Power, Temperature, and Overclocking
For as large as the chip is, the GTX 780 does not pull a tremendous amount of power. Part of that is simply due to the board having a max supplied wattage of 250W and another part is that the chip just is not clocked all that high. NVIDIA typically has excellent idle power consumption due to aggressive downclocking of the GPU and power gating.
As we see with these results, the GTX 780 has the second lowest power draw at idle. Once we crank things up we see it have the highest draw of the single cards. The AMD cards in CrossFire obviously pull a lot more power than a single GTX 780. Still, for as energy efficient as NVIDIA claims for their parts, AMD essentially matches power and performance.
The reference cooling on the GTX 780 is very good, but NVIDIA’s partners tend to do a little better when it comes to non-reference cooling. EVGA has obviously spent some quality time with their ACX coolers. In testing I rarely heard the fans spin up, and I can honestly say that I barely noticed any fan noise whatsoever in use.
It must have been a cold day in my lab when I took these readings. While I doubt it got down to 24C, it is certainly cooler running than the other cards used in this test. Now onto the load test.
EVGA’s ACX cooler does its job very, very well. It isn’t much hotter than the Asus DCII GTX 770, and it is quite a few degrees cooler than the Asus DCII based R9 280X. Looking back at my Asus DCII GTX 780 review from last summer, the EVGA card again runs significantly cooler than that particular implementation.
NVIDIA cards are sometimes a little strange with overclocking due to their use of boost modes. While this particular card is clocked at a base clock of 863 MHz and a boost of 902 MHz, it consistently showed a clockspeed of 966 MHz under load. This card is not labeled as being overclocked (FTW or SC in EVGA terms), but it certainly acted like one. This is just an effect of the boost modes and thermal/power headroom that NVIDIA and their partners allow their cards.
I used EVGA’s overclocking software, Precision X, to push this card as much as possible. I increased the power target as high as it would go and slightly increased voltage. I then tested in small increments with a repeating benchmark and monitored clockspeeds. Typically the boards would not hard lock unless they are clocked really high, instead the clockspeed would just settle into an area and rarely move up.
I was able to get the core to run at +110 MHz, giving a speed of around 1076 MHz. This is a very nice jump up from the claimed stock speed. Performance scales nicely with core clock and we see gains in the 10% to 14% range, depending on the title. The memory did not overclock very well in this instance. I was able to get an extra 55 MHz out of this board, which translates in overall clockspeed of around 6223 MHz effective. Users gain a little bit of performance from this overclock, but not a whole lot.