Best way to measure power consumption of graphics card?

[waterboyserver]

Many user compare measurements at the wall via a kill-a-watt, then subtract further to isolate graphics card wattage. How accurate are the sensor readings in GPU-Z and HWiNFO64? Is using a multimeter any better (being careful with probes, not to short anything)?

[1714111288]

1714111288

[1714111288]

1714111288

[ARapalo]

Hardware measurements are either equal or more accurate than software methods, but never less accurate. Just like if you had a motherboard that had voltage measuring points for multimeter readouts, that's generally more trustworthy than CPUz or any other software.

But if software is all you have, then go with that. Temperature readings for the GPU is fairly accurate because it is taken from the video card's sensors, but CPU temp readings are based on TJ max, so softwares can only guess what the temp is for CPU readings.

[twoBitBasher]

You'll need one heck of a multimeter to measure the power draw of a gpu pulling ~170w. as far as I understand it can't be done with equipement that people usually have at their disposal.

One could use a kill-a-watt with dedicated PSU(just one GPU no mobo) and then factor in the PSU efficiency. This is even more easier if you have powered extenders.

[rograz]

use a kill-a-watt since the power usage at the wall is what's interesting, with 7xxx series cards you can test with just one card since they pull around 3-5w at idle with the screen turned off.

For other cards what I like to do is testing a rig with 1 card first to see how much it draws, then add a second card and check again to see how much the new card draws. Another reason for testing with more than just 1 card is that when you have "low loads" like 1 card + cpu/mobo you can get higher numbers than you should due to PSU efficiency normally falling the lower the load is (peak efficiency for a PSU is normally in the 30-70% load area)

[AzN1337c0d3r]

use a kill-a-watt since the power usage at the wall is what's interesting, with 7xxx series cards you can test with just one card since they pull around 3-5w at idle with the screen turned off.

For other cards what I like to do is testing a rig with 1 card first to see how much it draws, then add a second card and check again to see how much the new card draws. Another reason for testing with more than just 1 card is that when you have "low loads" like 1 card + cpu/mobo you can get higher numbers than you should due to PSU efficiency normally falling the lower the load is (peak efficiency for a PSU is normally in the 30-70% load area)

You should do the second part with 7970 as well. The ZeroCorePower only works if you only have DisplayPort monitors plugged in. Also ZeroCorePower seems to behavior quite oddly when you change voltage or clock settings.