I've been playing with IBM's Watson Analytics, running Frostytech's test data through the wee cloud to see what it can predict, centered around these five criteria:

1. Heatsink dimension height
2. Manufacturer
3. dBA high (noise at high fan speed)
4. (150W R/A h) 150W Delta-T results at stock fan speed
5. (200W R/A h) 200W Delta-T results at stock fan speed

Up first, the correlation between Maximum fan RPM speed and noise output. It should be a no brainer that the faster the fan rotates, the higher the noise output. IBM's Watson Analytics reports the following:

Next up, the correlation between Maximum fan RPM and noise output with respect to averaged (150W R/A h) 150W Delta-T results at stock fan speed. IBM's Watson Analytics finds a sweet spot in the 2000-2200RPM fan speed range where noise output is between 35-37.6dBA and temperature results from the 150W thermal test platform hovered around 16.5 C over ambient.

Watson Analytics is a bit tricky to configure, seemingly going off in its own direction at times. Here, it finds a relationship between the (150W R/A h) 150W Delta-T results at stock fan speed and heatsink width. The sweet spot is apparently 120-126mm.

Below, we have the number of fans per heatsink running along the x-axis. On the y-axis, rise over ambient temperature for the 150W thermal test platform, for heatsinks operating at their minimum fan speed. Heatsinks with fans that operate at a fixed speed, would be excluded from this graph. The darker the dot, the louder the noise output.

At first glance it seems like two-fan heatsinks definitely have a thermal advantage, given the tight grouping between the 10C and 28C deltaT, while single fan heatsinks are much more spread out. Each of the 384 dots represents an individual heatsink. (Ignore the data formatting anomalies running along the zero x/y axis.)

Watson Analytics allows us to narrow the result set further, say to heatsinks producing more than 40 dBA noise (blue), leaving the remaining gray dots representing truly quiet coolers.

A question we're often asked, is which heatsink is best for a given height restriction. Here are the results, first for heatsink running at stock speed on the 200W thermal test platform, then for the 150W thermal test platform.

Note the line of water coolers along the vertical zero y-axis, as height is not recorded for these heatsinks. Going by the results, it seems the best performing heatsinks on the 200W thermal test platform have been between 145mm and 165mm tall. That lone dot at the intersection of 15.5 deltaT and ~160mm is the NH-D15.

Now let's see the relationship between heatsink height and the 150W thermal test platform. The IBM Watson Analytics cloud tool neatly allows us to filter the roughly 384 results. Down low at -48.2C is the Prometia Mach II GT :-). Again, water coolers line the vertical zero y-axis as height is not recorded for these heatsinks.

Neat tool.