Super Micro fan control for Linux (home) servers.
This is a systemd service running on Linux and is able to control fans in CPU and HD zones with the help of IPMI on Super Micro X10-X13 (and some X9) motherboards.
You can also run smfc in docker, see more details in Docker.md.
- a Super Micro motherboard with a BMC chip (i.e. ASPEED AST2400/2500/2600)
- Python 3.7+
- Linux OS with:
systemdpackagecoretempkernel module for Intel(R) CPUsk10tempkernel module for AMD(R) CPUsdrivetempkernel module (kernel version 5.6+ required) modules for SATA hard disks
bashipmitool- optional:
smartmontoolsfor the standby guard feature - optional:
hddtempfor the SAS/SCSI disks
- Set up the IPMI threshold values for your fans (see script
ipmi/set_ipmi_threshold.sh). - Optional: enable advanced power management features for your CPU and SATA hard disks for lower power consumption, heat generation and fan noise.
- Load kernel modules (
coretemp/k10tempanddrivetemp). - Install the service with running the script
install.sh. - Edit the configuration file
/opt/smfc/smfc.confand command line options in/etc/default/smfc. - Start the
systemdservice - Check results in system log
- Leave a feedback in issue #19
This service was planned for Super Micro motherboards installed in computer chassis with two independent cooling systems employing separate fans. In IPMI terms these are called:
- CPU zone (FAN1, FAN2, etc.)
- HD or peripheral zone (FANA, FANB, etc.)
Please note: the fan assignment to zones is predefined in IPMI, and it cannot be changed! On the other hand smfc implements a feature, called Swapped zones, in order to make the use of the fans more suitable.
In this service a fan control logic is implemented for both zones which can:
- read the zone's temperature from Linux kernel
- calculate a new fan level based on the user-defined control function and the current temperature value of the zone
- set up the new fan level through IPMI in the zone
The fan control logic can be enabled and disabled independently per zone. In the zone all fans will have the same rotational speed. The user can configure different temperature calculation method (e.g. minimum, average, maximum temperatures) in case of multiple heat sources in a zone.
Please note that smfc will set all fans back to 100% speed at service termination in order to avoid overheating!
The user-defined parameters (see configuration file below for more details) create a function where a temperature interval is being mapped to a fan level interval.
The following five parameters will define the function in both zones:
min_temp=
max_temp=
min_level=
max_level=
steps=
With the help of this function smfc can map any new temperature measurement value to a fan level. Changing the fan rotational speed is a very slow process (i.e. it could take seconds depending on fan type and the requested amount of change), so we try to minimize these kinds of actions. Instead of setting fan rotational speed continuously we define discrete fan levels based on steps= parameter.
In order to avoid/minimize the unnecessary change of fan levels the service employs the following steps:
- When the service adjusts the fan rotational speed then it always applies a delay time defined in configuration parameter
[IPMI] fan_level_delay=in order to let the fan implement the physical change. - There is a sensitivity threshold parameter (
sensitivity=) for the fan control logic. If the temperature change is below this value then the service will not react at all. - The configuration parameter
polling=defines the frequency of reading zone's temperature. The bigger polling time in a zone the lower frequency of fan speed change.
In some cases it is useful to swap IPMI zones. In this way the fans FAN1, FAN2, ... will cool the HD zone and the fans FANA, FANB, ... will cool the CPU zone. This feature could be useful if you need more fans for the HD zone since Super Micro motherboards have more fan connectors in the CPU zone usually. This feature can be enabled with [IPMI] swapped_zones=True configuration parameter, in default it is disabled.
For HD zone an additional optional feature was implemented, called Standby guard, with the following assumptions:
- SATA hard disks are organized into a RAID array
- the RAID array will go to standby mode recurrently
This feature is monitoring the power state of SATA hard disks (with the help of the smartctl) and will put the whole array to standby mode if a few members are already stepped into that. With this feature we can avoid a situation where the array is partially in standby mode while other members are still active.
