Three evil letters.

T

D

P

This is power draw. This is heat dissipation. This is NOTHING.

Intel defines TDP as blah blah fucking blah and AMD uses blah sensors to blah blah I DON’T CARE.

What I’m going to be talking about today is something people can actually measure on their computer that gives a value in Watts. There is a thing called CPU Package Power that is supposedly how much power in watts the entire package of the CPU uses. That is apparently the goal of this value but as you will see in this article that it might not succeed in this venture.

On the testing block today is:

  • Intel Core i7 8700k
  • Gigabyte Z370 Aorus Gaming 7
  • AsRock Z370 Extreme 4

I’m not going to beat around the bush and give an eloquent intro to all of this, that went out with the first “BLAH” and “fucking”. I just want some simple things to be understood here so I’m going to present some data and then leave.

The testing was done using Asus RealBench. RealBench works by testing the following things.

  • GIMP Image Editing: Test focuses on single threaded and memory performance. (Uses up to SSE4.2 CPU extensions)
  • Handbrake h.264 video compression: Test focuses on multi-threaded CPU and cache performance. (Uses up to AVX CPU extensions)
  • OpenCL: Test focuses on GPU accelerated computing using Luxmark OpenCL rendering test. Runs entirely on all available GPUs! (Uses any GPU with OpenCL acceleration)
  • Heavy Multitasking: Test uses a combination of the above to simulate a heavy multitasking scenario that loads the entire system! (Uses up to AVX CPU extensions)

The phases are broken up like this in the charts:

  • The tiny lead-in section before the big hump is the GIMP single threaded testing.
  • The big hump is the Handbrake encoding.
  • The long section is Luxmark OpenCL on the GPU only.
  • The last hump is the multitasking which runs every previous test at the same time.

8700k With Stock Auto UEFI Settings

Using historical data that I’ve collected from the AsRock Z370 Extreme 4 motherboard I was able to come up with this comparison with my current Gigabyte motherboard.

Mind you these were running on the same exact settings, verified. Let’s get drastic and compare something better.

8700k 5GHz Core 3.7GHz Cache At 1.340V

WHAT?

Generally, when power consumption goes up as much as we are seeing here doesn’t temperature go up also? Well, yes usually it does. In this case we are seeing the result of the differences in how Gigabyte and AsRock, at least for this generation of CPUs, handles the reporting of SVID.

The following is a quote from the ‘8th-gen-core-family-datasheet-vol-1’

Platform Power Control

The processor supports Psys (Platform Power) to enhance processor power management. The Psys signal needs to be sourced from a compatible charger circuit and routed to the IMVP8 (voltage regulator). This signal will provide the total thermally relevant platform power consumption (processor and rest of platform) via SVID to the processor.

When the Psys signal is properly implemented, the system designer can utilize the package power control settings of PsysPL1/Tau, PsysPL2 and PsysPL3 for additional manageability to match the platform power delivery and platform thermal solution limitations for Intel® Turbo Boost Technology 2.0. The operation of the PsysPL1/tau, PsysPL2 and PsysPL3 is analogous to the processor power limits described in Section 5.1.3.1.

  • Platform Power Limit 1 (PsysPL1): A threshold for average platform power that will not be exceeded - recommend to set to equal platform thermal capability.
  • Platform Power Limit 2 (PsysPL2): A threshold that if exceeded, the PsysPL2 rapid power limiting algorithms will attempt to limit the spikes above PsysPL2.
  • Platform Power Limit 3 (PsysPL3): A threshold that if exceeded, the PsysPL3 rapid power limiting algorithms will attempt to limit the duty cycle of spikes above PsysPL3 by reactively limiting frequency.
  • PsysPL1 Tau: An averaging constant used for PsysPL1 exponential weighted moving average (EWMA) power calculation.
  • The Psys signal and associated power limits / Tau are optional for the system designer and disabled by default.
  • The Psys data will not include power consumption for charging.

The part here in bold is the important thing in this case. I fully suspect that SVID isn’t implemented correctly in one of these motherboards and I’m not sure if it’s the one reporting the lower numbers or the higher numbers. If you are asking yourself what SVID is here is a little refresher from the same document.

VCC Voltage Identification (VID)

The processor uses three signals for the Serial Voltage IDentification (SVID) interface to support automatic selection of voltages. The following table specifies the voltage level corresponding to the 8-bit VID value transmitted over serial VID. A ‘1’ in this table refers to a high voltage level and a ‘0’ refers to a low voltage level. If the voltage regulation circuit cannot supply the voltage that is requested, the voltage regulator should disable itself. VID signals are CMOS push/pull drivers. Refer Table 7-19 for the DC specifications for these signals. The VID codes will change due to temperature and/ or current load changes in order to minimize the power of the part. A voltage range is provided in Section 7.2. 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 processor IA core frequency may have different default VID settings. This is shown in the VID range values in Section 7.2. The processor provides the ability to operate while transitionally to an adjacent VID and its associated voltage. This will represent a DC shift in the loadline.

That is pretty much all I got for you for now. Please argue with me on Twitter about this!