Hi all,

I've been running an overclocked AMD Phenom II X6 1055T @ 3.8GHz for some time without issue, however I'm a bit confused regarding temperature readings.

The motherboard I'm using is an Asus M4A89GTD PRO.

The CPU is in custom loop and at full load I think I'm achieving either maximum temps of around 55 degrees or a maximum of 45 degrees. The reason for my confusion stems from several posts like the one below. Generally suggesting a 10 degree difference between the socket temperature and the on chip temperature.
link to above post

When running lm-sensors I get two possible values for the CPU core temp. I get "CPU Temperature" which is generally 10 degrees higher at load than the other option of "temp1" listed under the K10temp-pci-00c3 section.

AMD state the the critical temp for the 1055t is 62 degrees, which reading should I use to judge how close I am to the maximum safe temp of the CPU?

Thanks for any help, hope the above is clear!

Alec

Attached Thumbnails

 

I will try to keep this simple & clear.

In my opinion, none of them is the 'core temp'. The actual issue is a junction temperature limit set by the manufacturer, which is usually 150 degrees C. Silicon cab get much hotter (perhaps 1100 degrees when being doped) but the naturally uneven internal heating imposes the lower limit. Otherwise, it cracks.

Somewhere near those millions of transistors is a big (by comparison) lump of a temperature sensor. As the outside of the package is cooled, there is a temperature gradient all the way into the core. It's a bit like having the temperature sensor for your house central heating in the porch.

Bottom line - 45/55 is fine; Over 75 is bad news and the device may become unstable; I regard this as a limit; 105 is a generally accepted upper limit and temperatures above that are likely to be caused by an excess of 150C in the core, which usually means permanent damage.

1 members found this post helpful.

Thanks for the info, the main reason for my enquiry was to establish which temperature sensors I should monitor to make sure I'm not getting near the thermal throttling limit of the CPU. The AMD spec sheet states 62 degrees as its max temp, so I assumed (maybe incorrectly) that this would be the temperature at which the CPU was throttled back to prevent damage.

I wish to push the processor a little more so trying to establish which value the cpu itself would use for throttling (or if one exists) is my main goal.

When I'm not using the machine, it runs BOINC simulations which take all 6 cores to 100% load. If I am to continue doing this, I want to make sure I'm under the 62 degree value as prolonged running at such temperatures have been suggested to shorten the life of CPU's. This one is going strong at 5 years old and until AMD pull their finger out and make a true 8 core desktop CPU, I don't want to upgrade!

I think the values you expressed sound like sensible rules of thumb, but with my extended runs at 100%, I'd like to play safe.

I am going to try installing Windows on a spare drive and see what its sensors pick up as the core temp. Perhaps that might give me a better idea of which sensor to use.

Thanks

Alec

You have 2 sensor readings; one is read from the mobo's io sensor (multiple sensors) and the other is from the cpu. Now the problem is reading their output correctly.

The sensor labeled "Cpu Temperature" is the cpu temp read by the mobo's sensor (its either under the cpu or read from the cpu internally or something). It is used by the BIOS to change fan speeds, shut down in case of emergency etc. Some sensors dont really provide accurate values until ~40c or so.
The other sensor, k10temp uses the cpu's internal temperature reportings. The problem with it is that the provided values arent necessarily real temperature values, but are used internally by the cpu for automatic speed throttling etc.

Besides the whole cpu temperature thing is relative since it really the location of the temperature sensor(s) affects readings.

If you want to see the real cpu speeds, use cpufreq-aperf:

"modprobe msr" is needed only first time. Alternatively "sudo cpufreq-aperf -o" will do a single reading instead of continuous reporting. It reports the actual cpu speed instead of the currrent speedstep (which may or may not be honored effectively).

1 members found this post helpful.

Thanks gradinaruvasile, your differentiation between sensor readings was what I was looking to confirm.

I've checked this assertion by putting the system under load for short periods of time and noted that the on chip 'temp1' sensor reading reacts quickly. However, the other reading reacts much slower. The slower reaction speed indicate a sensor on the motherboard socket rather than on the die.

I've also confirmed the 'on chip' sensor under Windows. I used the coretemp utility and prime95 to find the sensor output, then ran the same test under linux using mprime and the sensors readings were identical (a max of 43 degrees @ 3.8Ghz with ambient temperature of around 20 degrees).

The experience also confirmed that I won't be booting into Windows again on my ancient IDE hard drive for a long time, talk about slow!

Unfortunately, due to overclocking of my cpu, the 'cpufreq' command cannot get the data its looking for, I've used the following commands before with pretty good results on AMD and intel machines.Everyone's take on this confusing issue is very helpful!

Thanks again

Alec