antiflas
User
 Fresh Boarder
| Posts: 4 |   | Karma: 0
|
Re:Under-Volting Fermi (GTX 560 ti) for HTPC - 2012/01/07 00:29
Ok, I think I have finished for now. I tested stability using OCCT 4.0 that find "errors" in their GPU test. It started to find errors at 0.900V, so I ended up using 0.9125V. Even though at 0.9V a couple of games did not exhibit any artifact or instability (Metro 2033 and BF3), I would better go with OCCT errors as an early indication that it is time to stop decreasing the voltage. For lower performance states (2D) I got it down to 0.8750V. Only time will tell if that is too low or not because I have no way to stress 2D performance. Following are the final numbers: I used Metro 2033 benchmark to test load conditions before and after. The results are as follow: Core Clock: Original: 880Mhz, Modded: 820Mhz Reduction: -6.82% Mem Clock: Original: 2100Mhz, Modded: 2000Mhz Reduction: -4.76% Core Volt: Original: 0.987V, Modded: 0.912V, Reduction: -7.60% Load Temps: Original: 82C, Modded: 76C, Reduction: -7.32% Metro 2033 FPS: Original: 47.2FPS, Modded: 45FPS, Reduction: -4.66% Load Power: Original: 223.9W, Modded: 201.8W, Reduction: -9.87% Comparing the reduction in performance (-4.66%) with the reduction in power consumption (9.87%), for each percentage reduced in performance I got double percentage reduced in power. That is better performance per watt and more efficient card. Next for me is to redo the cabling inside the cramped HTPC to have better ventilation. Hopping the numbers can help someone in a similar situation. Thanks for the software and this very informative forum that enable me to do this.
|