with 2 absolute values for the temperature difference instead of a gradient as withTTim, whereby the sample temperature remains constant. And what is the point of all this? The behavior is similar to that of a graphics card, which has to manage without an IHS and where the delta is usually measured between the substrate and the water temperature. This can be projected quite well, because I test the temperature difference on the two surfaces between which the paste is located.
CPU simulation
Now I compare T3 of each of the tested products. If we normalize the values for the heater, we already have sufficient thermal resistance in the copper reference block to simulate the CPU temperature and its differences with different pads in comparison with each other and in relation to the thickness of the paste replacement. It is precisely this variable evaluation that no test on a CPU can offer, because the respective CPUs are bent differently and it is therefore not really reproducible. But the TIMA5 test does, because I can measure all distances, which is simply not possible on a single CPU.
TIMA5 control result for the bulk values
Let’s now compare the bulk thermal conductivity and the calculated interface resistances. The original from 2023 and the 3.43 W/m-K for the bulk thermal conductivity are not high-end, but they are certainly ok. I also removed the pinch at around 12 µm minimum BLT from the regression line because it deviates considerably
The poorer paste from 2024 is significantly more fluid and only achieves 2.92 W/m-K. However, we can see that there is already a performance loss of almost 17.5 percent in the bulk thermal conductivity compared to the paste from 2023, which cannot be explained by anything other than an influence (from whatever or whoever):
But what does the better paste from 2024 do now? Interestingly, the minimum BLT is once again fully in line with the regression line, which at least suggests that the mixture is more homogeneous. The bulk values of the paste from 2023 are not quite achieved, but this small difference will not be noticed in everyday life and can therefore be considered to be within tolerance with a clear conscience.
50 Antworten
Kommentar
Lade neue Kommentare
Veteran
Urgestein
Urgestein
Urgestein
Urgestein
Urgestein
1
Urgestein
Urgestein
Urgestein
Urgestein
Urgestein
Urgestein
1
Urgestein
Urgestein
Urgestein
Veteran
Urgestein
Alle Kommentare lesen unter igor´sLAB Community →