1

According to FAQ for Intel® Turbo Boost Technology, the turbo frequency is the same for all active cores. Intel's 8th and 9th Gen Datasheet, Volume 1 (PDF) also mentions that:

All active processor IA cores share the same frequency and voltage. In a multi-core processor, the highest frequency P-state requested among all active IA cores is selected.

So, it seems like all active cores of an Intel x86 CPU are in the same P-state at any instance of time, i.e. they are running at the very same frequency.

However, different hardware monitoring tools (e.g. HWiNFO64, Rivatuner Statistics Server) quite often display a different picture with different cores apparently running at different frequencies. Is this due to the fact that such tools simply cannot pool frequencies at different cores simultaneously and thus displaying values obtained at different instances of time or modern x86 CPUs can in fact run different active cores at different frequencies at the same time?

P.S. I was able to find out that starting from Haswell-EP per core p-states (PCPS) are supported, but I'm specifically interested in ordinary desktop CPUs and not server/HEDT products.

5
  • Going forward Intel will have products that have different core sizes. They just have not been released to market
    – Ramhound
    Aug 10, 2021 at 13:10
  • 1
  • 10th generation CPU have per-thread C-state, theoretically. The datasheet says that a power-saving transition occurs based on the per-core C-state resolved, but this state is the highest of the two threads (i.e. if one thread is in C1 and the other is in C6, the core C-state is C1). Similarly for the core-package C-state resolution. So yes, in theory, you have the machinery for per-thread C-state but in practice, it depends on the specific model. For 10th gen CPUs, per core C-state seems supported. Nothing prevents Intel from using a per-package C-state on low-end 10th gen models though. Aug 12, 2021 at 19:39
  • Note that if what you observe are fluctuations then that's probably a software artifact due to the unsynchronized sampling of the frequency on the different cores. If you see totally different frequencies (that may still fluctuate around their mean values) then it's per-code C-state power management. Aug 12, 2021 at 19:42

0

You must log in to answer this question.

Browse other questions tagged .