I have a new desktop computer with Intel i7-12700 32GB RAM.
I am doing build code stuff, I use the sensors command to check CPU temperature, and I found most of the cores are @ 100C.
Is that normal?
Will CPU hardware itself control the frequency to fit the temperature?
update
I checked dmesg and found many logs as below:
mce: CPUxx: Package temperature above threshold, cpu clock throttled
It looks like the CPU control itself not higher than 100C.
Whether it’s normal for your system depends on a number of factors; however 100°C is on the high end for a desktop system and you should try to address that. Typically, that would involve improving the system’s cooling: the overall airflow in the case itself (assuming your CPU isn’t water-cooled), the CPU cooler and its interface to the CPU, etc.
In any case, your CPU won’t cook itself: it knows its limits, and it will throttle itself (reduce its frequency) if it needs to cool down. If that happens, you’ll see corresponding messages in the kernel logs (sudo dmesg).
Is that normal?
No, that's very high.
Will CPU hardware itself control the frequency to fit the temperature?
Yes, but you don't want that. Going into thermal throttling means you're getting less performance from the CPU than you could, and spending a long time at elevated temperature will shorten the CPU's life. You want to remove the heat, to avoid both of those problems. You should check on your cooling setup, making sure that the heatsink is in good thermal contact with the CPU (with an adequate, but not excessive, amount of thermal interface material), that the heatsink isn't choked with dust, that all fans are operational, and airflow through the case is adequate.
Oh, and check to see whether someone has compromised your system and installed a cryptocurrency miner that's pegging the CPU and/or GPU to 100% and making things so hot.
Same sensors command should tell that 100°C is the high temperature threshold eg:
Core 0: +37.0°C (high = +100.0°C, crit = +100.0°C)
Once it reaches the high threshold, in addition to active cooling (eg: fan), other method(s) available for the CPU will be used and among them lowering the CPU speed: so the CPU will reach 100°C and stay there due to the negative feedback loop.
If you're worried about the temperature, or more probably annoyed by the fan speed at its maximum, some Intel CPUs (at least Haswell ~ 2013, Skylake or later) along quite recent enough kernel (probably >= 5.4) have a setting to artificially add an offset to lower the temperature threshold (high) at which it will start lowering its speed. In practice that will mean the CPU will reach the newly chosen temperature and stay there, and the fan having less heat to evacuate, will spin less noisily. Of course build time will suffer.
I couldn't find the right subsystem symbolic link to reach it, so here is how I find it (location will probably vary on each system):
$ find /sys -name tcc_offset_degree_celsius 2>/dev/null
/sys/devices/pci0000:00/0000:00:04.0/tcc_offset_degree_celsius
Then as root:
echo 30 > /sys/devices/pci0000:00/0000:00:04.0/tcc_offset_degree_celsius
will trigger CPU slowing effects at 100-30=70°C. You would now see cores reach, and stay at, 70°C instead of 100°C.
sensors will not know about this except displaying the temperature is now 70°C (which is really the case), but I understand the kernel-related tool turbostat has knowledge about TCC and should be the preferred tool to set the temperature (using --TCC) if it supports the feature for the given CPU (probably meaning it knows what subtraction to do).
high = +100.0°C as the actual temperature, or something similar.
In the case of your specific CPU is not normal at all. Make sure that your paste is correctly applied to the CPU and that the heatsink is correctly installed. It must touch the plate surface of the CPU. Also check if your pc case has proper airflow. You can also run a htop(sudo install htop) to see if the usage of your CPU and if it correlates with the high temperatures.
Contrary to some other answers, this is normal.
According to Intel's Ark page for the 12700K, the Junction Temperature is rated at 100C. The processor has been set to run up to that temperature limit, and it's considered by Intel to be safe.
That said, it's not ideal because it means there probably is some bottleneck somewhere in removing heat from the processor.
Processors generally run at variable speeds and will often attempt to go as fast as possible until they hit either the max speed set in the bios or the max-rated temperature. Once they hit one of those limits, they will stop going faster and potentially may even slow down to keep from over-shooting the temperature limit. Ideally, they should hit max speed before max temp.
