Here are the stock voltages/frequencies for the i5-4690K

[kalmquist]

This information doesn't seem to be available on the web anywhere, so I did some measurements. Here they are:

   Ghz   volts
   0.8   0.768
   1.0   0.784
   1.2   0.800
   1.4   0.816
   1.6   0.832
   1.8   0.848
   2.0   0.880-
   2.1   0.880
   2.3   0.896
   2.5   0.912+
   2.7   0.944-
   2.9   0.944+
   3.1   0.960+
   3.3   0.992-
   3.5   1.008-
   3.7   1.040  (4 cores active)
   3.8   1.056+ (3 cores active)
   3.9   1.088- (1 or 2 cores active)

The first column is the CPU core frequency. Most of the frequencies listed are the ones that can be selected by the operating system. The last three frequencies are the ones selected by the chip, based on the number of active cores, when the operating system requests maximum frequency.

The second column gives the core voltage at the specified frequency. These voltages should be taken as approximations for three reasons:

1) The granularity of the measurements is 0.016 volts. A trailing + or - means that I sometimes measured the next higher or lower value. For example, "0.880-" means that some readings were 0.880 and some were 0.0864, so the actual value is probably just above 0.872.

2) The voltages were measured using the Nuvoton NCT 6791D chip on the motherboard, which reads about 1.5% above the actual voltage (at least when measuring the power supply output).

3) These numbers only apply to my chip; Intel may program slightly different numbers into different chips depending on the results of factory testing.

 

[altairauditore]

Remembered once again how much load voltage has decreased from Sandy Bridge:

i5-2400

1.6Ghz ~ 0.888V
3.2-3.4Ghz ~ 1.236V

 

[know of fence]

Page 94, under 7.4 in Intel's Hawell documentation under Vcc Voltage identification states

The VID codes will change due to temperatureand/or current load changes to minimize the power of the part. A voltage range is provided in the Voltage and Current Specifications section. The specifications are set so that one voltage regulator can operate with all supported frequencies. Individual processor VID values may be set during manufacturing so that two devices at the same core frequency may have different default VID settings. This is shown in
the VID range values in the Voltage and Current Specifications section.

To enable comparisons one ought to provide temperatures as well. There may even be differences between the different cores. Once all these parameters (Temp, Core, V measured under Load/Idle) are fixed, I certainly would be interested to hear by just how much the processors can differ from each other.

 

[BSim500]

As you said, each CPU is different so 2x i5-4690K's may differ up to +/-300mv. Eg: My i5-3570 (non K) can hit 4GHz at a stable 0.996v and idle's at 0.696v (at 1.6GHz). I've seen other lesser well binned i5-3570's need 1.2v & 0.8v respectively for same 4.0 & 1.6GHz freqs.

However thanks for posting this as it gives a good average guide and is the kind of detailed "geeky" stuff you don't get in general reviews. :thumbsup:

 

[ShintaiDK]

The information is not avaliable because easch CPU got its own VIDs.

 

[Ed1]

As you said, each CPU is different so 2x i5-4690K's may differ up to +/-300mv. Eg: My i5-3570 (non K) can hit 4GHz at a stable 0.996v and idle's at 0.696v (at 1.6GHz). I've seen other lesser well binned i5-3570's need 1.2v & 0.8v respectively for same 4.0 & 1.6GHz freqs.

However thanks for posting this as it gives a good average guide and is the kind of detailed "geeky" stuff you don't get in general reviews. :thumbsup:

Yes, but what was the stock values , I am guessing they were around 0.9xx and 1.0xx or so w/o messing with voltage and LLC .
problem with processor that does 0.8 idle but needs 1.2 at load, at least with Z77 your pretty much screwed trying to get those results .
I guess you could try under-volting with - offset for low idle and then using additional turbo voltage if your bios supports it to get proper load voltage .

 

[kalmquist]

What Intel refers to as Vcc is the voltage feeding into the integrated voltage regulator. The numbers I posted were the voltages provided by the integrated voltage regulator to a core operating at the specified frequency.

On my system, measured Vcc is either 1.792 or 1.776, but mostly 1.792. It doesn't seem to change based on system load. For reference, requesting 1.76 volts in the BIOS gives 1.776 measured, and requesting 1.74 volts gives 1.744 measured.

I didn't record temperature data, but rerunning the most demanding test (4 cores active, 3.7 Ghz, 1.040 volts) yields temperatures of 38 or 39 degrees C. I take the measurements after the load has been running for one second, and the CPU loads in this case are simple counting loops in order to let me verify the CPU frequency. Using a more typical program to generate load, and letting it run longer would probably result in higher temperatures.

 

[Ed1]

What Intel refers to as Vcc is the voltage feeding into the integrated voltage regulator. The numbers I posted were the voltages provided by the integrated voltage regulator to a core operating at the specified frequency.

On my system, measured Vcc is either 1.792 or 1.776, but mostly 1.792. It doesn't seem to change based on system load. For reference, requesting 1.76 volts in the BIOS gives 1.776 measured, and requesting 1.74 volts gives 1.744 measured.

I didn't record temperature data, but rerunning the most demanding test (4 cores active, 3.7 Ghz, 1.040 volts) yields temperatures of 38 or 39 degrees C. I take the measurements after the load has been running for one second, and the CPU loads in this case are simple counting loops in order to let me verify the CPU frequency. Using a more typical program to generate load, and letting it run longer would probably result in higher temperatures.

Don't you mean VTT for voltage to integrated voltage regulator .I think this means different things depending 3 or 4 series core processors .

 

[kalmquist]

Don't you mean VTT for voltage to integrated voltage regulator.

VTT is probably the same concept, but in the Haswell documentation, Intel uses the name VCC for the power to the internal voltage regulator. VSS is ground.

 

[Idontcare]

Nicely done and thank you for taking the time to share your results.

Data collection is always a learning process, you'll find out things you wish you had considered before you started (such as recording the operating temperature for each corresponding voltage/GHz entry) but that is part and parcel to the journey.

 

[know of fence]

What Intel refers to as Vcc is the voltage feeding into the integrated voltage regulator. The numbers I posted were the voltages provided by the integrated voltage regulator to a core operating at the specified frequency.

Agreed. I guess the only thing to take away from Intels the VID table, that it works in 10 mV increments starting with 0.5 V.
The maximum VID can differ between 1.6 and 1.86 V between samples (table 48. page 102) and may give an indication of how well a CPU OCs. But this maximum VID is not be the same as Vcc, and who knows if it even can be determined without sub zero cooling. VIDs displayed are likely maxium voltages over all 4 cores, we are just seeing the tip of the iceberg here. I could have sworn I've seen screenshots with Voltages for individual cores, but I can't find any.

On my system, measured Vcc is either 1.792 or 1.776, but mostly 1.792. It doesn't seem to change based on system load. For reference, requesting 1.76 volts in the BIOS gives 1.776 measured, and requesting 1.74 volts gives 1.744 measured.

I understand there is circuitry to stabilize load current, and before the VRs were moved on die to become FIVRs, some motherboards allowed to mess around with it. http://www.anandtech.com/show/2404/5

I didn't record temperature data, but rerunning the most demanding test (4 cores active, 3.7 Ghz, 1.040 volts) yields temperatures of 38 or 39 degrees C. I take the measurements after the load has been running for one second, and the CPU loads in this case are simple counting loops in order to let me verify the CPU frequency. Using a more typical program to generate load, and letting it run longer would probably result in higher temperatures.

This clears things up a bit, maybe you should speak to how you managed to hold the frequency in place (Did you disable EIST, set it as max multiplier or what?), after all isn't the goal to be able to reproduce those results. Also, do the Nuvoton NCT 6791D Values differ from CPU-Z or other software readouts at all?

 

[Ed1]

HWinfo64 shows VID for each, core and for me there all not the same at any one time .

I think on Haswell it also shows Vcore voltage for each core , I am on IB and only one Global Vcore is shown