Power Consumption: Before And After
Our test system for power consumption measurements is a result of a cooperation with HAMEG (Rohde & Schwarz). It gives us the opportunity to present highly accurate values based on the use of a precise multi-channel oscilloscope with remote control.
We log every relevant channel, and all measurement values and graphs get saved to the scope. Our very nice clamp-on ammeters provide 100 mV/A, so we can easily derive current based on the measured voltages. The actual supply voltage is recorded and multiplied by the current later on.
For what it's worth, power consumption peaks do not always correspond to action happening on your screen, as you can see from the following sample video, which also demonstrates how the Radeon R9 290X's power consumption gradually falls, since the card isn't able to sustain consistent maximum clock rates.
Measurement Setup: | Non-contact DC-measurement at the PCIe slotNon-contact DC-measurement at the external PCIe power supply Direct voltage measurement 3.3 V / 12 V |
---|---|
Measurement Devices | Oscilloscope:HAMEG HMO 1024 four-channel DSO with storage function and Ethernet remote controlClamp-on ammeter adapter:HAMEG HZO50 (1 mA-30 A, 100 kHz, resolution: 1 mA), Voltcraft VC-511Probe:HAMEG HZ154 (1:1, 1:10), variety of additional adaptersDigital multimeter:Voltcraft VC-950 with multi-channel data logging function |
Bench Table: | Microcool Banchetto 101 |
Test System | AMD FX-8350 @ 4.5 GHz Corsair H100i closed-loop liquid cooler16 GB (2 x 8 GB) Corsair Vengeance DDR3-1866 Asus FX990 Sabertooth (modified PCIe attachment with current loops) |
Power Supply Unit | Corsair AX860i (modified outputs with measuring taps) |
Three Runs With AMD's Stock Thermal Compound
We start by running three successive test loops. The Radeon R9 290X hits its target temperature in the last quarter of the first loop, which is also where we see the highest power consumption (and peak performance).
The GPU temperature remains constant during the second loop, while power consumption falls considerably.
Although performance erodes again at the beginning of the third loop, we get consistent results from then on. Power consumption is reduced by another 10 W, and the peaks and troughs in this chart are clearly less intense.
Three Loops With Gelid's GX-Extreme
By using a higher-quality thermal paste, idle temperatures decrease considerably and we gain a slight increase in performance as power consumption rises by 20 W compared to the previous first loop.
Again, power consumption drops during the second loop as the GPU heats up. As soon as the Radeon R9 290X reaches its target temperature, Power Tune takes action and slams on the brakes, although the difference in average power consumption remains at 20 W.
And for the second time, loop number three yields a more consistent result at a lower performance level. The power consumption gap narrows to 12 W compared to this same run using AMD's stock thermal compound.
What have we seen so far? Well, we know that Power Tune takes action as soon as the GPU reaches its target temperature, but it takes some time after that before power consumption decreases significantly. How well is Power Tune working? A drop in power consumption of 11.5 and nine percent between the original and alternative thermal compounds is accompanied by a commensurate performance drop. And what about efficiency? Does replacing the interface material also yield an improvement there? We'll answer that question shortly.