The smfc service was originally designed for SATA hard drives, but from 3.0 version it is also compatible with NVME and SAS/SCSI disks. The following table summarizes how the temperature is read for different disk types:
| Disk type | Temperature source | Kernel module | Command |
|---|---|---|---|
SATA |
Linux kernel (HWMON) | drivetemp |
- |
NVME |
Linux kernel (HWMON) | - | - |
SAS/SCSI |
hddtemp |
- | hddtemp |
Some additional notes:
- For
NVMESSDs no kernel driver will be loaded the kernel itself can handle this disk type - For
SATAdisks thedrivetempkernel module should be loaded (this is the fastest way to read disk temperature and the kernel module can report the temperature during sleep mode!) - For
SAS/SCSIdisks thehddtempcommand will be used to read disk temperature (NO daemon mode is required forhddtemp!) - Different disks types can be mixed in
hd_names=configuration parameter but the power management (standy mode) and Standby guard feature will not be supported in this case. - Although
smfccan handle NVME SSDs, it is NOT RECOMMENDED to include NVME disks inhd_names=parameters, because the hard disks and NVME SSDs are operating in quite different temperature zones (e.g. 40 C vs 70 C). - The service can identify the disk types automatically based on the tags (
ata-/-SATA,nvme-andscsi-)
Originally this software was designed to work with Super Micro X10 and X11 motherboards with a BMC chip (i.e. ASPEED AST2400/2500) and IPMI functionality.
In case of X9 motherboards the compatibility is not guaranteed, it depends on the hardware components of the motherboard (i.e. not all X9 motherboards employ BMC chip).
The earlier X8 motherboards are NOT compatible with this software. They do not implement IPMI_FULL mode, and they cannot control fan levels how it is implemented in smfc.
X13 motherboards (with AST2600 BMC chips) seem to be compatible with smfc (see mode details in issue #33 about an X13SAE-F motherboard).
Fan control and IPMI_FULL mode are working properly. The only difference is in using thresholds, AST2600 implements only Lower Critical threshold, so setting up thresholds is different in this case.
Feel free to create a short feedback in issue #19 on your compatibility experience.
IPMI uses six sensor thresholds to specify the safe and unsafe fan rotational speed intervals (these are RPM values rounded to nearest hundreds, defined for each fan separately):
Lower Non-Recoverable
Lower Critical
Lower Non-Critical
Upper Non-Critical
Upper Critical
Upper Non-Recoverable
Like many other utilities (created by NAS and home server community), smfc also uses IPMI FULL mode for fan control, where all fans in the zone:
- initially configured to full speed (100%)
- then their speed can be safely configured in
[Lower Critical, Upper Critical]interval - if any fan speed oversteps either
Lower CriticalorUpper Criticalthreshold then IPMI will generate an assertion event and will set the all fan speeds back to 100% in the zone
Please also consider the fact that fans are mechanical devices, their rotational speed is not stable (it could be fluctuating). In order to avoid IPMI's assertion mechanism described here please follow the next steps:
- Per fan: check the minimum and maximum rotational speeds of your fan on its vendor website
- Per fan: configure proper IMPI sensor thresholds adjusted to the fan speed interval
- Per zone: define safe
min_level/max_levelvalues forsmfcrespecting the variance of the all fans in the IPMI zone (it could take several iterations and adjustments)
Here is a real-life example for a Noctua NF-F12 PWM fan:
Upper Non-Recoverable = 1800 rpm
Upper Critical = 1700 rpm
Upper Non-Critical = 1600 rpm
Lower Non-Critical = 200 rpm
Lower Critical = 100 rpm
Lower Non-Recoverable = 0 rpm
Max RPM = 1500 rpm
Min PRM = 300 rpm
max_level = 100 (i.e. 1500 rpm)
min_level = 35 (i.e. 500 rpm)
