Thermal Compound Poll

[darkswordsman17]

Generally speaking, no. You want effective dissipation along as many axes as possible, even for thin interface layers (such as TIM/thermal pads). If they're going to use graphite, they may as well use CNTs or something similar that can conduct heat well along multiple axes. Synthetic diamond would also be an option (and it's actually rather cheap; so is industrial-grade diamond). When you have hotspotting, you definitely need lateral dissipation, to help heat spread into underutilized parts of the thermal interface.

That's my point though, if you reduce the layers, you reduce the potential for issues like that, so I would think would be beneficial for that issue even. Funnel heat from the die to the largest component of the heat exchange as directly as possible. Let the heatsink spread the heat out. So you find a way to make the TIM as thin but even (and consistently able to), so that the heat from the die is making as direct of contact with the main and largest heat dissipation aspect as possible. You can spread the heat out with it, you simplify install (which will also help with cooling, especially over time).

I think you might be focused on the sponge part of what I was saying? Let's just talk about one of these thermal sheets between the die and heatsink, wouldn't that provide better cooling than including the solder (or other TIM) and heatspreader? That's what I'm saying.

 

[DrMrLordX]

@darkswordsman17

In the case of thermal pads, they require a certain amount of thickness to meet the physical specifications of the interface. Sometimes they're used as TIM and spacers combined. There are limits to how thin you can make them. I do find it interesting that graphite has been chosen for some of these pads, and that they do apparently work. But diamond/CNTs would work better . . .

The only other thing I can add is that to fully appreciate the need for multiple axes of dissipation, it's a good idea to look at a numerical analysis of heat transfer in three dimensions. In order to carry out such anaylsis, you create nodes with six thermal connections to other nodes. You use various formulae to model what the temperature will be at each node based on thermal conductivity, known temperatures, etc. Even if you have a relatively unidirectional flow of heat from a heat source (CPU/GPU die) to a target (HSF), if you massively lower the k value on four out of those six nodal connections, you'll get reduced heat flow between source and target in any scenario where the source has uneven temperatures. From what Radeon VII has shown us, temp variation between parts of the die can be as great as 30C on TSMC 7nm for that particular heat source.

In extreme cases, graphite sheeting might have a W/mK as high as 1950 along one axis but as low as 5.7 along the other two which is a huge difference. As it stands, the graphite thermal pad on Radeon VII allegedly has an overall thermal conductivity equivalent to, say, Kryonaut or GC Extreme. Indium solder would obviously outperform it, and even Conductonaut and CLU are outperforming the graphite pad when used as a replacement (which is difficult to do on Radeon VII).