During testing with various X570 motherboards and the Ryzen 2 and Ryzen 3 models, I noticed that there are considerable differences between the water temperature and the die temperature (Delta), even though I have set a fixed power limit of 95 watts in the Precision Boost Overdrive (PBO). Counter-tests as a measurement at the EPS connector resulted in the same power consumption and also what the sensors read out for the package power was almost exactly at the same level. That to blame the higher thermal density for the 7 nm chipt being much smaller would be far too short.
If you take a look at the picture below, you can see that the hotspot of the 8-core with only one chipplet is located at the top right, i.e. exactly where the 7nm chip is located. The larger I/O die is located at least horizontally in the middle, but in terms of performance the chip with the larger structure width is not a big item anyway. So the heat density (density) of the 8-core is concentrated only on a very small part of the heatspreader. With the 12- or later 16-cores, the highest thermal density will concentrate on a narrow vertical corridor.
If we now look at the cold plate of a normal CPU water block (e.g. EK Velocity AMD), we can see that the microchannels cover the entire heatspreader, but the water flow is not optimal, because the in- and outtake are centered. Ideally, the intake should also take place at the top right and at the end, the other, much larger side remains the smaller one in terms of cooling. This would also mean that the installation direction of the water blocks is fixed and cannot be changed.
Also air coolers, especially the cheaper models in DHT-design (Direct-Heatpipe-Touch) should not work optimally anymore, especially not if only a few heatpipes are used, which are all centered. The photomontage below illustrates this:
I am pretty sure that for Ryzen a special design of the Cold Plate could mean a significant improvement and the manufacturers are called upon to take up this new challenge as well. This rather accidental discovery can only be the beginning.
And by the way: we have also worked on a simple solution and will certainly be able to present the result shortly. Maybe already next week 😉