Single page Print

Overclocking
Thanks to Nvidia's GPU Boost technology, overclocking a GTX 1080 isn't quite as simple as moving some sliders and calling it done. As we saw with our out-of-the-box performance testing, the boost clock on Nvidia cards isn't an absolute maximum—it represents a typical speed under load, determined by factors like temperature limits, power limits, and voltage. The GPU can clock up past that boost number under the right conditions.

The last time we tested custom-cooled graphics cards, we relied on a FurMark variant in MSI's Kombustor app called FurryPQTorus. FurMark is a great stability tester, but the extreme thermal load it produces isn't representative of actual games. To investigate how the GTX 1080 Xtreme Gaming performs, I'm relying on Unigine's tried-and-true Heaven benchmark. Heaven is still an exceptionally demanding application, but the load it produces is closer to what one might see from an actual game than the one from FurMark.

As a point of reference for the overclock to come, here are the clock speeds and voltage numbers the GTX 1080 Xtreme Gaming exhibited in "Gaming Mode" and "OC Mode" after running Heaven for 10 minutes:

  GPU
base
clock
(MHz)
GPU
boost
clock
(MHz)
Memory
speed
(MT/s)
Heaven
GPU
voltage
Heaven
GPU
clock
(MHz)
Gigabyte GTX 980 Ti G1 Gaming 1152 1241 7010 1.193 1342
Gigabyte GTX 1080 Xtreme Gaming 1759 1898 10206 1.050 1987
GTX 1080 Xtreme Gaming (OC Mode) 1784 1936 10400 1.050 2012

As you can see from the "Heaven GPU clock" results above, the GTX 1080's specified speeds have practically no link to the real-world clock speeds it can deliver. Whatever parameters GPU Boost 3.0 is considering, it's apparently comfortable pushing the Xtreme Gaming card far above its rated clocks. I also kicked the card into its "OC Mode" to see how much Gigabyte's hottest profile would eke out of the card, and I observed a 25-MHz clock speed boost. Not much, but not nothing.

To overclock the Xtreme Gaming, I maxed out the voltage, power, and thermal limit sliders in the Xtreme Gaming Engine software before edging up the clock speed offset. This isn't a particularly refined approach, but it didn't seem to harm the card, either, so I rolled with it. I continued pushing clocks upward until I began to see artifacting or crashes in Heaven, and I double-checked my work by running Doom for a while. In my experience, Doom exhibits serious artifacting while running on a graphics card pushed past its limits, so it's a good real-world reference for the success or failure of an overclock.

  GPU
clock
offset
(MHz)
Base/
boost
with
offset
(MHz)
Memory
speed
(MT/s)
Heaven
GPU
voltage
Heaven
GPU
clock
(MHz)
Heaven
GPU
temp. (°C)
Gigabyte GTX 1080
Xtreme Gaming (OC)
+50 1759/1948 10868 1.094 2063 68

The table above records the maximum clock speed offset I was able to use while keeping the card stable. The "boost+offset" number describes the boost speed the Xtreme Gaming Engine software displays after clock tweaking, not an actual number teased out through performance testing. For that, we need to look at the voltage and "delivered" clocks, for lack of a better word, that the card exhibited after 10 minutes in Heaven with this overclock. Under an actual load, the GPU settled into an observed speed of 2063 MHz. At that speed, GPU-Z indicated that the card was voltage-limited.

Considering that firing up OC mode in the Xtreme Gaming engine resulted in a 2012-MHz stable speed, our particular piece of GP104 silicon only had about 2.5% more clockspeed left in it and about 4.5% more memory speed on top of the factory boost that Gigabyte dialed in. It's worth remembering that every chip and card is unique, however. Some of these GTX 1080s will fare better than our test unit, and others may fare worse. Even taking the silicon lottery into account, Gigabyte's easy-to-use tuning software and beefy custom PCB seem to offer no obstacles in taking a particular piece of GP104 silicon to its limits, and that's what we're really interested in when we overclock a graphics card.



To see just what those clock speed boosts mean for performance, I re-ran our Shadows of Mordor and Rise of the Tomb Raider benches. Depending on the resolution used, overclocking the card resulted in anywhere from about 2-4% better performance in RoTR and about 3.5-6% better performance in Shadows of Mordor compared to its results in Gaming Mode. Whether that's worth the trouble of manual overclocking is in the eye of the beholder, but if you're looking for every last bit of performance, it might be worthwhile to move some sliders around.

I also tested our system's power draw while the GTX 1080 was overclocked and found that it pulled about 30W more from the wall with the sliders maxed, or 310W in total. We'd expect nothing less from an overclocked card, but its effect on system power draw is still less than that of a custom GTX 980 Ti running at its default clock profile.