![]() ![]() Then, loading it with `make -j4` or `make -j16` on a 4-core causes substantially less than full load on all cores since there are always moments when the dependency chain isn't filled enough with runnable stuff. All these kernel turnaround ease off on the CPU. Then there is lots of userland/kernel space switched from all the system calls, both for disk blocks but also for tons of stat(2) calls, forks, execs, waits etc. These causes gaps in CPU load even if the block is already resident. Doesn't matter whether it is locks that are explicit to your program, whether it is cache sychronization done by hardware or whether it is VM page tables and TLB entries maintained by the OS.įirst of all there is disk involved. They need to be as separate as possible, any trace of synchronization will ease the load on the CPU. The only way to figure out how much silicon you use is to run a piece of code and look at the CPU temperature.Įdited to add: you also want individual processes, not threads on the different processors. You only use a small part of the CPU's silicon and the other parts will remain unused and hence untested - and they do not produce heat. ![]() You don't get a lot of stress from just tiny pieces of code. Thanks for the info!Īs has been explained in this thread before: I'll compare the heat of the CPU after both tests for about an hour and see if the temperature at all differs, since that's all I care about. Should probably be using the term "get hot" test. Sorry if I'm misusing the word "stress" here. I am under the impression that getting them to run at full load will get them the hottest. I only care about how hot it can get (testing thermal paste for example). Well, I basically wanted to make a small program that stresses all cores of the CPU while outputting the temps. Prime95 has excellent internal consistency checks, and in practice it is the first program to report errors. In a simple program with no checking on anything you'll never know, and you aren't likely to crash the OS right away. Presumably you also want to know when it starts giving out nonsense. ![]() In addition, it is pretty useless to just stress the CPU. There is a reason for it, and they look very closely, and they have been looking very closely for many years. There are tens of thousands of sports overclockers who constantly look for harder and harder test programs. In practice, prime95/mprime (and larger suites using prime95 such as OCCT) heat up the CPU most. All the non-hands-on theoretical thought given here and elsewhere goes right out the window once you actually look at what is happening. Whatever has the highest temperature is most stressful at least for that definition of stress. There is only one way to find out how stressful a given program is: ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |