I've recently taken some measurements from my Pentium D 915. Unfortunately I didn't have enough time to write an article about it, so I've got only the results for you with some short annotations in bad English.
Intel Pentium D 915 (2,8 GHz) VID=1,25V, C1 Stepping
GigaByte P35 DS4 with 6 phase CPU voltage regulator (Intersil ISL6317
1 GB Cellshock D9GMH
be quiet! E5 450W
Voltcraft VC920 (Vcore measurement)
Current Sensor (+12V CPU supply current measurement)
The ACS712 is not suitable to measure the real CPU supply current with values over 100 ampere. For this purpose a current clamp would be needed, to which I haven't got access at the moment. Thus I measured the current in the +12V supply line of the CPU VRM. The measured power consumption includes the CPU power consumption and the power loss in the voltage regulator.
Voltage regulators used for the CPU power supply have a typical efficiency at medium load of approx. 80% but up to 90%. The most important fact at this point is that VRM efficiency changes heavily with the CPU supply/VRM output current.
This graph by IR
shows some efficiency curves of different VRMs.
This graph makes also clear, that a higher phase count mustn't be better for efficiency. A lower phase count has less switching losses at light load, while a higher phase count has lower conduction losses at high load. Please keep this exemplary efficiency curves in mind!
Also the ratio between input and output voltage of a step down voltage regulator has an impact on its efficiency. The voltage on the 12V supply line can be considered as stable.
The comparably low efficiency of the VRM at low load can also be shown by these two measurements:
(1 core disabled in boot.ini, I hope it was really shut off!)
Real Vcore vs. Bios-Set Vcore (Drop & Droop) vs. Power Consumption at fixed frequency
Load was always tested with Prime95 large FFT setting.
Voltages are rounded to 3 decimal places
Real Vcore vs. Bios-Set Vcore (Drop & Droop) vs. Frequency vs. Power Consumption with fixed CPU supply voltage
Vdroop vs. CPU supply current
For everyone interested in Vdroop i recommend reading the Intel VRD 11 Specifications
I've done further testing with different multipliers, frequencies and voltages, but I don't think it's necessary to show them, until you are really interested in them...
I also measured the impact of CPU temperature on the CPU power consumption.
I'm sorry for not measuring the temperature directly at the CPU, but only using SpeedFan, which shows a temperature of about 10-15°C under the real value...
This measurement should be repeated with an Intel "Core" CPU with an accurate and internal temperature sensor with a fast response.
These curves should be similar in the negative temperature range until the so called "cold bug" is reached. If somebody is interested I could run some tests with my phase change cooler down to ca. -40°C.
Always keep your CPUs cool, undervolted and reasonably overclocked ;-)