Skip to content

Interpreting IPMI sensor thresholds that make no sense

Dedicated Servers by steveipw 5 replies 311 views
11 #1

Picked up a Supermicro X9 from auction last month. IPMI reads CPU at 82°C, marks it CRITICAL. Same chip in my other box idles at 78°C, runs for years. Something's off.

Checked the raw sensor data:

  • Lower Non-Critical: 5°C
  • Lower Critical: -5°C
  • Upper Non-Critical: 75°C
  • Upper Critical: 80°C

But the CPU Tjunction is 105°C. 80°C critical makes no sense for this chip.

Dug into SDR records. The threshold byte values don't match the sensor type. Looks like the SDR was built for a 65W TDP CPU from 2008, not the 120W Xeon currently installed. Vendor never updated the SDR when they refreshed the board revision.

For that money you get a board with sensors pretending it's still 2003. Anyone else seen this? Specifically looking at:

  • Supermicro X9/X10 with E5 v2/v3 CPUs
  • Intel S2600 series
  • Quanta/Hetzner rebadges

What's your workaround? I just ignore IPMI temp alerts and watch `coretemp` directly, but that defeats the purpose of out-of-band monitoring.

zfs send | zfs receive. repeat.
#2

Just use lm_sensors and stop trusting BMC firmware written by people who think systemd is "too modern"

Seriously though this is a skill issue on the vendor side. IPMI spec lets them set whatever thresholds they want, and they never validate against actual silicon. "legacy from 2003" is generous, probably copypasted from a Pentium 4 datasheet and nobody noticed because who reads IPMI logs until something actually melts

oops: 0000 [#1] SMP
#3

RTFM

The fix is in `ipmitool sensor thresh` but you need to know the exact sensor number

Ipmitool sensor list | grep -I temp
Ipmitool sensor thresh <num> <threshold> <value>

Or just automate it

- name: fix broken IPMI thresholds
Command: >
Ipmitool sensor thresh {{ item.num }}
{{ item.type }} {{ item.value }}
Loop: "{{ ipmi_temp_sensors }}"

Bullet points for OP:

  • SDR is stored in BMC flash, not board EEPROM on most Supermicro
  • You can flash updated SDR with `ipmitool raw` or vendor tool
  • Some boards need jumper set to allow SDR write
  • Always backup current SDR first: `ipmitool sdr dump backup.bin`

RTFM the IPMI spec chapter 35, it's all there

2 #4

Yea I got a $45 x9 from InterServer last year

Same thing, cpu "critical" at like 70c

I just disabled the alerts in ipmi web gui, costs nothing and works fine

For that money you get what u pay for, im not gonna spend hours fixing firmware on a $45 box

A lot of these auction boards are pulls from datacenters that never updated anything, definitely not the sdr

$3/year. 128MB RAM. Pure happiness.
#5

78°C idle sounds high to me

#6

Which X9 board variant?

No logs, no proof. I have logs.

Post a reply

You need an account to reply. Log in or register to join the conversation.

Post reply Preview Save draft