How to measuring the power consumption of CPU and memory

[waldar]

Hello,

I would like to quantify the energy consumption of the different subsystems of a PC (processor, memory, PIC, pCIe, etc.). The last two are relatively easy since I use raiser cards and intercept the power lines. But I have difficulty in (1) accurately measuring the energy consumption of the processor, and (2) separating the energy consumption of the processor from the memory. I use a D2461 Siemens/Fujitsu motherboard and it is not entirely clear to me how these subsystems are supplied with power. There are two Molex connectors (4-pole and 25-pole), but I am not able from which the meory draws its current. Moreover, it seems to me that the processor draws current from both connectors. Does anyone have some input there?

I appreciate.

Waldar

 

[duxducis]

the easiest is to use external whole system device to measure, and use program that can load CPU 100% internally, Then also you can put load on ram (internal ram to ram read write) that can give you some numbers but not precise
4-pin auxiliary connector is for CPU to draw extra power from
here few links
http://pinouts.ru/Power/atxpower_pinout.shtml

http://www.hardwaresecrets.com/printpage/Understanding-All-Voltage-Configurations-from-the-Motherboard/995

 

[hunter315]

The CPU is fairly easy to do, that 4 or 8 pin connector that goes to the motherboard provides all the power to the CPU so you can use a multimeter to check the exact voltage at the board and a current clamp around the yellow wires to measure the current, multiply those two numbers and you have the CPU power.
http://www.fluke.com/fluke/usen/accessories/current-clamps/i1010.htm?PID=56281

Memory is a bit more challenging, it feeds off the 3.3 V rail of the main 24 pin connector. You could work with adding and removing sticks of memory and see how the current draw down the 3.3 V lines change and then figure out the power per stick, but if you factor in about 5 W per stick its more than enough.

 

[waldar]

Dear hunter315,

Thanks. I tested the power consumption of the CPU under different loads and measured the current fluctuations on the 12V 4-pole as well as 12V and 5V of the 24-pole. It seems to me that the CPU takes current through all these lines. I am sort of frustrated because the motherboard manufacturer does not give me a hint how the CPU is powered. As to the MEM, I have tried what you suggested, but the change in power consumption is not applicable.

 

[duxducis]

the CPU does not work alone and such you could newer accurately measure its power draw, we can get close but, why are you asking? if CPU rated 125W its not going to consume more then that. Only engineers in INTEL can tell exactly how much,
you cant do it motherboard, but overall approximate

 

[waldar]

The idea is to estimate the power consumption of a "workload" and breakdown this cost in terms of the cost of the different subsystem: processor, mem, etc. That is why I need to measure the power consumption of these components while the workload is running.

 

[duxducis]

you are for the big long quest then i would rate 1 CPU 2 chipset 3 memory by watt usage.
Don't forget about efficiency as well, and energy loss from conversion (PSU+Motherboard)
Same CPU+memory plugged into different motherboard can account for ~50watt loss, based on that MB efficiency.

 

[hunter315]

The CPU only recieves power down the 4 pin connector, but it doesn't mean that a "CPU only" workload will only increase the CPU power draw. In order to feed the CPU with data the motherboard had to do more work which in turn increases the power draw from the 24 pin connector.


Unfortunately, nothing in a computer is independent. A memory heavy workload will also increase CPU power draw, a CPU heavy work load increases the load on the HDDs, memory, and the buses on the motherboard.

Since all of the subsystems are intertwined its impossible to break down the cost in terms of each subsystem for each work load, you best bet is to monitor overall system power draw in each workload with different hardware configurations. Try to find a PSU with a fairly flat efficiency curve to minimize your accuracy loss as the power loads change.