Proper Way to Measure CPU Temperature

[Stuka87]

Hey all,

So I am currently having an issue with a supplier and a server they supply us with as an OEM product. The chassis has some cooling issues. But they claim it is fine.

The difference comes down to how we are each measuring the temperatures.

I am using HWMonitor (Same as Anand uses) to get the temps. I am reading the average temps of the CPU's (Dual Xeon 5620's) in the upper 70's to lower 80's under full load. To me, this is too hot.

They are using a thermal probe on the heatsink. Which to me is a poor way to measure. As the heatsink is obviously going to be cooler than the CPU itself, because its job is to dissipate heat. They claim its temps are in the mid 50's.

They are claiming that using software to measure the temperature is inaccurate, and they claim Intel says not to rely on temps given by software.

So, my question is, which way is more correct? I can see the thermal probe being fine to measure differences provided the probe is always in the same place. But if software is not right, why is it so heavily used?

Thanks!

 

[Stuka87]

Wow, I thought for sure somebody here would know :/

 

[PlasmaBomb]

Get them to remount the heat sink...

If it isn't making good contact of course the CPU will be hot and the sink will be cool...

Your choice of monitoring apps is-
HWMonitor
Real Temp
CrystalCPUID
Core Temp
CPU Thermometer
Speedfan

Unfortunately Intel doesn't produce a monitor for their 5500 chipset, although if you check the MB vendor, they might. Tyan certainly support some boards.

A program such as TMonitor, which monitors for throttling may be of use to, although I don't know if it will work, there are others...

 

[Seero]

This issue was discussed before. Technically speaking, using thermal gun to measure temperature is accurate if you can shoot the beam on top of the die, but that is not practical. Those temp programs are not accurate because they are used to measure the distance away from TJMax, not temperature. People usually assume TJMax is a number, but it is not. We approximate the temp for that TJMax, let say that temp is approximately 100c celsius, if the Digital Thermal Sensors shows 40, then people think it is 100-40 = 60c celsius. Since the 100c TJMax is just an approximation, which makes the reading on the screen about the current temperature also an approximation as TJMax varies on each CPU.

The purpose of DTS is not for user to read, but for hardwares to react when things go wrong. The time it takes hardware to acquire info from DTS is around 15 us, but it takes much more than that to get to the screen (probably over 1 second), so it is not really accurate in terms of determining temperature.

Having said that, if you are to read distance to TJMax and try to determine rather it is too hot, then it is accurate, more accurate than to shooting gun at it. However, as long as the heatsink is hot, then it is working (radiating heat off CPU), and therefore no problems. I am not saying they are right, I am simply stating what their claim is.

BTW, if it is a server, do not remount the heatsink if you are not a technician as doing so means that you are responsible shall cooling fails. Simply test the system and if it fails, point your fingers towards the company that is responsible for it.

 

[Stuka87]

Thanks for responding.

The boards in question are SuperMicro, and as far as I know they don't provide any monitoring software. We used to use Tyan, but ended up with a large number of in field failures, although that was several years ago. The SuperMicro's have been solid though.

From what I have seen, the CPU does start to throttle. They also saw it throttle. To me, this means the temp is over what it should be. It should be able to run full boar without throttling due to CPU temps.

I am just trying to find a way to prove to them that monitoring via software is an approved way of doing so.

 

[Seero]

Thanks for responding.

The boards in question are SuperMicro, and as far as I know they don't provide any monitoring software. We used to use Tyan, but ended up with a large number of in field failures, although that was several years ago. The SuperMicro's have been solid though.

From what I have seen, the CPU does start to throttle. They also saw it throttle. To me, this means the temp is over what it should be. It should be able to run full boar without throttling due to CPU temps.

I am just trying to find a way to prove to them that monitoring via software is an approved way of doing so.

You are right in this case. Show them the delta(distance) from TJMax while the system is under full load. If it is around 5c, then it can be used as a prove of throttling.

 

[aigomorla]

CPUID.. or in another term Realtemp / coretemps, get a report off the diode in the cpu directly.

This temp is not a temp like thermo, but more of a tick clock that counts down how many C's you are away from thermal shutdown.

I say this because so far i have been able to break ambient on all the cpu's i have come across from I7 -> W35XX -> gulftowns.

The number u take @ cpuid is the most accurate number you need to make sure your cpu is safe.

It is this number that tells the bios in your board to throttle @ 90C, and thermal shutdown @ 11C.
This is the number which your boards' safety protocols revolve around.