EVGA Addresses GeForce GTX 1080 FTW PWM Temperature Problems

[InvalidError]

If, as you say, efficiently heat transfers from the chip to the PCB, then what ? Since the chip is exposed to air with no backplate any additional cooling is limited to the minute surface area of those VRMs where they are exposed on the back of the board.

The weakest link when putting a heatsink on top of a plastic package is the plastic package which adds a few degrees per watt.

The surface area of the chips themlselves is of no importance: in a properly done power PCB layout, the surface-mount power chips' thermal pad is soldered to a power or ground plane (copper) which spans a large board area, often across two or more layers, making the chips' effective heatsinking area is much larger than the chips themselves. On a 6-8 layers PCB like the GTX1080, there should be more than enough copper under the FETs to handle the 2-4W per chip.

Take a GTX1080FTW and try comparing it in original factory state without the pads on the FETs, with pads on the FETs and one last time, by removing the MMCP altogether. If EVGA did its PCB layout reasonably well, the results will be much closer to using thermal pads than the original factory setup without them.

Here's one example of how to heatsink high power surface-mount FETs properly:
https://youtu.be/pKX50E_14MQ?t=2741

(Of course, that's quite a bit more than 3-4W per FET.)

 

[bit_user]

the surface-mount power chips' thermal pad is soldered to a power or ground plane (copper) which spans a large board area, often across two or more layers, making the chips' effective heatsinking area is much larger than the chips themselves. On a 6-8 layers PCB like the GTX1080, there should be more than enough copper under the FETs to handle the 2-4W per chip.

But you're talking about a fairly thin sheet of metal that's sandwiched between plastic. I can follow your argument about more efficient heat transfer into the power or ground plane, but then I don't see how you propose that the heat be efficiently removed from it.

 

[InvalidError]

I can follow your argument about more efficient heat transfer into the power or ground plane, but then I don't see how you propose that the heat be efficiently removed from it.

The copper planes may be thin but for low power applications like switching FETs which are only intended to dissipate a few watts each at best, that much copper is enough to fan the heat out. In a high power application, most of the heat would move vertically through the PCB through thermal vias instead of fiberglass. There isn't much vertical movement involved either as the prepreg fiberglass layers on either side of the core PCB are very thin - the core board provides the bulk of mechanical stability and thickness, prepreg layers glued on both sides of that core layer provide additional power/ground planes and signal layers.

How thin can prepreg layers be? As little as 0.0254mm or 0.001", though 0.01" is more common for everyday PCBs since it is much cheaper (easier) to work with. You won't get much thermal resistance between layers on one side of the core board. If you want to move heat from one side of a board to the other, you drill thermal vias, add generous copper plating, optionally fill them with solder, and slap a heatsink on the back if the application is sufficiently high power to justify the extra steps.

Cooling-wise, sufficient cool air flow across a the PCB surface is all that you should need for ~3W per FET. If the PCB cannot handle that, it needs a better layout/stackup.