ipmi fanctrl: are my fans broken?

[asmodeus]

Hello,

I'm using fanctrl.pl from this forum to control all fans including CPU on a SuperMicro X11SSL-CF. With my new built, one model seems to be giving me trouble: the NF-P14s redux 1500 PWM. I'm running two of them with the board set to full mode

I've used Noctua's spec sheet to come up with thresholds:

ipmitool sensor thresh "FAN" lower 200 300 400
ipmitool sensor thresh "FAN" upper 1700 1800 1900

These however did not work. Even at 100% the fans would go into non-recoverable and setting a duty cycle did not have an effect.
I tried to massage them a bit but I cannot get them to exhibit consistent behaviour.

These settings work a bit better, but still have the problem of the fans going into 'nc' sometimes:

ipmitool sensor thresh "FAN4" upper 1500 1600 1700.

Also, as soon as I set a duty cycle to e.g. 37%, the fans stop spinning (confirmed by looking into the case), are then listed as not recoverable and only spin up again after a power cycle. I've tried increasing/decreasing the lower thresholds, e.g. to

ipmitool sensor thresh "FAN4" lower 0 100 200

and

ipmitool sensor thresh "FAN4" lower 400 500 600

but neither helps with the fan just stopping completely.

Another mystery is the CPU fan. I have the exact same stock intel fan in the same motherboard, where it's reporting speeds of 2000-2200 max and running fine with these settings:

ipmitool sensor thresh "FANA" upper 2000 2100 2200

On this motherboard, the CPU fan reports 3100/3200 RPM and also goes into nc sometimes. So I'm beginning to think something must be off with this unit? Any ideas what might be going on?

Edit: after a BMC reset, sometimes one of the fans completely disappears from ipmitool sensor list:

FAN4 | na | | na | na | na | na | na | na | na

 

[Kevin Horton]

Also, as soon as I set a duty cycle to e.g. 37%, the fans stop spinning (confirmed by looking into the case), are then listed as not recoverable and only spin up again after a power cycle. I've tried increasing/decreasing the lower thresholds, e.g. to

ipmitool sensor thresh "FAN4" lower 0 100 200

and

ipmitool sensor thresh "FAN4" lower 400 500 600

but neither helps with the fan just stopping completely.

Is this true for all attempts to set any duty cycle, or only if the duty cycle is lower than some value? I have found that for each fan type, there is some minimum acceptable duty cycle. If the duty cycle is set too low, the fan may stop spinning. I had to ensure my fan control scripts never set a duty cycle lower than some minimum value.

 

[asmodeus]

Is this true for all attempts to set any duty cycle, or only if the duty cycle is lower than some value? I have found that for each fan type, there is some minimum acceptable duty cycle. If the duty cycle is set too low, the fan may stop spinning. I had to ensure my fan control scripts never set a duty cycle lower than some minimum value.

I had this issue even with a duty cycle of 50%. I'm happy to say that I managed to solve it, but I don't really understand how.
First, I set the thresholds of the 14mmcm fans to

FAN4 | 800.000 | RPM | ok | 0.000 | 100.000 | 200.000 | 1700.000 | 1800.000 | 1900.000

and my CPU fan to

FANA | 1800.000 | RPM | ok | 300.000 | 500.000 | 700.000 | 3300.000 | 3400.000 | 3500.000

I also replaced my 12mmcm fans with more powerful ones. I am now able to control the fans without them stopping and can use a duty cycle as low as 30%.

 

[jgreco]

my 12mm fans

(raises eyebrows, amused)

 

[asmodeus]

(raises eyebrows, amused)

Haha, indeed ;-) That would be something for https://www.reddit.com/r/thingsforants/

 

[asmodeus]

For the record, I'm using a few Noctua fan models and these are the IPMI threshold settings that seem to work well with them:

- NF-F12 PPC 3000 PWM (12cm):
lower 400 500 600
upper 3300 3400 3500

- NF-S12B redux 1200 PWM (12cm):
lower 100 200 300
upper 1300 1400 1500

- NF-P14s redux 1200 PWM (14cm):
lower 100 200 300
upper 1300 1400 1500

- NF-P14s redux 1500 PWM (14cm):
lower 100 200 300
upper 1700 1800 1900

HTH!

 

[joeschmuck]

12mm fans do exist, but maybe not for this application:
https://www.newark.com/c/cooling-thermal-management/fans-blowers?brand=multicomp&fan-frame-size=12mm

 

[asmodeus]

After rebooting the box, I'm looking at weird things again. FAN2 and Fan4 are just not reporting any data after a BMC reset. I really wonder if i should replace the mainboard, since it's only two months old...

 

[joeschmuck]

After rebooting the box, I'm looking at weird things again. FAN2 and Fan4 are just not reporting any data after a BMC reset. I really wonder if i should replace the mainboard, since it's only two months old...

You were playing with the FAN settings, true?
If true then I'd reset the system to factory defaults and ensure you are not having a self-induced problem. If the problem still exists then ensure it's not just a faulty fan before assuming the motherboard.

My advice is if you want slower running fans then run the intake and exhaust fans on 7VDC vice the motherboard. While it does take a tiny amount of work to do this, it's worth it in the end, well for me it is and I do this for all my systems. The CPU fan would be the only one connected to the motherboard. This will make a quieter system while preserving good airflow, assuming you have a case that supports good airflow.

