Comparison

General Information

Manufacturer

Alphacool

Polartherm

Designation

Apex

Manufacturer Specifications

Accessories

Spatula

Spatula, Applicator

Container

Tube, bag

Box

Notes and Recommendations

CPU

GPU

Difficulty

Conclusion

Medium performing paste with better durability, based on the original DOWSIL TC-5026

Favourable entry-level paste at an attractive price and easy to use. Rather unsuitable for graphics cards and not as durable as the X10 (same base, but lower fill level).

Measurements

Thermal Conductivity (W/m·K)

2.9

3.1

+6.5 %

Min BLT

Interface Resistance

8.2

5.2

Heat Conducting Particles and Matrix

Al2O3, ZnO, Silicone

Particle Size

<= 10 µm

<=12 µm

Minimum Possible Layer Thickness

That's exactly why I wanted to see how far we could go with a bit of pressure and how much a paste can still be compressed. Here, I use the usual 60 psi (41 N) on the measurement surface of 1 cm², which is more than sufficient and higher than what, for instance, a GPU cooler would achieve.

Thermal Resistances R_th

Let's start with the most important aspect, the thermal resistance R_th. The key feature of R_th is that it correlates linearly with the layer thickness, while thermal conductivity follows a different curve and is far from linear. But experienced readers already know this. We are mainly interested in layer thicknesses of 200 µm or less for CPUs, and usually 100 µm or much thinner for GPUs, depending on bending.

I have now prepared a bar chart comparing the relevant layer thicknesses from 50 to 400 µm for R_th.

Effective Thermal Conductivity and Cooling Simulation

As is always the case in my other reviews: once you have R_th, λ_eff (effective thermal conductivity) is not really needed. We can also see how the values change across BLT (Bond Line Thickness), though we can't expect a linear curve anymore due to the included area and BLT.

I have now prepared a bar chart comparing the relevant layer thicknesses from 50 to 400 µm for λ_eff.

GPU Simulation

Let's start with the values for T3 and T4, which indicate the temperatures at the respective contact surfaces between which the paste sits. These curves are no longer entirely linear because the interface resistance changes slightly. And we’re not working with six points anymore but rather two absolute values for the temperature difference, instead of a gradient as in T_Tim, with the sample temperature remaining constant. And why do we need all this? The behavior is quite similar to a graphics card that operates without an IHS (Integrated Heat Spreader), where the temperature delta is measured between the substrate and the water temperature. This can be projected fairly well, as I measure the temperature difference at the two surfaces where the paste is located.

CPU Simulation

Now, I compare T3 values for each of the tested products. When the heater values are normalized, we already have sufficient thermal resistance in the copper reference block to simulate CPU temperatures and their differences with various pads in comparison to one another, depending on the layer thickness as a substitute for paste. This type of variable assessment cannot be achieved in a CPU test because individual CPUs bend differently, making it non-reproducible. But in the TIMA5 test, I can measure all distances, something that simply isn't possible on a single CPU.