Notes:
-
Use the following
ipmitoolcommand to display the current IMPI sensor thresholds for fans:root@home:~# ipmitool sensor|grep FAN FAN1 | 500.000 | RPM | ok | 0.000 | 100.000 | 200.000 | 1600.000 | 1700.000 | 1800.000 FAN2 | 500.000 | RPM | ok | 0.000 | 100.000 | 200.000 | 1600.000 | 1700.000 | 1800.000 FAN3 | na | | na | na | na | na | na | na | na FAN4 | 400.000 | RPM | ok | 0.000 | 100.000 | 200.000 | 1600.000 | 1700.000 | 1800.000 FANA | 500.000 | RPM | ok | 0.000 | 100.000 | 200.000 | 1600.000 | 1700.000 | 1800.000 FANB | 500.000 | RPM | ok | 0.000 | 100.000 | 200.000 | 1600.000 | 1700.000 | 1800.000 -
Use the following
ipmitoolcommand to list assertion events:root@home:~# ipmitool sel list 1 | 10/19/2023 | 05:15:35 PM CEST | Fan #0x46 | Lower Critical going low | Asserted 2 | 10/19/2023 | 05:15:35 PM CEST | Fan #0x46 | Lower Non-recoverable going low | Asserted 3 | 10/19/2023 | 05:15:38 PM CEST | Fan #0x46 | Lower Non-recoverable going low | Deasserted 4 | 10/19/2023 | 05:15:38 PM CEST | Fan #0x46 | Lower Critical going low | Deasserted 5 | 10/19/2023 | 05:20:59 PM CEST | Fan #0x46 | Lower Critical going low | Asserted -
Use the following
ipmitoolcommands to specify all six sensor thresholds for FAN1:root@home:~# ipmitool sensor thresh FAN1 lower 0 100 200 root@home:~# ipmitool sensor thresh FAN1 upper 1600 1700 1800 -
You can also edit and run
ipmi/set_ipmi_treshold.shto configure all IPMI sensor thresholds -
If you install a new BMC firmware on your Super Micro motherboard you have to configure IPMI thresholds again
-
If you do not see fans when executing
ipmitool sensors, you may want to reset the BMC to factory default using the Web UI or usingipmitool mc reset cold -
Noctua specifies the variance of minimum and maximum fan rotational speeds (e.g. see the specification of Noctua NF-F12 PWM). For example:
Rotational speed (+/- 10%) 1500 RPM: 1350-1650 RPM intervalMin. rotational speed @ 20% PWM (+/-20%) 300 RPM: 240-360 RPM interval
Please note that LNA/ULNA cables or Y-cables can modify the rotational speed calculations here and the required IPMI sensor thresholds too.
You can read more about:
- IPMI fan control: STH Forums and TrueNAS Forums
- Change IPMI sensors thresholds: TrueNAS Forums
If low noise and low heat generation are important attributes of your Linux box, then you may consider the following chapters.
Most of the modern CPUs has multiple energy saving features. You can check your BIOS and enable them in order to minimize the heat generation.
Intel(R) CPUs:
- Intel(R) Speed Shift Technology
- Intel(R) SpeedStep
- C-states
- Boot performance mode
AMD(R) CPUs:
- PowerNow!
- Cool`n`quiet
- Turbo Core
With this setup the CPU will change its base frequency and power consumption dynamically based on the load.
In case of SATA hard disks, you may enable:
- advanced power management
- spin down timer
With the help of command hdparm you can enable advanced power management and specify a spin down timer (read more here):
hdparm -B 127 /dev/sda
hdparm -S 240 /dev/sda
In file /etc/hdparm.conf you can specify all parameters in a persistent way:
quiet
/dev/sda {
apm = 127
spindown_time = 240
}
/dev/sdb {
apm = 127
spindown_time = 240
}
...
Important notes:
- If you plan to spin down your hard disks or RAID array (i.e. put them to standby mode) you have to set up the configuration parameter
[HD zone] polling=minimum twice bigger as thespindown_timespecified here. - In file
/etc/hdparm.confyou must define HD names in/dev/disk/by-id/...form to avoid inconsistency.
We need to load the following important Linux kernel modules:
coretemp: temperature report for Intel(R) CPUsk10temp: temperature report for AMD(R) CPUsdrivetemp: temperature report for SATA hard disks (available from kernel 5.6+ version)
Use /etc/modules file for persistent loading of these modules.