 

[pro lamer]

run the intake and exhaust fans on 7VDC vice the motherboard. While it does take a tiny amount of work to do this, it's worth it in the end

Did you mean some DIY works? With or without soldering?

Sent from my phone

 

[joeschmuck]

Did you mean some DIY works? With or without soldering?

Sent from my phone

Actually you can do it without soldering if you use a power fan header. If you are using a 4 pin Molex to 3 pin fan header you just swap a few wires around. So the stuff to pay attention to is the fan header can only have 2 power wires (typically black and red) and a third wire is for the Tach and not used, actaully it shouldn't even have a third wire if it's just a power header. Tracing the 3 pin connector red wire out it should land on the yellow (+12VDC) 4 pin Molex power supply connector. What you would do now it trace the 3 pin connector black wire to the black (0 VDC/Ground) 4 pin Molex power connector. Now on the header cable 4 pin Molex connector you remove the black wire and push it into the red pin location (+5VDC) which should be an empty spot. What this does is use the voltage differential between the +5VDC and +12VDC = +7VDC. Now a note of caution, while I've never seen two black wires on the molex connector of a fan header, if that were to exist then you would have to remove/cut the jumper between them. For anyone who has no idea what I'm talking about, don't try this. If you short out the power supply you may fry it. Anyone who has a clue what I'm talking about should be fine. And this is generally reversible. I do this for all my case fans, it runs then at a slower speed and if you have a well designed air flow through your case then all should be good.

 

[jgreco]

What this does is use the voltage differential between the +5VDC and +12VDC = +7VDC.

YIKES. OMG PLEASE DO NOT DO THIS.

Get a frickin' miniature buck converter and do it right. Bonus: you can set arbitrary voltages, not just 7V. And when you run into a fan stall you won't totally hose voltage regulation by the power supply.

 

[joeschmuck]

YIKES. OMG PLEASE DO NOT DO THIS.

Get a frickin' miniature buck converter and do it right. Bonus: you can set arbitrary voltages, not just 7V. And when you run into a fan stall you won't totally hose voltage regulation by the power supply.

You can do that too. But if you have a fan stall, that would be pulling down the +12VDC line so it still effects the power supply and ultimately the computer, assuming the stalled fan can pull that kind of current in the first place. I've not seen to date a current digital fan that can pull that kind of power just because it's stalled. It may continue to try to start the fan rotation but the current is very low. I'm sure there are exceptions for possibly high CFM fans such as in commercial servers however I wouldn't myself recommend slowing server fans down, there is a reason they are high CFM.

I provided the advice and if someone does not feel comfortable doing it, that is fine but it is reliable. Here is a commercial product which does exactly the same thing and uses a single DPDT switch (as indicate by the switch photo) to route either 12VDC or 7VDC to your fans.

And of course there are many fans speed controllers out on the market which come pre-built and many are adjustable. These are nice if you are looking for an adjustable fan speed.

Lastly I should have added a large WARNING banner to the above posting. Not for the faint of heart, only if you know what you are doing, and double check your work before powering on.

 

[jgreco]

You can do that too. But if you have a fan stall, that would be pulling down the +12VDC line so it still effects the power supply and ultimately the computer, assuming the stalled fan can pull that kind of current in the first place.

So we probably agree on that. Now, the worrying part... what do you think happens to the +5VDC line with your 7V strategy and something bad happens?

 

[joeschmuck]

So we probably agree on that. Now, the worrying part... what do you think happens to the +5VDC line with your 7V strategy and something bad happens?

Your asking if the +5VDC line voltage level rises above the say +5.25VDC level (the standard max limit). The +5VDC over voltage sensor will shut down the power supply. That is how a regulated computer power supply operates. The ATX 2.2 spec is +/- 5% (4.75 to 5.25) for the 5VDC rail so it must maintain that band or it must turn itself off.

I am perfectly fine that you don't agree my method are smart and someone should use a prebuilt device but I'm also a hobbyist and we do things like this after figuring out the risks and any facts we can obtain. Again, if this type of little project makes someone uncomfortable then they shouldn't do it.

 

[jgreco]

Yes, but it's probably going to be easier to drag that in the high direction. Normally the PSU is in control of the voltage and the switching circuitry is a self-contained feedback loop. The sensing circuitry is designed with the understanding that the current flows in a loop from +5V to ground (or if you are extra-pedant physicist true-electron-flow ... fine, from ground to +5V). If the voltage sags, the PSU compensates by changing the pulse width (duty cycle). A SMPS involves a fairly complicated design involving the careful interplay of the switching circuitry with inductors and capacitors and rapid reaction to state changes in order to maintain the voltage. Bleeding off current to what is effectively another supply could have unintended consequences.

Using a supply in a manner that it was not designed for, and that the designers are unlikely to have anticipated, opens you up to a lot of risk. A SMPS is basically taking AC line current and transforming it to system level DC. There's a nontrivial chance that this could cause component stress in ways that are outside the design parameters, and if you cause a failure in an inductor or other component, the scary thing about SMPS is that there's the potential for it to deliver stupid-high voltages into your gear. It just seems to me like you're taunting your power supply into doing something bad.

If this was a simple AC transformer, hey, yeah, have fun with secondary winding taps. This isn't that though.

 

[joeschmuck]

Welcome back, hope your vacation was relaxing.