Increasing surface area of this interface does
not facilitate that, if the cross section remains the same. Deviations from flat cannot increase the total conductive cross section. Deviations from smooth might result in more contact with the thermal paste, at the cost of less metal on metal contact and larger voids for the paste to fill.
Having as little excess material as possible, in as small a void as possible, does. This is why improving flatness and smoothness work.
The trend in the electronics packaging industry is towards smaller, more powerful devices. With these small, high power components, however, there are also higher heat fluxes. As a result, engineers must find ways to minimize the thermal resistance from the electronic device junction to ambient air.
www.boydcorp.com
Lapping the CPU in a heatsink test platform is probably a controversial move that’s bound to provoke reactions. Funny thing is, it was done a year ago, and photos of the CPU showing the copper top exposed by the lapping have been featured in many of our reviews. Yet, not a single comment. This...
silentpcreview.com
A 7C lower delta allows for 25-50MHz more across all cores or lets me shave almost 15w off the CPU's power consumption during peak loads at the same clock speed (both because of lower resistance from simply being cooler and from the lower voltage requirement for stability). The former is quite modest, but performance is performance, and every tweak adds up. The latter is important because I'd like to save as much power budget as possible for the GPU.
It also allows me to meet performance targets with a smaller cooler than I would otherwise need, which is a big help within the space constraints of this SFF case.
A benchmark is a diagnostic tool. It shows performance is actually being gained, or that there are problems that need to be addressed.
That's less of a concern in my use, and increased current and voltage at the same temperature (where I usually end up burning any temperature gains) will actually result in a reduction in lifespan, all other things being equal. I kill plenty of processors, but they usually last as long as I need them to.
That said, if the goal was to maximize longevity, a 7C reduction is not insignificant either. Every 10C or so roughly doubles the lifespan of ICs (see the Arrhenius equation and failure mechanisms like electromigration) and components like capacitors. Smaller temperature swings also means less mechanical stress from thermal cycling, so if current and voltage remain the same, and longevity is a concern, lowering peak temperatures is a great way to improve that.
