AMD Opteron 6276 CPU Temp is below Ambient

[ChaosBlades]

The ambient temperature in the room this server is in is around 21C so this is clearly wrong. I know over the last several months there have been some updates to TrueNAS to correct these kinds of issues but it has not fixed this CPU. Does anyone have any information specifically on this generation of CPU. Is this CPU to old to bother with opening a support ticket for? The motherboard is a Supermicro H8SGL-F.

 

[jgreco]

What is the system reporting its temperature as? If the underlying system is reporting bullchips, this wouldn't even be a TrueNAS issue. This is unfortunately somewhat commonplace especially if you've sourced a mainboard from some of the common consumer-grade vendors.

 

[Ericloewe]

An Opteron 6276 is pretty old and from a time when AMD wasn't at the top of its game, so it's fairly likely nobody ever paid that much attention to this sort of detail.

 

[ChaosBlades]

To be clear, I don't care either way. I am only using this old system for backups of backups. I put it together out of scrap parts for a temporary solution. It will be fully retired in a few months once SCALE comes out. I figured while I had it up I might as well report a bug if this is one.

What is the system reporting its temperature as? If the underlying system is reporting bullchips, this wouldn't even be a TrueNAS issue. This is unfortunately somewhat commonplace especially if you've sourced a mainboard from some of the common consumer-grade vendors.

TrueNAS is reporting 18.4C with a 21C ambient on a Supermicro H8SGL-F, which is absolutely not some consumer-grade vendor. It is a name brand server board. A fairly high end one at that.

FreeBSD is reporting the same temp as TrueNAS. The below temp is different than the original post because I took it at a different time ;)

root@Manda[~]# sysctl -a | grep temperature
dev.cpu.15.temperature: 17.6C
dev.cpu.14.temperature: 17.6C
dev.cpu.13.temperature: 17.6C
dev.cpu.12.temperature: 17.6C
dev.cpu.11.temperature: 17.6C
dev.cpu.10.temperature: 17.6C
dev.cpu.9.temperature: 17.6C
dev.cpu.8.temperature: 17.6C
dev.cpu.7.temperature: 17.6C
dev.cpu.6.temperature: 17.6C
dev.cpu.5.temperature: 17.6C
dev.cpu.4.temperature: 17.6C
dev.cpu.3.temperature: 17.6C
dev.cpu.2.temperature: 17.6C
dev.cpu.1.temperature: 17.6C
dev.cpu.0.temperature: 17.6C

An Opteron 6276 is pretty old and from a time when AMD wasn't at the top of its game, so it's fairly likely nobody ever paid that much attention to this sort of detail.

Both TrueNAS and FreeBSD was around when this board and CPU was in data centers and I don't remember the temp being wrong years ago when I used this before it was EOL and still under warranty (the board anyway). My question is would be go under a regression or unsupported/EOL.

Edit: I used this CPU/Board back on FreeNAS 9.2

 

[Ericloewe]

It would be a regression, but you’d need to find someone who cared and help them test things (or someone who has the same hardware).

 

[jgreco]

which is absolutely not some consumer-grade vendor. It is a name brand server board. A fairly high end one at that.

While I'm a fan of Supermicro, and was largely responsible for the Supermicro advocacy push here in these forums years ago, this is probably a mischaracterization of Supermicro. Companies like HP and Dell are high end server manufacturers, with deep integrations and product designs that are carefully targeted. Supermicro is more of a buffet; you can make your own server options, doing things Supermicro never expected or offered as a prebuilt. This is very handy for ZFS, because so often the prebuilts by HP and Dell come with adverse crap like RAID controllers.

Supermicro H8SGL-F,

The Supermicro H8 stuff was not a popular line. I would equally believe that it worked or it didn't. There were problems with some of these. However, even if it did, the FreeBSD support for CPU temps may well have changed or rotted as new CPU's were added in the last decade. PC temperature reporting is crappy.

The sysctl calls make it clear that this is not a FreeNAS/TrueNAS issue as such, since it is just reporting what FreeBSD is reporting. It would probably be interesting to see what was reported if you booted up an old copy of FreeBSD 9...

 

[QonoS]

edit:

Low temperature readings with 'Brisbane', 'Lima', 'Sparta' - ALCPU Forums

"Low temperature readings with 'Brisbane', 'Lima', 'Sparta'" aka Brisbane Temperature Bug
"It applies to all CPUs based on the 65nm K8 core (Revision G):
Athlon64
Athlon64 X2
Athlon64 X2 BE
Athlon64 BE
Turion64
Turion64 X2
Sempron"
AMD listed this as "Inaccurate Temperature Measurement" (erratum 319) ( https://www.amd.com/system/files/TechDocs/41322_10h_Rev_Gd.pdf )
Since a AMD Opteron 6276 (Interlagos) is mentioned i'm not sure if it applies.


edit2:
there is no known temperature bug (erratum) in opteron 6200 cpus listed by AMD : https://www.amd.com/system/files/TechDocs/48063_15h_Mod_00h-0Fh_Rev_Guide.pdf
So this might be a mainboard or software issue after all.

 

[jgreco]

Well we know for certain that there's temperature sensors. This is more a case of the old nugget about standards, so many to choose from. I imagine someone spent time ten years ago getting this working on the H8 boards, and then that may have been overcome by more recent developments that could have broken this. Could be the BIOS/IPMI/BMC, could be FreeBSD. That's why I suggested trying a boot of an old FreeBSD 9 to see what it reported.

 

[joeschmuck]

To the best of my knowledge the temperature sensors on any computer be it the motherboard or the CPU itself are not calibrated. From my perspective they are just a good indication of temperature and not to be taken as highly accurate. We often hear stuff like +/-3% accuracy for things like thermometers or pressure gauges, well I figure this falls into the same category, but I could be wrong, just a thought. I also read an Intel paper that a Core Duo has a 1.5C thermal accuracy. This was just provided as an example, I did not look up your specific CPU but you might find it if you really want to. I wouldn't be surprised if there were exceptions and special purpose computers like special military projects or something for outer space use could have highly accurate thermal sensors.

My advice, use it as a guide.

 

[ChaosBlades]

To the best of my knowledge the temperature sensors on any computer be it the motherboard or the CPU itself are not calibrated. From my perspective they are just a good indication of temperature and not to be taken as highly accurate. We often hear stuff like +/-3% accuracy for things like thermometers or pressure gauges, well I figure this falls into the same category, but I could be wrong, just a thought. I also read an Intel paper that a Core Duo has a 1.5C thermal accuracy. This was just provided as an example, I did not look up your specific CPU but you might find it if you really want to. I wouldn't be surprised if there were exceptions and special purpose computers like special military projects or something for outer space use could have highly accurate thermal sensors.

My advice, use it as a guide.

Definitely not a calibration issue. I think jgreco is right, if it was a calibration issue it would be at least around 20-30C too low in this case. From what I remember when I ran this setup (same cpu, mobo, cooler, and case) back on FreeNAS 9.2 when the temp was correct I was getting around 35-45C.

The number of people still running these vs. the development time both from a time till the update actually makes it into production TrueNAS and from the work hours involved isn't worth the effort to even submit a bug report which would most likely be on the FreeBSD end. I mainly made this post to make sure it was not something known.

This system will be migrated over to SCALE once it goes into release as a testing platform. It will be interesting to see if linux reports this correctly. I'll update if I remember.

 

[jgreco]

see if linux reports this correctly

Yeah, there's that too. Really, though, if you get a moment, please do consider just booting a FreeBSD 9.2 installer ISO, do not actually INSTALL anything, just go into singleuser/recovery mode, and run "sysctl -a" and grep for those objects.