Here are some sample HWMON file locations for these kernel modules:
coretemp:/sys/devices/platform/coretemp.0/hwmon/hwmon*/temp1_inputk10temp:/sys/bus/pci/drivers/k10temp/0000*/hwmon/hwmon*/temp1_inputdrivetemp:/sys/class/scsi_disk/0:0:0:0/device/hwmon/hwmon*/temp1_input
Notes:
smfcis able to find the proper HWMON file automatically for Intel(R) CPUs and SATA hard drives, but users of the AMD(R) CPU should specify manually (seehwmon_path=parameter in the config file)- Reading
drivetempmodule is the fastest way to get the temperature of the hard disks, and it can read temperature of the SATA hard disks even in standby mode, too.
For the installation you need a root user. The default installation script install.sh will use the following folders:
| File | Installation folder | Description |
|---|---|---|
smsc.service |
/etc/systemd/system |
systemd service definition file |
smsc |
/etc/default |
service command line options |
smsc.py |
/opt/smfc |
service (python program) |
smsc.conf |
/opt/smfc |
service configuration file |
but you can use freely any other folders too. The service has the following command line options:
root@home:~/opt/smfc# ./smfc.py --help
usage: smfc.py [-h] [-c CONFIG_FILE] [-v] [-l {0,1,2,3,4}] [-o {0,1,2}]
optional arguments:
-h, --help show this help message and exit
-c CONFIG_FILE configuration file
-v show program's version number and exit
-l {0,1,2,3,4} log level: 0-NONE, 1-ERROR(default), 2-CONFIG, 3-INFO, 4-DEBUG
-o {0,1,2} log output: 0-stdout, 1-stderr, 2-syslog(default)
You may configure logging output and logging level here and these options can be specified in /etc/default/smfcin a persistent way.
Edit /opt/smfc/smfc.conf and specify your configuration parameters here:
#
# smfc.conf
# smfc service configuration parameters
#
[Ipmi]
# Path for ipmitool (str, default=/usr/bin/ipmitool)
command=/usr/bin/ipmitool
# Delay time after changing IPMI fan mode (int, seconds, default=10)
fan_mode_delay=10
# Delay time after changing IPMI fan level (int, seconds, default=2)
fan_level_delay=2
# CPU and HD zones are swapped (bool, default=0).
swapped_zones=0
[CPU zone]
# Fan controller enabled (bool, default=0)
enabled=1
# Number of CPUs (int, default=1)
count=1
# Calculation method for CPU temperatures (int, [0-minimum, 1-average, 2-maximum], default=1)
temp_calc=1
# Discrete steps in mapping of temperatures to fan level (int, default=6)
steps=6
# Threshold in temperature change before the fan controller reacts (float, C, default=3.0)
sensitivity=3.0
# Polling time interval for reading temperature (int, sec, default=2)
polling=2
# Minimum CPU temperature (float, C, default=30.0)
min_temp=30.0
# Maximum CPU temperature (float, C, default=60.0)
max_temp=60.0
# Minimum CPU fan level (int, %, default=35)
min_level=35
# Maximum CPU fan level (int, %, default=100)
max_level=100
# Optional parameter, it will be generated automatically for Intel CPUs and must be specified manually for AMD CPUs.
# Path for CPU sys/hwmon file(s) (str multi-line list, default=/sys/devices/platform/coretemp.0/hwmon/hwmon*/temp1_input)
# hwmon_path=/sys/devices/platform/coretemp.0/hwmon/hwmon*/temp1_input
# /sys/devices/platform/coretemp.1/hwmon/hwmon*/temp1_input
# or
# hwmon_path=/sys/bus/pci/drivers/k10temp/0000*/hwmon/hwmon*/temp1_input
[HD zone]
