Question regarding CPU temp reporting

[ACHU]

Hi, We used BurnInTest to test systems, but would like to know some more details about the event log. We have the CPU temp recorded every 10 minutes. The event logs will show current and max CPU temp. (Please refer to "event log.jpg")

1. Am I right that the max CPU temp is the highest CPU temp recorded starting from the very beginning till when the event log is, rather than the highest temp between 2 event logs?
2. If so, what's the interval that BurnInTest would record the max CPU temp?
3. Can we adjust the interval? If yes, how?

[Ian (PassMark)]

The maximum temperature reported is from the start of a test run (not from when BurnInTest is started and not between 2 interim result reports). The temperature is monitored every x seconds, where x is configured in Preferences->Temp / Battery, "sample every x seconds". The maximum temperature is the highest temperature sample from the start of the test run. The interval where the maximum temperature is measured is not configurable and is always from the start of the test run.

In saying this, if you needed to measure the maximum temperature for just a specific test period, then you could write a BurnInTest script and have temperature monitoring turned off for the tests you don't want to monitor, and just turn on temperature monitoring for the period of testing you were interested in.

[ACHU]

The interval where the maximum temperature is measured is not configurable

=> What's the default interval for the maximum temp. measured?

[Ian (PassMark)]

The default interval for temperature monitoring is 1 second.

The default interval from which the highest temperature is taken is from the start of the current test run until the current reporting time.

[ACHU]

So, is it possible that the event log showed the CPU max was 98C, but the CPU might actually reached 100C if the 100C happened between the 2 seconds monitored?

[Ian (PassMark)]

If temperature logging was configured to sample the temperature every 1 second (for example), as per the default, then it is possible that the CPU temperature was higher between the 1st and 2nd temperature samples. e.g. at 1 second it could have been 98C, and at 2 seconds it could have been 98C, but at 1.5 seconds it could have been 100C. The high temperature is based on the BurnInTest samples at 1 second and 2 seconds, i.e. 98C.

In this example, a reading is taken every 1 second and if the temperature sample is hovering around 98C to 100C, then chances are that you are going to get a sample of 100C (unless it really was a one off instantaneous reading of 100C).

[ACHU]

Am I right to state this:

"Since the interval of the max CPU temp monitoring can't be configured, the CPU max is monitored every second. If CPU reaches a higher temp within one second, this temperature won't be recorded in the log."

[David (PassMark)]

Any CPU with a heatsink will have a fairly large thermal mass, but have a limited amount of power input during a 1 second period. So is should be near impossible to move the temperature dramatically during a 1 second period.

This doesn't exclude the possibility that some CPUs don't measure their own temperatures very accurately however. e.g. two reads 1ms apart might return a slightly different value.

[ACHU]

Here's the scenario. Intel tells us that for certain CPU when its Tj reaches 100C, the CPU will throttle. We ran BurnInTest and the log file showed CPU max is 98C. However, it still throttled. We'd like to know whether it's because BurnInTest didn't record the "ever-happened 100C".

[Ian (PassMark)]

Am I right to state this:

"Since the interval of the max CPU temp monitoring can't be configured, the CPU max is monitored every second. If CPU reaches a higher temp within one second, this temperature won't be recorded in the log."

Further to David's comment.

Not quite, the CPU max is not monitored. The current CPU temperature is monitored, as mentioned this is configurable in preferences from 1 second upwards. The max CPU temperature is the highest CPU temperature sample from the start of the test to the current reporting time.

[Ian (PassMark)]

Here's the scenario. Intel tells us that for certain CPU when its Tj reaches 100C, the CPU will throttle. We ran BurnInTest and the log file showed CPU max is 98C. However, it still throttled. We'd like to know whether it's because BurnInTest didn't record the "ever-happened 100C".

It is possible, but I guess there are also other possible explanations.

[ACHU]

Not quite, the CPU max is not monitored. The current CPU temperature is monitored, as mentioned this is configurable in preferences from 1 second upwards. The max CPU temperature is the highest CPU temperature sample from the start of the test to the current reporting time.

=> Take the event log I posted for example. The current CPU temperature is monitored every 10 mins. The current CPU temperature of CPU 0 core 0 stays 84.0C from event 13 to event 14. However, the max CPU temperature of CPU 0 core 0 went up from 91.0C to 92.0C. That means, BurnInTest still kept track of the max CPU during this interim. Can you tell me how often the max CPU temperature is recorded in the software? (As you said, it's not cofigurable. But what's the default?")

It is possible, but I guess there are also other possible explanations.

=> What are the other possibilities?

[David (PassMark)]

We don't know exactly how Intel implemented the throttling.
It might be they implemented it in such a way to keep the temperature under 100C. That is to say they effectively throttle slightly in advance of 100C.

The max min and current are all updated at the same time.

[ACHU]

The max min and current are all updated at the same time.

=> The max and current are updated at the same time "in the log file". But BurnInTest should have recorded the max at a more frequent rate. Otherwise, the max CPU at event 14 shouldn't go up from 91.0C to 92.0C but the current CPU stays 84C at both event 13 and 14. Correct?

[orioon]

Most of the time the system vendor enables throttling before reaching 100C, on some system (mostly custom build) this is configurable.
On custom build systems you should expect that throttling won't happen before Intel's limit.
Is this a laptop or a desktop system? Vendor?

Often this is done to reduce temperature issues with other components like the HDD or the chipset.
Even an overheating GPU might cause the CPU to downclock, most typically when they're connected with the same heatpipe (happened to some Dell E6400/E6410 laptops I have access to)