Not sure if anyone else posted this but JayzTwoCents posted a quick mod that drastically reduced the fan noise on one of the CRS312 switches.

The fans used were the Noctua NF-A4x20, at US$15 (retail) each these would easily get mikrotik into silent home use. I don't care to be honest, I am perfectly fine with doing such a mod myself but if mikrotik even charged the extra $60 for the better fans (I'm sure large orders would get it cheaper than that) allot of people would jump at it. As Jay mentions this is one of the few good 10G switches available on the market at the moment and would get Mikrotik into high-end consumer networks (who end up being the points of recommendation for the average consumer).

The video:
https://www.youtube.com/watch?v=hhJvcRSLpiE

I haven't seen the fans in CRS312 but they look like Delta type fans, if so then I'd be careful changing them to any other fans as they might not have enough airflow to keep the unit cool! Interesting to see what temp the unit reports before and after the mod!

Bear in mind, Delta fans have 4 times more airflow than Noctua ones!

After finding the video myself, it raises a question, is the board capable to take advantage of PWM fans in controlling them?
As the original fans were not PWM and he replaced them with PWM ones.

I checked with a multi-meter. It controls the fans with voltage, regardless if they are 3 pin DC or 4 pin PWM. If the fans haven't maxed out and CPU is within limits IMHO there's no reason to worry about temps. Router would just ramp up the speed with increased temperature. But seeing how the Noctuas have top speed of 5000 RPM i'd keep an eye on that. Not sure they wouldn't max out and overheat under load as most 40mm fans have 10k RPM or more.

Also i I think it was in this forum, where one user put a 3D printed wind tunnel over CPU. Reported temps were good, fans got slower, BUT i suspect everything else besides the CPU got hotter. So i wouldn't modify airflow like that. Anyone remember, have link?

Since my warranty had already expired I've turned to more destructive approaches. Armed with a voltmeter, I spent my holidays poking around fans in quest for quiet. It turns out the bugger in red circle is connected to fans. It has the same voltage on it as all the fans. But most interestingly, it has voltage appearing and disappearing on it as the fans turn on and off! BEFORE YOU CONTINUE READING, I MUST WARN - what I did will definitely void warranty, can possibly break switch or worse ( other switches, cause fire ). Proceed at your own risk.

For the sake of experiment i cut the wire showing fan on status. Turns out it makes the fans permanently on. They still change speed as before, but instead of turning off when switch cools they just stay at 3500 RPM with fans getting uninterrupted 6V (previously they dropped to 0V): I'll leave it like this for a while to see if it fixes the dreaded on-off cycle, but i'll keep an eye on it to avoid any mishaps with stalled fans, temperature spikes, crashes or other potential problems.

@hyppen
Thank you for taking your time to look into this.
I am also looking for a way to disable this stupid zero rpm behavior, especially when https://mikrotik.com/product/crs312_4c_8xg_rm states MTBF Approximately 200'000 hours at 25C. (no clue if that refers to ambient or CPU actual temperature)
Can you please explain what

For the sake of experiment i cut the wire showing fan on status. Turns out it makes the fans permanently on.

means?

Are you referring to the sense (3rd) wire on the fan itself?

No, the one going in the middle of chip. It measures 3.3V when fan is on and 0V when fan is off. When i cut the wire going to chip ( irreversible change ! ) the fans always stay on and the pin on chip always measure approx. 4 V. I assume that is safe since the chip receives 12 V but a more robust confirmation would be preferred. Or connecting it to 3.3V trough resistor to be safe. I assume it turns the power on and off.

Can't the middle pin be desoldered and raised a bit instead of cutting ?

Trying to lift such a fat leg will break chip that small. IMHO if you want to avoid permanents cuts, it's better to remove the chip altogether, cover wire with isolation and put the chip back. That requires more skill and equipment. I felt more confident just cutting the wire and patching it with wire and iron in case of problems.

Just opened my CRS213 to to try to do something about that pin and...I got fooled by the zoomed in image from above.
The actual chip and pins are way smaller. Way too small for my skills, soldering iron and no magnifying glass.
Guess I am at the mercy of MT support to fix this crappy zero rpm sometime during my lifetime.