JetsonHacks

The Heat Sink Chronicles – Salvation

Enzotech CNB-R1 be gone! Bring in the big guns and be done with heat sink fiddlin’.

After installing an Enzotech CNB-R1 onto a NVIDIA Jetson TK1, the results were a little disappointing. That made me start to wonder, how does the stock fan get better results? While it is a noisy little whiner, it does get the job done. I removed the thermal paste to take a peak to see what was underneath. Here’s the video:

The results surprised me, I assumed that there would be some type of metal underneath the fan to help with the thermal transfer. There was just the plastic of the bottom side of the fan there and a big dollop of thermal paste. Wow, really?

Since my main objection to the stock fan is the fan whine, I decided to order an
Enzotech SLF-40(mm) Forged copper 1100 Fan & Heatsink. One of the issues that I had with the CNB-R1 is that the hole to hole mounting distance was 59mm, while the Jetson board is drilled for 61mm. Plastic pins are used and can be forced to make the fit, but it puts undo stress on the board.

The SLF-40, on the other hand, is adjustable from 59mm to 63mm so it should fit ok. The fan runs at 6300 RPM and moves 5.4 CFM of air. Unfortunately I don’t know how that compares to the stock fan, but my guess is that it is equivalent. The addition of the copper heat sink seemed like a plus. Also, the SLF-40 is about the same height as the stock fan, the CNB-R1 is quite a bit taller. With the copper, the SLF-40 probably weighs a little more than the all plastic stock fan. Here’s the installation video:

The results? The pins that hold the heat sink in place are a little bit fatter than the pins on the stock fan, and need some “persuasion” to seat. The thermal paste is less viscous than the paste provided with the CNB-R1, so it was much easier to spread.

The fan seems to my ear to be much less annoying, with a lower pitch to the SLF-40. The annoying whine is gone, and is certainly something I can live with. In addition, after running the CUDA Smoke demo for 20 minutes, the SLF-40 equipped board runs 5 or 6°C cooler than the stock fan and tops out around 47°C past 20 minutes. The CUDA Smoke demo doesn’t run the system flat out, but it does stretch its legs.

Now it turns out that there are things called “math” and “physics” which folks can use to select appropriate cooling devices. Seems like a lot of mumbo-jumbo to me, it seems much better to just randomly order a variety of parts and force them onto a board and act surprised and perplexed when they don’t work.

Executive Summary

I’m definitely in the “if it ain’t broke, don’t fix it” camp when it comes to things like heat sinks and fans. This just shows you how annoying I found the fan whine of the stock setup. Up front, if the noise doesn’t bother you (you probably wouldn’t notice if you’re in a lab) this isn’t worth doing. However, if you are like me with it close by in a quiet room while you’re writing code and it’s annoying you, the SLF-40 is a good choice.

There are many alternatives, but this one works for me.