Benchmark Results: Power And Heat
How about that for an interesting chart? The GF100-powered GeForce GTX 480 sits on top, ahead of the dual-GPU Radeon HD 5970, under load. The GeForce GTX 470 is the third-highest consumer here, with the Radeon HD 5870 and 6870 behind.
Notice that the GeForce GTX 580 isn’t even represented. The protection circuitry that Nvidia added prevents the card from running FurMark without throttling down. So, the maximum power shown appears to be lower than the GTX 470, though that’s clearly not realistic. AMD and Nvidia have long despised FurMark for its ability to inflate consumption figures, and now it looks like they're finally going to get to see it phased out.
Instead of leaving out power measurement altogether, we logged consumption in Metro 2033 across three runs of the demanding title’s built-in benchmark. The result is telling for a few reasons. First of all, the GeForce GTX 580 and 480 sit within 9 W average power use across the tests. This is because Nvidia shot for higher clocks and all 16 SMs enabled with GF110, bringing power back up to GTX 480 levels. Fair enough.
More interesting is that the Radeon HD 5870 and GeForce GTX 480/580 are separated by only about 100 W here. FurMark would have us believe that this gap is close to 160 W. That’s still a pretty significant number, though. We continue to be impressed with the performance AMD’s Cypress GPU delivers compared to the power it draws.
Measuring temperatures similarly becomes difficult without the ability to apply a consistent, demanding load, and FurMark again falls short there. We’re left with the scores gleaned from the GeForce GTX 480, which should be somewhat similar given the 580’s comparable TDP rating.
What’s not taken into account, however, is the redesigned cooling mechanism, which is more effective. GeForce GTX 480 presented Nvidia’s engineers with a challenge in this regard, and they came up shy of “good.” The enthusiast community lit a metaphorical fire under their asses, and we have a card that’s much more capable of coping with similar thermal output.
The difference in SLI is most telling. Our Falcon Northwest-based test platform’s Scythe fan was noisier than the two 580s under load in Metro. You’re still encouraged to space both cards three slots apart, but at least now we’re making a concession that yields usable acoustics under load.