PMD Accuracy

#1 · 2022-06-07, 15:03

Hi =)

I currently use the PMD every day to work and to determine CPU consumption. However, I have noticed a very large deviation the lower the CPU consumption. For example, in a test setup with a Core i7-12700K, Hwinfo reports a consumption of 3 watts (Idle). However, the PMD then reports a consumption of 32 watts to me. That is almost a deviation of 1000 percent! I once looked into the data sheet and noticed that you have a deviation of at least 2.2 amps as well as a maximum of 2.2 amps at 10 amp load. Also, is there any information on what the deviation is at 5 amps, 30 amps or 100 amps? This information would help me a lot. Beyond that, are there ways to increase the precision of the values? On average, I measure deviations of the CPU consumption between 15 and 30 percent. I find this to be a lot. Thank you very much for your time.

Translated with http://www.DeepL.com/Translator (free version)

#2 · 2022-06-08, 13:23

Quote from elmor on 2022-06-08, 13:23

Quote from PCGH_Dave on 2022-06-07, 15:03

Hi =)

I currently use the PMD every day to work and to determine CPU consumption. However, I have noticed a very large deviation the lower the CPU consumption. For example, in a test setup with a Core i7-12700K, Hwinfo reports a consumption of 3 watts (Idle). However, the PMD then reports a consumption of 32 watts to me. That is almost a deviation of 1000 percent! I once looked into the data sheet and noticed that you have a deviation of at least 2.2 amps as well as a maximum of 2.2 amps at 10 amp load. Also, is there any information on what the deviation is at 5 amps, 30 amps or 100 amps? This information would help me a lot. Beyond that, are there ways to increase the precision of the values? On average, I measure deviations of the CPU consumption between 15 and 30 percent. I find this to be a lot. Thank you very much for your time.

Translated with www.DeepL.com/Translator (free version)

Hi Dave,

There's a couple of items to consider here:

The typical measured current offset of PMD of +/- 0.3 A has more impact at lower currents. There's a built-in calibration function to improve this offset (see the datasheet https://1drv.ms/b/s!Atmpv-6qHr_61q039WXkPJHhVLyp9A?e=ejZRSf)

When comparing with HWInfo you need to consider the difference between input power and output power. At mid to high loads, typical motherboard VRM's are about 85% efficient. It depends on the board design which rails are powered from the EPS 8-pin input (ex Core, SOC/SA/IO and similar). Each of those will be measured together through PMD.

At light loads most VRM's are very inefficient meaning there will be a large difference in input vs output power (see https://www.reddit.com/r/overclocking/comments/bjn9qn/are_vrm_efficiencies_fixed/)

Offsets for current and voltage can be programmed through I2C. However there's no gain calibration available. It will be added on future PMD-USB

Max deviation for PMD at other currents:

5A 2.1A
10A 2.2A
30A 2.6A
60A 3.2A

Quote from PCGH_Dave on 2022-06-07, 15:03

Hi =)

I currently use the PMD every day to work and to determine CPU consumption. However, I have noticed a very large deviation the lower the CPU consumption. For example, in a test setup with a Core i7-12700K, Hwinfo reports a consumption of 3 watts (Idle). However, the PMD then reports a consumption of 32 watts to me. That is almost a deviation of 1000 percent! I once looked into the data sheet and noticed that you have a deviation of at least 2.2 amps as well as a maximum of 2.2 amps at 10 amp load. Also, is there any information on what the deviation is at 5 amps, 30 amps or 100 amps? This information would help me a lot. Beyond that, are there ways to increase the precision of the values? On average, I measure deviations of the CPU consumption between 15 and 30 percent. I find this to be a lot. Thank you very much for your time.

Translated with http://www.DeepL.com/Translator (free version)

Hi Dave,

There's a couple of items to consider here:

The typical measured current offset of PMD of +/- 0.3 A has more impact at lower currents. There's a built-in calibration function to improve this offset (see the datasheet https://1drv.ms/b/s!Atmpv-6qHr_61q039WXkPJHhVLyp9A?e=ejZRSf)
When comparing with HWInfo you need to consider the difference between input power and output power. At mid to high loads, typical motherboard VRM's are about 85% efficient. It depends on the board design which rails are powered from the EPS 8-pin input (ex Core, SOC/SA/IO and similar). Each of those will be measured together through PMD.
At light loads most VRM's are very inefficient meaning there will be a large difference in input vs output power (see https://www.reddit.com/r/overclocking/comments/bjn9qn/are_vrm_efficiencies_fixed/)

Offsets for current and voltage can be programmed through I2C. However there's no gain calibration available. It will be added on future PMD-USB

Max deviation for PMD at other currents:

5A 2.1A
10A 2.2A
30A 2.6A
60A 3.2A

PCGH_Dave has reacted to this post.

#3 · 2023-05-10, 06:26

On the topic of PMD accuracy, I notice that in the latest PMD-USB datasheet, the error specs have improved greatly from previous versions (for example 10A +/- 1.0A in latest version vs +/- 2.2A in previous version). Does the latest PMD-USB have design improvements in this regard over previous versions? If so it's a great improvement.

#4 · 2023-05-10, 10:59

Quote from shawn on 2023-05-10, 06:26

On the topic of PMD accuracy, I notice that in the latest PMD-USB datasheet, the error specs have improved greatly from previous versions (for example 10A +/- 1.0A in latest version vs +/- 2.2A in previous version). Does the latest PMD-USB have design improvements in this regard over previous versions? If so it's a great improvement.

Hello Shawn, the circuitry for power measurement is unchanged. However we've improved our QC and calibration steps to improve the accuracy starting with PMD-USB.