# Fan controller enabled (bool, default=0)
enabled=1
# Number of HDs (int, default=1)
count=1
# Calculation of HD temperatures (int, [0-minimum, 1-average, 2-maximum], default=1)
temp_calc=1
# Discrete steps in mapping of temperatures to fan level (int, default=4)
steps=4
# Threshold in temperature change before the fan controller reacts (float, C, default=2.0)
sensitivity=2.0
# Polling interval for reading temperature (int, sec, default=10)
polling=10
# Minimum HD temperature (float, C, default=32.0)
min_temp=32.0
# Maximum HD temperature (float, C, default=46.0)
max_temp=46.0
# Minimum HD fan level (int, %, default=35)
min_level=35
# Maximum HD fan level (int, %, default=100)
max_level=100
# Names of the HDs (str multi-line list, default=)
# These names MUST BE specified in '/dev/disk/by-id/...' form!
hd_names=
# Optional parameter, it will be generated automatically based on the disk names.
# List of files in /sys/hwmon file system or 'hddtemp' (str multi-line list, default=)
# hwmon_path=/sys/class/scsi_disk/0:0:0:0/device/hwmon/hwmon*/temp1_input
# /sys/class/scsi_disk/1:0:0:0/device/hwmon/hwmon*/temp1_input
# hddtemp
# Standby guard feature for RAID arrays (bool, default=0)
standby_guard_enabled=0
# Number of HDs already in STANDBY state before the full RAID array will be forced to it (int, default=1)
standby_hd_limit=1
# Path for 'smartctl' command (str, default=/usr/sbin/smartctl).
# Required for 'standby guard' feature only
smartctl_path=/usr/sbin/smartctl
# Path for 'hddtemp' command (str, default=/usr/sbin/hddtemp).
# Required for reading of the temperature of SAS/SCSI disks.
hddtemp_path=/usr/sbin/hddtemp
Important notes:
-
[HD zone} hd_names=: This is a compulsory parameter, its value must be specified in/dev/disk/by-id/...form (the/dev/sdaform is not persistent could be changed after a reboot). -
[CPU zone] / [HD zone] min_level= / max_level=: Check the stability of your fans and adjust the fan levels based on your measurement. As it was stated earlier, IPMI can switch back to full rotational speed if fans reach specific thresholds. You can collect real data about the behavior of your fans if you edit and run scriptipmi/fan_measurement.sh. The script will set fan levels from 100% to 20% in 5% steps and results will be saved in the filefan_result.csv:root:~# cat fan_result.csv Level,FAN1,FAN2,FAN4,FANA,FANB 100,1300,1300,1200,1300,1300 95,1300,1300,1100,1200,1300 90,1200,1200,1100,1200,1200 85,1100,1100,1000,1100,1100 80,1100,1100,1000,1100,1100 75,1000,1000,900,1000,1000 70,900,900,800,1000,900 65,900,900,800,900,900 60,800,800,700,900,800 55,700,700,700,800,700 50,700,700,600,700,700 45,600,600,500,700,600 40,500,500,500,600,500 35,500,500,400,500,500 30,400,400,300,400,400 25,300,300,300,400,300 20,1300,1300,1200,1300,1300My experience is that Noctua fans in my box are running stable in the 35-100% fan level interval. An additional user experience is (see issue #12) when Noctua fans are paired with Ultra Low Noise Adapter the minimum stable fan level could go up to 45% (i.e. 35% is not stable).
-
[CPU zone] / [HD zone] hwmon_path=: This parameter is optional for Intel(R) CPUs and SATA drives (i.e.smfccan identify automatically the proper file locations), but must be specified manually for AMD(R) CPUs. In case of SAS/SCSI hard disks (wheredrivetempcannot be loaded) you can specifyhddtempvalue. You can use wild characters (?,*) in this parameter andsmfcwill do the path resolution automatically. -
Several sample configuration files are provided for different scenarios in folder
./src/samples. Please take a look on them, it could be a good starting point in the creation of your own configuration.
