Measuring CPU power usage

porina · April 17, 2019

Haven't done any serious testing in a while and with the impending arrival of Zen 2, I'm thinking an in depth look at actual performance-per-watt of representative current Intel and AMD CPUs would be a fun one to do. The difficulty will be measuring CPU power usage in a meaningful way. Currently I have a few ways to do it, but none are ideal.

One is the reported power from the CPU itself, via software. No idea how accurate that is, or if Intel and AMD use comparable measured. I doubt it somehow.

I can also measure at system level, but the problem is, it is at system level. I can measure either at the wall, or with certain PSUs like Corsair HXi series, they will report power on each rail. In this case, it would be difficult to isolate other system components so they would have to be somehow controlled as best possible. Measuring at the wall will also be impacted by PSU efficiency, but the HXi reporting does measure DC side so that could be ok.

The other way is to get more serious measuring kit. I think I've seen some put current clamps on the EPS12V cable for example, maybe some power from ATX12V also? This would still be impacted by VRM efficiency so again not entirely the CPU, but this might be about as close as I can realistically get. If I went this route I'd probably have to get a data logger to help with the readings.

Measuring after the VRM might not be impossible but it will probably exceed my motivation to implement. It will almost certainly get messy to do well and require some serious kit.

At the moment I'm thinking measuring the EPS power might be the most balanced way of doing it. It will include VRM impact but that is generally unavoidable to a user anyway. It would need testing of a sample of mobos with the same CPU to get a feel for how much impact that has. Hmm... this is starting to sound like more of a pain than I thought already.

Any other ideas welcome.

Mira Yurizaki · April 17, 2019

You can create a reasonably accurate current measuring setup using a shunt resistor circuit:

porina · April 18, 2019

7 hours ago, Mira Yurizaki said:

You can create a reasonably accurate current measuring setup using a shunt resistor circuit

Thanks for the suggestion. I know of the concept but I'm not sure it is the best choice in this case. I'd have to put the sense resistor into circuit. The EPS connector has three 12V wires for example and it would take care to not adversely affect flow if I were to common them up. At high currents the voltage drop would also have to be considered.

Mira Yurizaki · April 18, 2019

6 hours ago, porina said:

Thanks for the suggestion. I know of the concept but I'm not sure it is the best choice in this case. I'd have to put the sense resistor into circuit. The EPS connector has three 12V wires for example and it would take care to not adversely affect flow if I were to common them up. At high currents the voltage drop would also have to be considered.

If you need something accurate, I'd argue poking at the wires directly using something like a shunt is the best way to do it. Current clamps can be finnicky and considering it uses electromagnetic fields, it could be thrown off by anything else in the computer as there could be a lot of power flowing around elsewhere.