A New Level of Power Testing
Maxwell is an impressively efficient architecture and we saw that with our original GTX 750 Ti review and then again with the GTX 980 and GTX 970 release. Not only was the GTX 980 able to outperform the likes of the Radeon R9 290X but it was able to do with substantially less power. While power consumption alone might not seem appealing to enthusiast gamers, you have to realize that indirectly translates into heat in your system and noise from it. Lower power allows for smaller cards, quieter coolers and an overall better user experience.
Let's see how the GTX 960 stacks up in our existing power testing infrastructure.
Well things are looking up for NVIDIA again in this department – the GTX 960 is using 46 fewer watts than the GTX 760 under a gaming load, and 80 watts less than the Radeon R9 285! These are not small numbers guys – the TDP of the GTX 960 is only 120 watts. You are using 60-70% of ANOTHER GTX 960 to get similar performance out of the Radeon R9 285 or Radeon R9 280!
But that method of power testing isn't as easy to decipher, isn't as direct. After all, we are measuring WHOLE SYSTEM power consumption above. What if we had a new way?
PC Perspective Advanced Power Testing Preview
In the coming weeks we are going to detail a brand new set of power testing methods we are using at PC Perspective to directly measure power of…well, everything if we can. This includes and starts with graphics cards.
How do we do it? Simple in theory but surprisingly difficult in practice, we are intercepting the power being sent through the PCI Express bus as well as the ATX power connectors before they go to the graphics card and are directly measuring power draw with a 10 kHz DAQ (data acquisition) device. A huge thanks goes to our very own Allyn Malventano for getting the setup up and running in time for us to get a sample of the kind of data we'll be showing in the future.
Click to Enlarge
What you see here is the combined total power of GPU power usage during a run of the Metro: Last Light benchmark. The beginning on the lines on the left (and the ending lines on the right) are the GPUs running at idle and you can see the quick spikes up during game loading and load.
So what stands out? The GTX 960, even in this overclocked state, is only using 90-95 watts by the end of the test run (which consisted of 175,000 samples). The GTX 760 was using 150-160 watts at its peaks. The Radeon R9 285 crosses the 200 watt barrier several times. This is made more interesting by the fact that the R9 285 claims to have a 190 watt TDP or typical board power, but we are definitely above that here.
The data we gathered does prove NVIDIA's claims that the GTX 960 was almost never approaching the TDP it set at 120 watts – and in fact in this one example we only cross 100 watts in a couple of quick blips.
Click to Enlarge
With this testing methodology we'll be able to get new details we weren't able to see before. Here is a zoom in of the beginning of our sample above showing idle graphics power. The GTX 960 is sitting at just around 7-8 watts, the GTX 760 hovers in the 13 watt range while the R9 285 from Sapphire is between 11-12 watts with a very regular jump every few seconds. What all of this means has yet to be determined but clearly our power testing will be much more interesting going forward.
Another thing we can check is NVIDIA's claim that you can play a MOBA like League of Legends or DOTA 2 at 30 watts with silent operation.
While I didn't fire up LoL, I did start up DOTA 2 and run through a quick single player example and the results were nearly on par. We ran the game at 1080p with maximum in-game quality settings and turned V-Sync on, limiting the frame rate to 60 Hz.
Click to Enlarge
The GPU power does start out nearly hitting the 40 watt level at the beginning of the in-game sequence but levels out at around 33 watts during game play. (This was measured with 241,000 data points.) I didn't hit the 30 watt mark in DOTA 2 that NVIDIA claimed would work in LoL but it is still easy to be impressed by a result like this.