EDIT:
Aida numbers were wrong. Aida power consumption is core voltage x VRM current. It is possible that Aida doesn't take power over PCIe into account, or is just simply rough approximation.
So, to clear things up I did some testing with system power consumption using Kill-A-Watt.
I disabled in BIOS everything I don't need. Turned off QnQ so it doesn't screw with my results by changing CPU speeds widely.
Here is a graph that shows system power consumption and GPU VRM current (from AIDA):
First things first.
Mining is less intensive than furmark by a lot. Not only system power usage is 30-45Watts lower, but I could cut another 0,1Volts off core.
Fan speed have a big impact on power consumption! It probably have a lot to do with VRM temps which go from 60'C when mining @0,95V to 95'C during furmark @1,25V. Not sure why the difference between quiet and uber custom fan profile is much wider during furmark than mining.
Idle power consumption was 93Watts.
mining system power consumption @0,95V was 179(Uber)/ 183(Quiet) Watts
mining system power consumption @1,25V was 249(Uber)/ 258(Quiet) Watts
Furmark power consumption @1,05V was 216(Uber)/ 230(Quiet) Watts
Furmark power consumption @1,25V was 277(Uber)/ 303(Quiet) Watts
Now isolating GPU power consumption by taking out Idle power consumption:
mining GPU power consumption @0,95V was 86(Uber)/ 90(Quiet) Watts
mining GPU power consumption @1,25V was 156(Uber)/ 165(Quiet) Watts
Furmark GPU power consumption @1,05V was 123(Uber)/ 137(Quiet) Watts
Furmark GPU power consumption @1,25V was 184(Uber)/ 210(Quiet) Watts
I have 82% efficiency (or lower at that load) Chieftec CFT-750 watt power supply:
mining GPU power consumption @0,95V was 70,52(Uber)/ 73,8(Quiet) Watts
mining GPUpower consumption @1,25V was 127,9(Uber)/ 135,3(Quiet) Watts
Furmark GPU power consumption @1,05V was 100,86(Uber)/ 109,6(Quiet) Watts
Furmark GPU power consumption @1,25V was 147,2(Uber)/ 172,2(Quiet) Watts
Comparing these values to aida current readings:
mining GPU power consumption @0,95V was 70,52(Uber)/ 73,8(Quiet) Watts AIDA(Uber):55AMPS(52,5Watts)
mining GPUpower consumption @1,25V was 127,9(Uber)/ 135,3(Quiet) Watts AIDA(Uber):92AMPS(115Watts)
Furmark GPU power consumption @1,05V was 100,86(Uber)/ 109,6(Quiet) Watts AIDA(Uber):83AMPS(78,87 Watts)
Furmark GPU power consumption @1,25V was 147,2(Uber)/ 172,2(Quiet) Watts AIDA(Uber):118AMPS(147,5 Watts)
While not most accurate, these results are very close to each other (Aida vs Kill-A-Watt).
Decreasing core voltage by 24%(16% for furmark) decreased:
mining power consumption by
44,9(Uber)/
45,5(Quiet)%
Furmark power consumption by 31,5(Uber)/36,3(Quiet)%
Fan speed BY A LOT!
Test sample(Golden 7870 of Golden 7870's) Club3D HD7870 Coolstream:
Shoutout to Club3D for not voltage-locking thier cards!*
*These test were not sponsored by Club3D
I think voltage tuning is very interesting aspect of tweaking hardware. With proper tools this could be looked at closer and could possibly lead to many distressing conclusions. I think this matter deserves a separate thread, not only to stop derailing, but it completely changes the nature of a card.
Also, this little card took 172 Watts?! Holy moly, that is a lot for such small heatsink, VRM, ETC.
If the card takes 70Watts, can I safely unplug 2x6pin? Can something bad happen?
TL,DR: Bunch of numbers on how golden sample of golden samples is close to becoming Perpetuum-mobile heat(and crypto-coin) generator .