Hi all,
I am trying to read the CPU temperature (and case temperature if possible) on an Intel Atom 330. According to the processor datasheet, the CPU has an on-die digital thermal sensor which I verified with CPUID.
The problem is that IA32_THERM_STATUS MSR register, which is used for getting the DTS reading, indicates that the temperature reading that I get from the DTS is not valid (i.e. bit 31 is not set). Actually, the IA32_THERM_STATUS MSR has a value of 0x85F0000.
I also read the IA32_THERM_INTERRUPT MSR register and I found it to have a value of 0. That means that the value in both the temperature thresholds (Threshold #1 Value, Threshold #2 Value) is zero. I suspected that maybe my DTS reading is indicated as invalid because it is above the temperature thresholds limits which are both zero.
I’ve set a value to both temperature thresholds equal to Tjmax but I still get an “invalid reading” indication when reading the IA32_THERM_STATUS register.
Something last: it’s mentioned in the processor datasheet that the DTS will output a temperature relative to the maximum supported operating temperature of the processor (Tjmax).
Because it’s not implicitly mentioned, is it safe to assume that Tjmax is equal to the catastrophic processor temperature which is 125°C (it is mentioned that the processor will stop all execution when the junction temperature exceeds approximately 125°C)?
Do you guys have any idea what could be wrong?
Thanks in advance.
UPDATE: When I set the Threshold #1 and Threshold #2 values to 125°C, the bits (Thermal Threshold #1 and Thermal Threshold #2 bits) that indicate if the actual temperature is currently higher than or equal to the value set for each threshold respectively, are set. That means that it is indicated that currently the temperature is above 125°C!
Reading CPU temp on an Intel Atom 330
-
Rudster816
- Member

- Posts: 141
- Joined: Thu Jun 17, 2010 2:36 am
Re: Reading CPU temp on an Intel Atom 330
The DTS is rather tricky. The first wrench Intel throws at you is that they don't give you the TJ Max. IIRC, they don't even guarantee that the TJ Max will be the same between CPU's with the exact same Family\Model\Revision\etc, so its never safe to assume any given TJ Max. I've seen the DTS readings on CPU's that have "gone past" zero (I dont see a place where Intel explicitly says the RO value is unsigned, but since its 7 bits, I believe it is) without any frequency throttling. I even have personal experience with my CPU's core temps going above TJ Max with no throttling, so relying on a throttle\shutdown temp is definitely not reliable.
Download this if you're on Windows
http://www.techpowerup.com/realtemp/
This is the most accurate DTS software that I've yet to encounter, as well as a lot of other overclockers\benchers. Check to see what readings it gives you (if any). There might just be a problem with your CPU's DTS, they aren't the most reliable things.
On a side note, unless you're looking to report the core's (each CPU core has it's own DTS) temp to the user, there really isn't a point to reading the value unless you want to do rather sophisticated thermal management. Strictly from a CPU safety standpoint, they can "take care of themselves" without you configuring anything. That being said, manual thermal management could be very beneficial in a server like environment.
One more thing, geting none DTS temps is chipset dependent. Typically you can get readings from the Super I\O chips on motherboards. Problem is that pretty much all the datasheets are under NDA's, but here is a project that should prove useful if you want to program these chips
http://openhardwaremonitor.org/
Download this if you're on Windows
http://www.techpowerup.com/realtemp/
This is the most accurate DTS software that I've yet to encounter, as well as a lot of other overclockers\benchers. Check to see what readings it gives you (if any). There might just be a problem with your CPU's DTS, they aren't the most reliable things.
On a side note, unless you're looking to report the core's (each CPU core has it's own DTS) temp to the user, there really isn't a point to reading the value unless you want to do rather sophisticated thermal management. Strictly from a CPU safety standpoint, they can "take care of themselves" without you configuring anything. That being said, manual thermal management could be very beneficial in a server like environment.
One more thing, geting none DTS temps is chipset dependent. Typically you can get readings from the Super I\O chips on motherboards. Problem is that pretty much all the datasheets are under NDA's, but here is a project that should prove useful if you want to program these chips
http://openhardwaremonitor.org/
Re: Reading CPU temp on an Intel Atom 330
Thanks for your reply.
1) “Unless specified otherwise, all specifications for the processor are at TJ = 90°C.”
2) The processor will stop all execution when the junction temperature exceeds approximately 125°C.
