Strange CPU temp readings

[icsy7867]

So I am currently using an AMD FX-8120 processor. This is a Bulldozer and I have been trying to do a lot of research on how to get this reading by using

Code:

sysctl -a | grep -i temp

I came across this little nugget from couple of years ago:
https://forums.freenas.org/index.ph...reenas-transition-and-hardware-upgrade.14529/

and I added to my loader.conf via tunables:

Code:

amdtemp_load="YES"
hint.acpi_throttle.0.disabled="YES" 

and this to sysctl via tunables:

Code:

dev.amdtemp.0.sensor_offset="-6" 

Now it would appear that this offset may change per computer. So I would definitely need to fiddle with that to get it to match the temp in my BIOS. I also ran a memtest last night which had the same reading as my bios.

Now here is were things get a little odd... If I run:

Code:

sysctl -a | grep -i temp

I do get some helpful output!

Code:

dev.cpu.0.temperature: 28.3C
dev.cpu.1.temperature: 28.3C
dev.cpu.2.temperature: 28.3C
dev.cpu.3.temperature: 28.3C
dev.cpu.4.temperature: 28.3C
dev.cpu.5.temperature: 28.3C
dev.amdtemp.0.%desc: AMD CPU On-Die Thermal Sensors
dev.amdtemp.0.%driver: amdtemp
dev.amdtemp.0.%parent: hostb4
dev.amdtemp.0.sensor_offset: -6
dev.amdtemp.0.core0.sensor0: 28.3C

Which is great. It appears to be reading something. But I can't trust these numbers. If I run the command over and over very quickly I see a very very strange behavior.

The temps are DESCENDING from ~31 Degrees to ~25 Degrees. And once it hits 24-25 (I can only run the command so fast, but the numbers are VERY consistent) it starts back over to ~31 Degrees and this cycle just keeps going.

Oddly enough the starting and ending temps are about 6 degrees apart. I will try and change the offset value, but this shouldnt be causing this loop behavior (I wouldnt think!).

Anyone have any ideas? I read in a couple places that FreeBSD 10 should have better AMD support for the bulldozer processors. This isnt a big deal. I just have been trying to shut off cores on my 8120 and underclock/voltage it to 2.6GHz since 8 cores @ 3.1ghz is a little overkill.

As always, I am loving freenas and learning a ton! I really appreciate all of your help!

 

[icsy7867]

Just a note... while googling I found this
https://lists.freebsd.org/pipermail/freebsd-stable/2013-March/072825.html

Additionally someone suggested that AMD CPU's just poll more frequently and that these numbers may not be wrong...

So with this I thought I would do some testing. It seems with the cpu is idling (Or at least less than 1.00 CPU load), the temperatures seem to cycle through those temperatures. Perhaps it is something to do with the AMD Cool N' quiet feature I enabled.

So to drive up CPU utilization I found that seeking to different places on plex constantly my CPU zooms off to fill the video buffer. This actually seems like an easy way to max out my CPU utilization. Now upon doing this I see a different behavior when running

Code:

sysctl -a | grep -i temp

Now when repeatedly seeking in plex i see the temperature jump up. I am seeing increases in the 0.2-0.5 range every couple of refreshes of the command, and this seems to max out at about 60 degrees celsius and the "countdown" like behavior ceases to happen.

Once i stop stressing the CPU, that behavior resumes. at around 35-40 degrees celsius. Now I am more convinced that these temp readings may actually be working correctly, but that may need to adjust that offset value. I would also like to stress test the CPU for 10 minutes or so to see where these temperatures reach.

Anyone have any idea of any CPU stress test tools that I can boot into?

 

[MrToddsFriends]

Anyone have any idea of any CPU stress test tools that I can boot into?

Have a look at prime95, which is the hardest (read: best) CPU stress test I know of.

http://www.mersenne.org/download/

Use your favorite boot environment and the download from mersenne.org that fits best to it.

If you are brave you can even run prime95 in FreeNAS. I wouldn't recommend that for a box in a production environment, but it's an option for a burn-in test.

p95v279.FreeBSD64.tar.gz (64-bit FreeBSD 8.0 version) runs on FreeNAS 9.3.x if you do a

Code:

ln -s /usr/local/lib/libcurl.so.7 /usr/local/lib/libcurl.so.6

prior to starting mprime, which will ask a few questions upon start.
15. Options/Torture Test
2 = In-place large FFTs (maximum heat and power consumption, some RAM tested).
are suitable choices.

The soft link /usr/local/lib/libcurl.so.6 can be deleted afterwards.

 

[icsy7867]

Thanks for this!

I actually found a lot of other threads around the internet
(see http://www.overclock.net/t/1343392/help-cpu-too-cold-amd-fx-6300-idles-at-10c-10-degrees-celsius/10)

Code:

Why should I use "Core Temp" and when?
AMD designed this equation to accurately read peak (45C+) and load temps. It has an equational offset to determine said temps which equalizes at 45C. Since it's designed for peak temps and is a non-physical temperature it cannot read idle temps or account for ambient temps correctly.

Although I see very little officially from AMD. I see this in lots of places. Which would seem to fit with what my FX Bulldozer is doing.

I found the the Ultimate Boot CD has Prime95 already built into it. Since I am unsure if freebsd is reading those temps correctly or not, I think trying it in a completely different environment may be best. I think I still have room to undervolt as well since I have cores disabled and I have underclocked to 2.5GHz.

I have a busy week of work and school, so I will try this on thursday and report back. I know most users are using Intel's Atoms, Xeons and i Series so this may not relate to very many people, but maybe it will help someone!

 

[icsy7867]

Interstingly enough, I booted my machine into windows, and downloaded coretemp which showed the exact some behavior. So it would appear that when idling, it seems to cycle through temperatures. Maybe it's because I have Cool N' Quiet enabled, but I am not quite sure. Once the CPU starts getting some usage the temperatures go up and with hold at values and read just fine (although I am still not sure if there is an offset issue, or if the numbers are high or low). But at least it would seem freenas isnt being too crazy while reading these values!