And supposedly keeping temps significantly lower can help improve the lifespan of parts (I believe it but I've also never seen rigorous data on this for modern systems)
There are a few things that can be done to stop this:
Just to emphasize this: even pre-built systems from major manufacturers can have issues. Gamers Nexus on Youtube occasionally purchases pre-builts from manufacturers anonymously to review their build quality. They've seen case fans installed backward, cables not fully plugged in, etc. They even had an advanced CPU cooler come with two fans on it facing in opposite directions, effectively canceling each other out.
It's not quite normal. You probably need to set PL1 and PL2 power values to something reasonable in your firmware (BIOS). Some motherboards ship with ridiculously high defaults which basically mean unlimited.
The CPU you have can draw a lot of power (above 200W IIRC) and it takes a huge cooler to manage that. Most PCs have a more average cooling solution and depend on soft controls to keep temperatures reasonable. That's what PL1 and PL2 settings do. Without proper values, the CPU will hit its critical temperature limit (100°C) and throttle itself. While this causes no immediate damage, it is apparently not good for long term stability. It also wastes a lot of power for a tiny performance improvement.
I don't know how much truth there is to the possible damage at 100°C. There are no widespread reports of early CPU failures that I know of. You can do nothing, find out, and let us know in a few years.
Or you can adjust the power limits. There's a lot of information about this on the web but the short version is: dial down PL1 until your long build jobs give you a lower stable temperature. I like 80°C but pick whatever you want. The amount of power you set will be what your heat sink can handle and your case fans move out of the way, at that temperature.
Then you can set PL2 somewhat higher than PL1, for short peaks. Think starting an application, displaying a web page, etc. How much higher depends on the termal mass of your heat sink, how long you set the tau parameter, how high you set PL1. Again, the goal is to keep temperature reasonable.
To test PL2, let the CPU idle for a short while then start some CPU intensive job and watch temperature and CPU frequency. You should see temperature rise and eventually frequency drop when PL1 kicks in. If the temperature doesn't rise too much before frequency drops, you're good. If it gets too high to your liking, reduce PL2 or tau.
All the above assumes your cooling system is working correctly. If you're hitting 100°C at low power values (e.g. 65W), you likely have a cooling problem.
having a high temperature on hardware (CPU/GPU) got noting to do with kernel it self it's about the task intensity of the program. compiling a code is a high cpu intensive task.
my recommendation is to basically get a better cpu cooler or better thermal paste how ever if you already have above mid-range cooling system make sure you screwd cooler tightly to the mother board and applied enough thermal paste. lastly make sure the case it self has good & correct airflow. you can always increase fan speed but it's noisy.
if you already did all of this i highly recommend to under-volt the cpu from bios. you can also under-clock cores by reducing clock speed if you want to extremely reduce temps but you lose performance.
yes your motherboard will reduce cores clock based on cpu thermal throttle settings in bios but remember you loose performance. i recommend to keep hardware temps below 85°C on max performance if you are planning for long term use.
cpupower. A few hundred mhz slower won't have a huge impact on performance but lowers temps substantially You could even set up a keyboard shortcut to instantly throttle for when unexpected loads arise.
For Intel 12th gen CPUs, this is not normal. The relevant specification (careful - PDF download) found at the Intel's website clearly states (with minor edits for brevity):
The PROCHOT# (processor hot) signal is asserted by the processor when the TCC [Thermal Control Circuit] is active. [...] When any DTS [Digital Thermal Sensor] temperature reaches the TCC activation temperature, the PROCHOT# signal will be asserted.
And then (emphasis mine):
With a properly designed and characterized thermal solution, it is anticipated that PROCHOT# will only be asserted for very short periods of time when running the most power intensive applications. The processor performance impact due to these brief periods of TCC activation is expected to be so minor that it would be immeasurable. However, an under-designed thermal solution that is not able to prevent excessive assertion of PROCHOT# in the anticipated ambient environment may:
- Cause a noticeable performance loss.
