Single page Print

Overclocking and the silicon lottery
In a normal graphics card review, this is where I'd share the results of our efforts to wring out every last Hertz of clock speed from our Strix. Sadly, that won't be the case today. We lost the silicon lottery with the particular GP104 GPU that's soldered to this GTX 1080.

As we've already seen, Nvidia's GPU Boost technology can make the marked clock speeds on recent GeForces an understatement of their actual capabilities, and that's especially true of Pascal-powered cards. To establish the actual clock-speed baseline for these cards, we've taken to running Unigine Heaven for 10 minutes to let cards heat up before making any records of observed clock speeds.

Since the Strix card comes with multiple clock speed profiles baked into its firmware, we figured we'd start our testing by observing the clock speeds the card reached in its default Gaming Mode and the slightly-boosted OC Mode before moving to manual tweaking. Though Gaming Mode was perfectly stable in our tests, switching over to the baked-in OC Mode immediately led to hangs and artifacting, suggesting our particular GP104 chip didn't even have the headroom to handle that modest overclock.

This instability was surprising to see. I've never used a graphics card that couldn't maintain stability with its factory clock profiles, so I asked Asus about this behavior. The company told me that while its Gaming Mode profiles are guaranteed to be stable, OC Mode stability may vary from card to card. Asus suggests that if OC Mode doesn't work, owners should just flip back to Gaming Mode.

To be honest, that response surprised me—no such disclaimer is in evidence on any Asus product or retail page that I can see. What's worse, Newegg (and other retailers) present that OC Mode clock speed as the one the card will hit, not as a provisional bonus. Furthermore, Asus has sent out cards to reviewers in the past that were locked into OC Mode by default. If OC Mode isn't guaranteed for stability on Asus cards, that decision would mean the cards in reviewers' hands might not reflect the performance buyers will get off the shelf—even if it is just a tiny difference. We'd urge Asus to clarify its position on just what OC Mode means for its graphics cards so that buyers don't end up disappointed if their cards can't hit those clocks.

After our initial discovery, I tried to nurse this particular piece of GP104 silicon to higher clocks with power, temperature, and voltage limit increases, but our chip just wasn't having it. Even single-digit clock speed increases over the Gaming Mode defaults introduced instability, and that behavior persisted even with artificially high fan speeds that ruined the Strix's noise character and drove temperatures far below the typical load levels we've seen on GTX 1080s so far. Even a careful application of premium thermal paste didn't help. At that point, we wrapped up our overclocking efforts.

To be clear, we don't think these results are a black mark on the Strix GTX 1080's design or quality. Chip-to-chip variance is a fact of life with all semiconductor products, but this is the first time we've run into such a dud. It's possible that with different luck, one might see better overclocking results with a Strix, and both TechPowerUp and HardOCP had better luck with their Strixes. We just can't test that headroom with our particular sample, so caveat emptor.