This is why I was curious if he could measure it. I have the tools to do so but got a flush potted package that’s all flat, so no difference here to check.
If software temps are to be believed the HBM2 gets plenty warm, and the real effects can be seen in performance as the timings increase under throttling. Since the memory is on the package with the GPU I suspect it absorbs at least some of that heat as well which may help explain issues in this regard.
Now that my 56 is under water I can find out for myself how it helps, or if it does at all.