The catastrophic processor temperature is 125°C (maximum)
What you would consider as the Tj Max?
AFAIK. the super I\O chip take the reading from the DTS, isn't that right? If not, it uses an even less accurate sensor I assume.
Btw, does anyone know if ACPI uses DTS as well for taking the temp reading or not? By looking at APCI spec, I got the impression that it is much easier to get an accurate reading from there (_TMP (Temperature) Thermal Object).
Thanks!
Well, at least for my processor I found 2 things in the datasheet:Rudster816 wrote: The DTS is rather tricky. The first wrench Intel throws at you is that they don't give you the TJ Max. IIRC, they don't even guarantee that the TJ Max will be the same between CPU's with the exact same Family\Model\Revision\etc, so its never safe to assume any given TJ Max.
1) “Unless specified otherwise, all specifications for the processor are at TJ = 90°C.”
2) The processor will stop all execution when the junction temperature exceeds approximately 125°C.
The catastrophic processor temperature is 125°C (maximum)
What you would consider as the Tj Max?
I see...do you know have any guess on why my readings are indicated as invalid? What I mean is: what causing a reading to be noted as invalid? I assured that the temp is not below the threshold levels but the reading is still flaged as invalid.Rudster816 wrote: I've seen the DTS readings on CPU's that have "gone past" zero (I dont see a place where Intel explicitly says the RO value is unsigned, but since its 7 bits, I believe it is) without any frequency throttling. I even have personal experience with my CPU's core temps going above TJ Max with no throttling, so relying on a throttle\shutdown temp is definitely not reliable.
That seems to be a very good program indeed. Do you happen to know a similarly good program for Linux?Rudster816 wrote: Download this if you're on Windows
http://www.techpowerup.com/realtemp/
I know that it is not to accurate but I just wanted to be able to take some temp measurements for testing purposes (for my own OS).Rudster816 wrote: On a side note, unless you're looking to report the core's (each CPU core has it's own DTS) temp to the user, there really isn't a point to reading the value unless you want to do rather sophisticated thermal management. Strictly from a CPU safety standpoint, they can "take care of themselves" without you configuring anything. That being said, manual thermal management could be very beneficial in a server like environment.
One more thing, geting none DTS temps is chipset dependent. Typically you can get readings from the Super I\O chips on motherboards. Problem is that pretty much all the datasheets are under NDA's, but here is a project that should prove useful if you want to program these chips
http://openhardwaremonitor.org/
AFAIK. the super I\O chip take the reading from the DTS, isn't that right? If not, it uses an even less accurate sensor I assume.
Btw, does anyone know if ACPI uses DTS as well for taking the temp reading or not? By looking at APCI spec, I got the impression that it is much easier to get an accurate reading from there (_TMP (Temperature) Thermal Object).
Thanks!
Re: Reading CPU temp on an Intel Atom 330
> Do you happen to know a similarly good program for Linux?
The only option that i know under linux:
http://www.lm-sensors.org
it can be easily found in many distros.
At first you run sensors-detect and answer it's questions, then you can use userspace app sensors to read what kernel modules have measured.
> I got the impression that it is much easier to get an accurate reading from ACPI
as for me, on most hardware i received too less information from ACPI, only 1-2 parameters and they didn't even had useful names.
talking to superIO/MSRs seems a better way nowadays. Although, i'm always using rather old hardware, maybe on recent machines ACPI is full of information.
The only option that i know under linux:
http://www.lm-sensors.org
it can be easily found in many distros.
At first you run sensors-detect and answer it's questions, then you can use userspace app sensors to read what kernel modules have measured.
> I got the impression that it is much easier to get an accurate reading from ACPI
as for me, on most hardware i received too less information from ACPI, only 1-2 parameters and they didn't even had useful names.
talking to superIO/MSRs seems a better way nowadays. Although, i'm always using rather old hardware, maybe on recent machines ACPI is full of information.
Last edited by Nable on Mon Mar 26, 2012 9:07 am, edited 1 time in total.
Re: Reading CPU temp on an Intel Atom 330
Depending on what exactly you are looking for (which I am guessing at), you might want to have a look at ksysguard (KDE-based). Together with ksysguardd (sp!) and an SSH-connection, you can visualize the CPU temp. of remote (Linux) boxes, too.limp wrote:Do you happen to know a similarly good program for Linux?
Every good solution is obvious once you've found it.

