GTX1080-Feature2
A few posts on TechPowerUp have highlighted issues with GTX 1080 and 1070 cards from Asus and MSI. Specifically, both vendors have been caught shipping cards to reviewers that were configured for overclocking mode out of the box, while retail cards are shipping at base clocks by default.
In this case, the clock speed differences are very small, at roughly 1.5%, which means they’re only likely to produce ~1% of difference, if that.
GPU-Z-Settings
Image by TechPowerUp
Asus responded to PC Perspective’s inquiry on this issue by noting that reviewers and buyers alike can adjust GPU clock speeds via its GPU Tweak II utility, and that “The press samples for the ASUS ROG Strix GeForce GTX 1080 OC and ASUS ROG Strix GeForce GTX 1070 OC cards are set to “OC Mode” by default. To save media time and effort, OC mode is enabled by default as we are well aware our graphics cards will be reviewed primarily on maximum performance.”
The truth is, vendors have been pulling tricks like this for well over a decade. In the old days, they’d overclock CPU buses slightly, pushing a 133MHz base clock up to 136MHz. On a 2.13GHz CPU with a 16x base clock, that’s enough for a roughly 2% clock speed increase. Other scenarios have been more egregious — we’ve seen motherboards that would automatically enable optimized CPU overclocking settings when XMP memory profiles were enabled. In this particular case, that meant all CPU cores were set to run at the maximum Turbo frequency normally reserved for a single-threaded scenario.Optimizations like this can impact measured performance by significant amounts, much more than the 2% we mentioned earlier.
Asus claims that these changes are made to “help” reviewers, but that’s a secondary reason at best. Yes, we evaluate cards based on maximum performance, including overclocking performance — but what this is really about is securing top placement on a comparison graph between multiple vendors.
Consider, after all, the plight of companies like Asus, MSI, Zotac, Gigabyte, EVGA, and the other various GPU or motherboard vendors. They know that pricing is at least as important as brand when it comes to convincing users to buy a GPU. The problem is, many buyers do buy on price. The only way to justify asking an extra $10 to $20 is to offer something the other guy doesn’t have. Cooling, overclockability, and quiet operation are all ways to influence customer decisions, but those features only work if they can establish meaningful differences. Overclocking always varies by card and a GPU family may not be particularly loud or hot by default.
A card that turns in consistently higher performance is a card that’ll tend to either be at the top of the stack or will be highlighted in a different color. It’ll be the component that catches the eye, one way or the other.
There is a scintilla of truth to Asus’ statement. Because reviewers often review many cards at once, making certain that you’ve configured every piece of OEM software required to enable a given feature can be confusing. Since a review is a presentation of a product under objective testing conditions, Asus can make the argument that they want to make certain the product is tested in the right conditions. It’s not completely wrong. The problem is, those “right conditions” may be just as applicable to the end-user, who may not bother installing or configuring OEM software, either — particularly if they have the long-standing opinion that OEM software is more or less garbage.
Is a 1.5% overclock a fundamental betrayal of customers? No. We routinely accept much larger variations in products we buy. But the problem with pushing the envelope like this, beyond the fact that it looks pretty bad, is that it can lead to instability or other problems. In the motherboard case we mentioned above, the system would crash at full CPU load because the CPU we were using wasn’t a particularly good overclocker and couldn’t run all four cores at the single-thread Turbo Mode clock without a voltage nudge. Said nudge wasn’t programmed into the UEFI, which meant the chip seemed unstable until we hunted down the actual cause of the problem.
In some cases, even tiny increases cause issues. While our Fury X GPU runs rock-solid at stock speed, nudging it upwards even by 3% caused instability last year. The bottom line is that manufacturers should keep stock speeds stock and offer overclocking modes through clearly communicated alternate settings — not preloaded BIOSes pulled for reviewers.