Power, Temperature, and Overclocking
Last year when NVIDIA released their Kepler architecture it was a point of contention that they had a more efficient architecture when it came to gaming performance. It certainly seems that perhaps after some minor fixes on AMD’s side they are right there when it comes to power/performance efficiency.
At idle all of the cards are pretty closely matched, with the 760 being lower than everyone else. AMD is not quite able to get their idle power down as low as what NVIDIA does, but it is still pretty low. In CrossFire mode AMD is able to turn off one card altogether, so there are only a few watts difference between single and multiple cards.
At load we see the 770 and 280X right at the same level. The 270X on the other hand runs over 50 watts less than the 760. Once CrossFire is enabled, then we see a pretty significant increase in power consumption at the wall socket for the AMD cards.
Asus has done a mighty fine job with their DCII coolers. All of the cards at idle with the fans spinning silently all are in the 30C range. Once things heat up a bit, we still see a pretty compressed range of temperatures for the cards. The only real outliers are the boards in Crossfire.
I have to admit that I was slightly disappointed in the overclocking potential of these cards. The results weren’t bad, but they certainly were not beyond the norm.
The 280X board is clocked at 1070 MHz out of the box, but I was only able to get the board stable to 1135 MHz with a small voltage tweak. The memory did have a bit more headroom as I took it from 6400 MHz up to 7300 MHz. Overall this lead to some nice performance gains, but nothing all that impressive for such a beefy cooler.
The 270X boards was also disappointing in that I had previously taken the MSI R7870 HAWK to 1300 MHz core. I was expecting as good, if not better, overclocking performance from this card. It was not to be. I was able to take the board from the 1100 MHz stock speed to 1225 MHz. It is a better overclock than what we saw with the 280X, but it does not reach the heights that the year+ older R7870 did. The memory on this board also did not overclock all that well, especially as compared to the 280X. I was able to take it from 5600 MHz up to 5900 MHz. This is not a huge overclock by any stretch of the imagination for GDDR-5.
I was honestly expecting a lot more out of both solutions, but I guess I still can’t complain about free performance with just a little tweaking.