Temperature and cooling
Thermal problems rarely announce themselves. They show up as a machine that is mysteriously slower than it was, and only sometimes.
Read the sensors
apt install lm-sensors && sensors-detect, then:
sensors
And from the management controller, which sees inlet and exhaust as well as the board:
ipmitool sdr type temperature
ipmitool sdr type fan
See IPMI and the BMC. The BMC's view is the authoritative one, because it is what the fan control actually responds to.
What normal looks like
- Inlet air
- 18–27 °C in a properly cooled rack. A rising inlet temperature is a facility matter and therefore ours.
- CPU under load
- 50–75 °C is unremarkable. Throttling usually begins somewhere above 90.
- NVMe
- Often the hottest component in the chassis, and the first to throttle. Sustained writes on a drive without a heatsink will do it.
- Spinning disks
- Below about 45 °C. Consistently hotter shortens their life measurably — check with SMART.
Spotting throttling
The symptom is a machine that benchmarks well for thirty seconds and badly for ten minutes. Watch the clocks under sustained load:
watch -n1 "grep MHz /proc/cpuinfo | head"
or more precisely, turbostat, which shows the reason for every drop.
For NVMe:
nvme smart-log /dev/nvme0 | grep -i 'temperature\|throttle'
Non-zero throttle counters mean the drive is protecting itself and your sustained write figure is not what you measured in a short test. See benchmarking a new server for why short tests mislead.
What you can do, and what you cannot
On your side: do not disable the kernel's thermal handling, keep the firmware current, and be realistic about sustained workloads on dense hardware. If an application pins every core continuously, say so when you order and we will specify a chassis with the airflow for it.
What you cannot do remotely is clean a filter, reseat a heatsink or replace a fan. That is ours.
When to open a ticket
- Inlet temperature climbing over days
- A fan reported at zero RPM, or one spinning far faster than its neighbours
- CPU temperatures high at low load — that is a mounting or paste problem, not a workload one
- Thermal events in
ipmitool sel list - Clocks dropping under load on a machine that used to hold them
Send the sensors and ipmitool sdr output with the machine
identifier. Fans and heatsinks are replaced from stock in Amsterdam by our own
engineers — see hardware failure.
Still stuck? Mail support@novogara.com — an engineer answers, at any hour. Back to the knowledge base