Noise/Performance Heatsink Curves

Noise pollution turns $80 heatsinks into $30 heatsinks — Joe

SUMMARY: Noise pollution turns $80 heatsinks into $30 heatsinks.

I’ve been getting a lot of mail about noisy heatsink fans; we did a Fan Survey which found a LOT of dissatisfied Delta 38 users, so dissatisfied that most of them trashed the Delta 38 for a slower, less noisy fan.

There is an excellent website called The Silent PC which has gobs of source material on PC noise. This quote from the European Environment Agency nicely summarizes the issue:

“Noise is a serious issue – levels over 40 Ldn dB(A) affect our well-being, while there is evidence that levels over 60 Ldn dB(A) can affect our physical and psychological health.”

Most folks who have been in the same room with a Delta 38 screamer can relate to this.

I’ve been thinking about how best to represent heatsink noise as part of our testing; serendipitously, Andy Lemont over at Millennium Thermal (the Glaciator guy) worked up graph that, I think, shows some very interesting data. This is Andy’s data, not mine, but the concept should hold up with other data.

What Andy has done is to compare fan noise (dBA) and heatsink performance (C/W) for a range of fan rpms:

Noise vs C/W

12 Volt RPMs: AX7 – 5992; Glaciator – 5694; SK6 – 6837

38 dBA line: AX7 – 3622; Glaciator – 5694; SK6 – 5300est.

The A1-A2 line is for Thermalright’s AX7 with 80mm fan, G1-G2 Glaciator II with 60mm, and S1-S2 Thermalright’s SK6 with 60mm. The end-points are for fan speeds run at 7 and 16 volts, with subsequent C/Ws and dBAs at each point. I also indicated the 12 volt point for each curve, which represents “as shipped” noise/performance levels.

The light blue bar represents 38 dBA – this is the “Generally Accepted” noise level for office PCs.

A few things pop out:

  • The slope of the lines look about the same. If it holds up (and I think it will), this means that for each heatsink, there is a fairly predictable, linear relationship between rpm (= airflow through the heatsink, or velocity) and performance.
  • Increasing Performance = Increasing Noise. This is a “DUH!” in my book, but seeing it visually drives the point home.
  • Twice the RPMs does NOT equal twice the performance. About twice the rpms gets you about 30% more in C/W (depicted by the slope of the lines). Anything beyond what’s depicted will increasingly show diminishing returns (the line goes vertical).
  • The “Tolerable Performance/Noise” level appears to be around 0.34 C/W. Heatsinks may be “Hitting a Wall” at this point. There is no such thing as a free lunch – if you want performance, you need surface area and velocity through the heatsink.
  • Air cooling solutions beyond a C/W of about 0.34 may not be practical; liquid cooling, once the exclusive province of the overclockers, may be the most practical solution (note the emergence of integrated watercooling cases).

  • Heatsink “detuning” for noise (by using lower rpm fans) effectively throws money out the window. If you buy an $80 heatsink and couple it with a 3000 rpm, 80mm fan to decrease noise, effectively you turn it into a $30 heatsink (like buying a Ferrari and never going over 55mph).

The compelling point of all this is “What Price Performance?” Some of our readers have no problem with Delta 38s (although their “significant others” usually do), but the heatsink performance derby is increasingly pushing into noise levels that are intolerable to the majority of consumers.

One of the more interesting things we found in the Fan Survey was that people exposed to annoyingly loud fans became “noise sensitized” – not only did they elect to throw out a Delta 38, but many thought even heatsinks that were tolerable before using a Delta 38 now were intolerable.

What I like about Andy’s chart is the “pick a noise level” approach – if noise is a concern, you can see where each heatsink lies on its noise/performance curve. This meets the concern I get from readers about selecting fans for heatsink.

I have to purchase some additional test gear to do a good job on this, and once acquired, I have a fair amount of retesting to do. What I’ll do as soon as I can is to begin reporting noise/performance data for a range of values as opposed to a single point, accumulating data as I test new heatsinks.

I wish I could do this overnight, but be patient and stay tuned.

Addendum: Graph Plot Points

Glaciator			AX7				SK6
46	7365	0.31		58	7285	0.26		51	8342	0.35
43	6724	0.32		57	6977	0.26		50	8085	0.35
41	6362	0.33		56	6655	0.26		49	7689	0.36
40	5988	0.34		54	6321	0.27		47	7270	0.37
38	5694	0.35		53	5992	0.28		46	6837	0.38
36	5327	0.35		51	5571	0.28		45	6354	0.38
33	4981	0.37		48	5130	0.29		42	5885	0.39
32	4594	0.37		46	4651	0.30		39	5215	0.40
30	4160	0.39		41	4202	0.32		37	4718	0.42
28	3622	0.42		38	3622	0.34		32	4027	0.45

Voltages range from 16 to 7 volts, descending order; C/Ws rounded to two places.

Email Joe

Be the first to comment

Leave a Reply