We just got our MikroTik CRS804 switch, and attached 4x sparks - the fans are LOUD and spinning at max speed. The sparks are idle!
Should we tweak the fan curve? Or just replace them with 2x Noctua NF-A4x20 PWM fans?
Does anyone have similar issues? Thanks!
Thanks for the response!
This is from Claude:
I went down exactly that path (update first, then look at temps). Unfortunately it didn't resolve it, so let me post a full write-up in case it helps others, because I think this is a fan-control bug rather than a real thermal event.
Hardware: CRS804-4DDQ-hRM RouterOS: upgraded to 7.23.2 (stable) ā and the RouterBOOT firmware (/system/routerboard upgrade) to 7.23.2. No change in fan behavior.
Symptom: both chassis fans ramp to ~15,800 RPM (max) while the switch silicon is cold. Critically, the fan speed inversely tracks temperature ā the fans spin up as the chip cools. It oscillates in bursts every few minutes (loud ā calmer ā loud).
Representative /system/health print during a loud burst:
cpu-temperature 50C
switch-temperature 56C
phy-temperature 34C
board-temperature2 50C
fan1-speed 15780
fan2-speed 15960
fan-state ok
So the silicon is ~50 Ā°C below the 105 Ā°C overtemp threshold, yet the fans are pinned at max. (The PSU fans sit at ~4,500 RPM and aren't the problem.)
Fan controls available (/system/health/settings):
fan-full-speed-temp: 65C <- hard-capped; can't raise it:
"value of fan-full-speed-temp out of range (-273..65)"
fan-target-temp: 58C
fan-min-speed-percent: 12%
fan-control-interval: 30s
Everything I tried, and the result:
fan-target-temp=65 (the max, since fan-full-speed-temp is capped at 65): quiets for a few minutes, then the runaway returns.fan-min-speed-percent ā this is instructive:
5% made it worse (bigger swings, slammed to max harder)40% held a steady ~6,300 RPM for ~5 min, then burst to max anyway12% (default) ā bursts return within minutes/system reboot: clears it for ~5 minutes, then it reasserts.It is not thermal. I confirmed by removing all nearby heat-generating gear from on/around the unit ā chip stayed cool, fans stayed at max. The controller drives the fans to full regardless of the (cold) reported sensors, and RouterOS exposes no fixed-PWM or max-speed override to force otherwise. It behaves like a broken control loop ā a limit cycle that keeps integrating to 100%.
Questions for the thread / MikroTik:
fan-full-speed-temp capped at 65 Ā°C? On these Marvell-based switches the silicon is fine well above that, so the tiny 58ā65 band gives the controller almost no room and seems to encourage this oscillation.Check if those QSFP56 modules report temperatures (that info will probably be under /interface ... not under /system/health) and what they are. AFAIK fan speed is affected by modules (SFP/SFP+/QSFP/...) temperatures as well, these need to be kept at some decent working temperature as well.
It is known that ROS has problems with many types of modules (wrong readings, wrong autonegotiation, etc.) so I wouldn't be surprised if ROS reads erroneous temperatures off your modules and consequentially spins fans (too) fast.
Good thought, but checked and ruled out: all four modules are passive Amphenol DACs (power-class 1) and expose no DDM data at all ā /interface/ethernet/monitor shows no sfp-temperature field, so there's nothing for ROS to misread. I've also logged the fan controller against /system/health at 10 s resolution: every fan event tracks switch-temperature exactly. With all four QSFP ports disabled the switch chip still needs ~4,650 RPM to stay under 65 Ā°C ā the heat source is the 98DX7335 itself. Idle draw is ~41 W at 4Ć100G (28 W bare per STH's review). What I'd love to compare: your switch-temperature and fan RPM at idle, and whether your Sparks linked at 100G or 200G.
Please open a support ticket at https://help.mikrotik.com so we can assist you further.
When creating the ticket, please attach the supout.rif file