This systemd service can be started and stopped in the standard way. Do not forget to reload systemd configuration after a new installation or if you changed the service definition file:
systemctl daemon-reload
systemctl start smfc.service
systemctl stop smfc.service
systemctl restart smfc.service
systemctl status smfc.service
● smfc.service - Super Micro Fan Control
Loaded: loaded (/etc/systemd/system/smfc.service; enabled; vendor preset: enabled)
Active: active (running) since Fri 2021-09-17 23:28:10 CEST; 1 day 19h ago
Main PID: 1064180 (smfc.py)
Tasks: 1 (limit: 38371)
Memory: 7.4M
CPU: 41.917s
CGroup: /system.slice/smfc.service
└─1064180 /usr/bin/python3 /opt/smfc/smfc.py -c /opt/smfc/smfc.conf -l 2
Sep 19 17:12:39 home smfc.service[1064180]: CPU zone: new level > 39.0C > [T:40.0C/L:61%]
Sep 19 17:12:42 home smfc.service[1064180]: CPU zone: new level > 33.0C > [T:35.0C/L:48%]
Sep 19 17:48:14 home smfc.service[1064180]: CPU zone: new level > 38.0C > [T:40.0C/L:61%]
If you are testing your configuration, you can start smfc.py directly in a terminal. Logging to the standard output and debug log level are useful in this case:
cd /opt
sudo smfc.py -o 0 -l 3
All messages will be logged to the specific output and the specific level.
With the help of command journalctl you can check logs easily. For examples:
-
listing service logs of the last two hours:
journalctl -u smfc --since "2 hours ago" -
listing service logs from the last boot:
journalctl -b -u smfc
Most probably the rotational speed of the fans went above or below of a threshold value You can check the current fan rotational speeds:
ipmitool sdr
and you can also check event log on Super Micro remote web interface (Server Health > Health Event log). If you see Assertions log messages for fans:
Fan(FAN1) Lower Critical - going low - Assertion
Fan(FAN1) Lower Non-recoverable - going low - Assertion
Fan(FAN1) Lower Non-recoverable - going low - Deassertion
Fan(FAN1) Lower Critical - going low - Deassertion
Fan(FAN4) Lower Critical - going low - Assertion
Fan(FAN4) Lower Non-recoverable - going low - Assertion
then you must adjust your configuration (i.e. threshold values) because IPMI switched back to full rotational speed.
Q: I would like to use constant fan rotational speed in one or both zones. How can I configure that?
You should configure the temperatures and levels with the same value.
min_temp=40
max_temp=40
min_level=60
max_level=60
With this setup there will be a constant 60% fan level in the specific zone. The temperature value is ignored, steps parameter is also ignored.
The problem is that the specific file cannot be found in HWMON system. The potential reasons behind this issue could be:
drivetempdriver cannot support your disks (it support only SATA hard disks). In case of SAS/SCSI hard disks you can usehddtempinstead ofdrivetemp. See more details in issue #21.- Maybe you specified the
hwmon_path=parameter manually and it contains an invalid path. You can correct it.
The configuration is the following:
-
64 GB ECC DDR4 RAM
-
Fractal Design Node 804 case, with separate chambers for the motherboard and the hard disks:
-
Debian Linux LTS (actually bookworm with backported Linux kernel 6.5)
-
8 x WD Red 12TB (WD120EFAX) hard disks in ZFS RAID
-
3 x Noctua NF-12 PWM fans (FAN1, FAN2, FAN4) in CPU zone
-
2 x Noctua NF-12 PWM fans (FANA, FANB) in HD zone
Further readings:
- BMC IPMI User's Guide 1.1b (X10/X11/H11)
- BMC resources
- IPMI Utilities
- IPMICFG download
- IPMICFG User's Guide 1.15
- [STH forums] Reference Material: Supermicro X9/X10/X11 Fan Speed Control
- [TrueNAS forums] How To: Change IPMI Sensor Thresholds using ipmitool
- [TrueNAS forums] Script to control fan speed in response to hard drive temperatures
- [Pcfe's blog] Set fan thresholds on my Super Micro H11DSi-NT
- [coretemp] documentation
- [drivetemp] documentation and its GitHub respository
- How to install hddtemp from a source package
- [GitHub] Kevin Horton's nas_fan_control
- [GitHub] Rob Urban's fork nas_fan control
- [GitHub] sretalla's fork nas_fan control
- [GitHub] Andrew Gunnerson's ipmi-fan-control
Written with StackEdit.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent