@crashtech Well I have not seen any Intel documents that are explicit other than to mention LFM transitions at package level C-States and cores running below the requested performance state for a short time when woken. For instance from the LGA-2011 data-sheet
From this it's not clear IMO that if all cores reside in C3 or higher whether that constitutes an LFM request. However I can provide some examples.
Depending what software you use to measure clocks it may either be done at an instant or averaged over the sampling period. If the clocks are constantly changing then the instantaneous measure isn't showing the whole picture.
Here's IVB with clocks at 4.6GHz, high performance with C3/C6 core and package states set and C1E disabled.
As you can see the clocks, depending at what time it was measured, shows either 1.6GHz (LFM) or 4.6GHz for the instantaneous measurements of HWiNFO64 and CPU-Z. My own software shows the average ratio's over time which in this example will be an average of clocks run at 1.6GHz and clocks run at 4.6GHz during that time.
With just core C-State C3 reporting enabled. Package states, C1E and C6/C7 core reporting disabled
Now lets try with setting BIOS to disable C1E, package states and reporting of core states C3/C6/C7. C7 in this case is not so important as it's not included in the ACPI tables.
Clocks are now reported at 4.6GHz with no sign of LFM clocks.
IVB is interesting in that it can clock below the LFM (16x) down to 8x.
Running fixed 8x multi with BIOS set to disable C1E, package states, and reporting of core states C3/C6/C7
Now can you guess what happens if we run like this with just say core C-State C3 reporting enabled? Well...
What happens now is that clocks will do the opposite and go up to the LFM! Pretty damning evidence really. It may not be as noticeable as the CPU lowers the clocks before lowering the voltage so it can happen a lot quicker than the other way around where increasing the clocks requires the voltage to be raised first. Notice that the average ratios are a little above 8.0.
I noticed you posted a Haswell CPU in your reply. These aren't the same as SNB/IVB in that they can be set to run C-States without the noticeable effect of LFM clocks. Here's my i7-4700MQ with Core C-States C3/C6/C7 reported and package level C6 set, C1E disabled.
As you can see maximum clocks can be had and because C3 or higher is reported it is possible to run the higher stepped turbos when running less than 3 cores. Turbo bins 34,34,35,36. Also in this case is does not matter what Windows plan is being used or what processor percentage is selected. Only throttling will lower clocks in the C0 state (active).
Anyway, hope that's what you were looking for.
From this it's not clear IMO that if all cores reside in C3 or higher whether that constitutes an LFM request. However I can provide some examples.
Depending what software you use to measure clocks it may either be done at an instant or averaged over the sampling period. If the clocks are constantly changing then the instantaneous measure isn't showing the whole picture.
Here's IVB with clocks at 4.6GHz, high performance with C3/C6 core and package states set and C1E disabled.
As you can see the clocks, depending at what time it was measured, shows either 1.6GHz (LFM) or 4.6GHz for the instantaneous measurements of HWiNFO64 and CPU-Z. My own software shows the average ratio's over time which in this example will be an average of clocks run at 1.6GHz and clocks run at 4.6GHz during that time.
With just core C-State C3 reporting enabled. Package states, C1E and C6/C7 core reporting disabled
Now lets try with setting BIOS to disable C1E, package states and reporting of core states C3/C6/C7. C7 in this case is not so important as it's not included in the ACPI tables.
Clocks are now reported at 4.6GHz with no sign of LFM clocks.
IVB is interesting in that it can clock below the LFM (16x) down to 8x.
Running fixed 8x multi with BIOS set to disable C1E, package states, and reporting of core states C3/C6/C7
Now can you guess what happens if we run like this with just say core C-State C3 reporting enabled? Well...
What happens now is that clocks will do the opposite and go up to the LFM! Pretty damning evidence really. It may not be as noticeable as the CPU lowers the clocks before lowering the voltage so it can happen a lot quicker than the other way around where increasing the clocks requires the voltage to be raised first. Notice that the average ratios are a little above 8.0.
I noticed you posted a Haswell CPU in your reply. These aren't the same as SNB/IVB in that they can be set to run C-States without the noticeable effect of LFM clocks. Here's my i7-4700MQ with Core C-States C3/C6/C7 reported and package level C6 set, C1E disabled.
As you can see maximum clocks can be had and because C3 or higher is reported it is possible to run the higher stepped turbos when running less than 3 cores. Turbo bins 34,34,35,36. Also in this case is does not matter what Windows plan is being used or what processor percentage is selected. Only throttling will lower clocks in the C0 state (active).
Anyway, hope that's what you were looking for.