- Result in prolonged operation at or above the specified maximum junction temperature and affect the long-term reliability of the processor.
- May be incapable of cooling the processor even when the TCC is active continuously (in extreme situations).
IF all these measures fail, and there is a risk of physical damage to the product, another signal will go off:
[...] the package will automatically shut down when the silicon has reached an elevated temperature that risks physical damage to the product. At this point, the THRMTRIP# signal will go active.
Relevant conditions are not elaborated, but typically you could expect a "hard" shutdown as late as 115-125°C and as early as 105°C (lower limits are coming from my anecdotal experience). At any rate, temperatures above 125°C are generally seen as extreme, and are a part of industry-standard stress tests such as JEDEC HTOL [high temperature operating life] (pdf download!).
It is important to note that for modern CPUs, there is more to it than just temperatures. Intel tries to cool off long before reaching Tjmax (maximum junction temperature), and the throttling behavior is entirely centered around that temperature. In particular, when trying to go over it, a complex network of regulation of clocks, power limits, C-states and so on will be engaged. For Alder Lake desktop CPUs, Tjmax is 100°C, and this is the temperature referred to in the datasheet above.
By contrast, Ryzen 7000 series reaches Tjmax early and dances around it, boosting opportunistically and throttling as needed, relying on your thermal solution to do its job. Here is a promo/explanatory post from an AMD representative elaborating on this behavior. It also makes sense: larger temperature differences mean more efficient cooling. The main reason this approach is not widespread is because it becomes harder to control, with tiny margins for error. AMD is confident in their ability to keep temperatures under control, and time will tell whether they are right to do so. The reason I bring this up is because "high temperatures=bad" approach is currently being hotly contested: it is kind of like taking a corner in a race car, for a reasonable driver, there is no point in slowing down late - benefits are slim, and the price to pay is high. But you could go faster by taking more risks, which is what AMD does, and the question then becomes whether they can reliably handle it or not.
Either way, whether a given behavior is fine for a given product or not can and should be determined from the relevant design documents. And in your case, hitting Tjmax for a prolonged period of time is specifically described in the datasheet as an indication of an under-designed thermal solution.
No, that's very hot. Plenty of BIOSes offer temperature checks that stop at 100 degrees.
As other answers said, make sure your CPU is properly cooled; heatsink and fan positioned correctly, air can flow through with minimum obstruction from wires, etc.
There's also maintenance; dust will build up. I typically power off, unplug, dissemble and vacuum my computer at least once yearly. Normally, need to replace the thermal paste on the CPU as well.
To determine if your system is thermal throttling, check the CPU multiplier is close to your CPU expected frequency.
(On Windows, I would recommend CPU-Z. A CPU multiplier of 8 is 800MHz, so a 3.4GHz processor should have a multiplier of 34 when being heavily used.)
If you want to run cooler but quite a bit slower:-
Create a VM with a reduced number of CPU's and do your compiles in the VM.
I was forced to do a similar thing in a VM for a transcoding application which tried to run 16 threads at 100% each. This caused the temperature to rise above 100°C in less than 10 seconds!
Don't kill your new computer.
taskset to assign cores to a process. A VM is horrible overkill and will cause extra heat due to its own overhead.
Is it normal? That your computer gets this hot, that is not normal. That it slows down the clock speed and does other things to reduce the temperature in this situation, that is normal.
Does your computer get ventilation / access to cool air? Do the fans work properly? Are they effective? Is there anything blocking cold air from getting to the CPU? Does the temperature get measured correctly? (We know that some thermometer believes it is 100C, that doesn't mean your computer actually is that hot).
Take a thermometer and measure the temperature outside and inside your computer. If it's 40C outside, then move the computer to a cooler place.
Until 70ºC is hot, but acceptable. 90ºC or more is just critical. Most CPUs won't reach 100ºC, they shutdown automatically before that.
EDIT 1
Yes, it does damage, I didnt mean cpu dies imediatelly, but making it constant makes your cpu to die so much faster
