Having never owned an SFP cable, I excitedly bought one of these and a pair of ConnectX-3 40gbe cards. The performance is sweet but as someone that lives/works next to his equipment the sound is not. This unit powers up ~silent but uses some sort of coarse fan-power-adjustment control loop to regulate its internals. The key tell is on the Mikrotik wiki:
Who likes voiding warrantees? I do!
One of the user comments on the original review piece was to consider upgrading the fans to Noctua NF-A4x20 FLX and I agree, however once installing them it was clear that they weren’t sufficient. They ran basically flat out (5.2krpm) the entire time to just barely keep the “cpu-temperature” at the desired 58C setpoint while sitting on my nicely air-conditioned desk. While still relatively quiet, I was worried that once I started adding connections they’d be overwhelmed.
I presume that “cpu-temperature” is actually the temperature of the switch chip under the biggest heatsink. The air has to be pulled by the fans in the back clear from the front of the unit, gently whispering by this clearly thermally limited part. I considered fashioning an angled bracket to give that air a boost when I noticed the clearance from the top of the heatsink to the case was ~13mm – just enough to sneak in a 10mm fan! I fashioned a thin mounting bracket out of a drink can and with a dab of hot glue attached a NF-A4x10 FLX to the top of the “cpu” heatsink. I plugged it into the open “Fan4” socket and to my delight the tachometer readout even showed up as an added row in the RouterOS Health screen.
The whole system now bops between a functionally-silent 3krpm and a very-quiet 4krpm. Watching the progression of RPMs I may have added too much gain to their control system and might in the future wire in a speed reducer to the “cpu” fan. If you have a 4pin fan available it might be worth trying that to see if it just works, as a linear PWM control would be more pleasant than the coarse control I currently have.
I hope this is useful to anyone else. If you wanted to do this professionally Id also consider replacing the entire heatsink with a true heatsink-fan for better mechanical stability, but for my lab a little bit of hot glue was fast and effective. I fear removing whatever thermal epoxy is between the heatsink and the switch chip might damage the bga.
I see exactly that behavior: after a pleasant ~20 minutes of relative quiet, the fans alternate between a few minutes of “pretty annoying” and a few minutes of “maybe that 8x1gbps tplink switch wasn’t so bad”.
Who likes voiding warrantees? I do!
One of the user comments on the original review piece was to consider upgrading the fans to Noctua NF-A4x20 FLX and I agree, however once installing them it was clear that they weren’t sufficient. They ran basically flat out (5.2krpm) the entire time to just barely keep the “cpu-temperature” at the desired 58C setpoint while sitting on my nicely air-conditioned desk. While still relatively quiet, I was worried that once I started adding connections they’d be overwhelmed.
I presume that “cpu-temperature” is actually the temperature of the switch chip under the biggest heatsink. The air has to be pulled by the fans in the back clear from the front of the unit, gently whispering by this clearly thermally limited part. I considered fashioning an angled bracket to give that air a boost when I noticed the clearance from the top of the heatsink to the case was ~13mm – just enough to sneak in a 10mm fan! I fashioned a thin mounting bracket out of a drink can and with a dab of hot glue attached a NF-A4x10 FLX to the top of the “cpu” heatsink. I plugged it into the open “Fan4” socket and to my delight the tachometer readout even showed up as an added row in the RouterOS Health screen.
The whole system now bops between a functionally-silent 3krpm and a very-quiet 4krpm. Watching the progression of RPMs I may have added too much gain to their control system and might in the future wire in a speed reducer to the “cpu” fan. If you have a 4pin fan available it might be worth trying that to see if it just works, as a linear PWM control would be more pleasant than the coarse control I currently have.
I hope this is useful to anyone else. If you wanted to do this professionally Id also consider replacing the entire heatsink with a true heatsink-fan for better mechanical stability, but for my lab a little bit of hot glue was fast and effective. I fear removing whatever thermal epoxy is between the heatsink and the switch chip might damage the bga.
