There is a talk by Sophie Wilson (primary ARM architect)/ there is a google video available if you search, where she talks about how modern multicore CPU’s shut down a core to help disipate heat. She jokes that for the first time in history, people update their CPU’s to newer versions, only to have less enabled CPU’s than previously. Thermal dissipation is a new barrier in CPU design. Locking a process to a single core will cripple your performance on newer hardware. No OS scheduler today is even aware that CPU cores become suspended by hardware.
Would better heatsinks prevent this “thermal shutdown” of unused cores (it could throttle down a given “in use” core, but not shut it down completely as that would kill processes, crash the OS, etc.), since I assume it’s triggered by sensed overheating of the die?
If better heatsink would work, why do you think engineers would go with the much more complex solution of shutting down cores?
How is my question invalid, when this was the statement I was addressing? Unless Zenja’s statement is wrong, I don’t see how having a better heatsink wouldn’t help.
It would limit the problem, but it would not solve it completely. Otherwise, we would just have bigger/better heatsinks on computers and we would not have engineer taking the trouble of making it possible to shut down cores when they overheat.
Switching cores on and off as needed in relation to CPU usage is good for both thermal performance and power management. Extending battery life on portable machines is just as important as keeping the computer cool.
So there are more sides to the story.
1 Like
The processes won’t die if a core shut down, as CPU0 will relocate the processes to another core automatically, afaik.
Assuming an ideal case setup, with proper cooling, manufacturer’s specified heatsinks/fan, moderate room temperature (70 degrees), etc.
If you ran all cores on a given CPU at 100% (exactly; not 1% higher), would/could the CPU ever overheat (assuming the heatsink never came off, fan failed, heatsink grease degraded, etc.)? If so, how/why?
I tend to over-analyze everything when opportunity presents itself! 
What is the “proper cooling” for you? You can fit some expensive liquid cooling system into your tower or buy a decent cooler (stock heatsink is usually not that good, but costs less) and use some liquid metal compound. In these cases you probably won’t be able to throttle on 100% load, but this is possible only because the tower has plenty of space and you have some extra money to spend. You can’t fit a “proper cooling” into a laptop or smartphone.
If you use the manufacturer specified cooling system (usually the one shipped with the CPU, to make sure you don’t use something worse), you would probably stil hit this. Disabling or slowing down cores is now something CPU manufacturers do as part of the normal run of the CPU (in the early days it was only a security feature to avoid hardware damage).
In a marketing twist, Intel calls it “Turbo Boost”. When there is a single core used by the OS, they can make it run very fast (past the 3GHz mark). But when there are multiple cores running, they generate too much heat, so they slow down the frequency to compensate.
So, the CPU won’t overheat, but it will adjust its own speed and/or stop cores to stay inside the allowed temperature range.
Any app recommended to monitorize the heat and cooling?
1 Like