The composition, shape, conductivity and quality of thermal pads greatly affect not only the performance of a graphics card, but also its durability or useful life. When choosing these components, the manufacturer must also take into account something as irreverent as the curvature of the GPU chip itself, the PCB or the heatsink, and it is something that they do not seem to take into account precisely.
Why are there temperature problems on the NVIDIA RTX?
NVIDIA’s larger Ampere architecture chips are pretty warped, and we’ve talked about that in the past. The concave curvature was already present in Turing, but not as extreme as now and also the total height of the package was much more uniform. To test this, if it is measured with a special device (which shows the report you see above) at 20º C the curvature of the matrix remains at 0.068 mm (according to NVIDIA, within the specifications), a value that is in the range tolerable but depends on model, because 0.08mm has been measured in some.
That can still be handled pretty well with thermal paste if it is not too runny, but there comes an extreme where the gap is too great and the thermal paste is simply not enough to bridge the gap and for heat transfer to proceed smoothly. efficient way.
At the same time, if you scan the dimensions accurately and take out a 3D graph you can see that in many places the curvature of the die is 0.08mm or even more, which certainly raises questions about how such a thing can happen. Often the entire board (including the die package) is already deformed at the factory and gaps of up to 0.156mm can be seen in some parts.
The package has certain stresses and height tolerances, so this is still completely normal. However, NVIDIA specifies a tolerance range of up to 0.3mm, which is very high compared to Turing.
The following evaluation of the height differences shows a simultaneous bulge and deviation from the actually measured heights of the 3D file that we mentioned earlier. You can also see the flex of the pack and the underlying board of the graphics card very well, and in this case, we see that even the circuit board had to bend significantly due to screwing in order to make contact.
So we see a deviation of about 0.3mm from the norm, which NVIDIA considers the limit in their RTX graphics but is still practicable for thermal paste according to them. These 0.3mm don’t seem like a lot at first, but in practice you will have big problems transmitting heat.
Thus, these deviations and tolerances lead you to have to resort to thermal pads to transmit heat, and this also explains why they are becoming thicker and softer. Thermal pads 1 mm thick are then converted into products of up to 2 mm, with greater thermal resistance and worse performance, but that at least “reach” from one place to another to cover the gap and be able to transmit heat.
All this leads us to the fact that manufacturers choose thermal pads much thicker than normal to be able to cover all these large gaps, but also softer than normal so as not to exert too much pressure in the less concave areas. And that is precisely where the temperature problems come in the NVIDIA RTX.
The problem affects all manufacturers
From the looks of it, these issues aren’t just from NVIDIA RTX benchmark GPUs, they affect all or almost all manufacturers. Basically, due to the curvature and imperfections of the NVIDIA chips they have been forced to use thicker and softer thermal pads than it should, causing cooling efficiency to suffer a lot. This coupled with the fact that, as we mentioned at the beginning, manufacturers tend to lower costs, leads to really serious problems and especially in VRAM chips, which suffer from extremely high temperatures.
In fact, the problems go further. In the images that we have placed above and below these lines we can see the carelessness with which the thermal pads are handled during assembly. In the image above, from an AORUS RTX 3080, we can see that the capacitors are putting extreme pressure on the thermal pad and that the thermal pad is in fact bent and the last capacitor is not making contact. In the image below, corresponding to a PowerColor GPU, we can see that they did not even remove the protective plastics (even though it is seen that they have written “Remove before use”, remove before use).
At this point, we can only ask manufacturers to regain a little traction and to care more about the quality of their products. It is not admissible to pay 1,000 euros for a graphics card and run into serious temperature problems that affect its consumption, noise and performance because the manufacturer has not bothered to remove the seals from the thermal pads.
Less RGB and more quality under the hood, please.