Power, Heat, and Noise
Both Nvidia and AMD hate artificial stress tests like FurMark, and have initiated workarounds that allow the graphics card to downclock whenever the GPU exceeds amperage limits or thermal specifications. While some editors believe that this is a good thing for increasing GPU capability, this editor believes the same technology can cause “frame rate inflation” in actual games. The reason of course is that if a game were capable of pushing the GPU’s limit (an admittedly hypothetical scenario), FPS would only drop during the tough parts of the map—where the extra performance most needed.
Rather than drop FurMark completely, we added Just Cause 2 to today’s power tests. Because the game does not generate a consistent load, its charted numbers represent the highest power level held for at least three seconds.
Sparkle’s Calibre X560 appears to be the most affected by Nvidia’s overcurrent protection scheme, which is an entirely optional inclusion on GeForce GTX 560 Ti cards, drawing more power than its Gigabyte competitor under Just Cause 2 but less than that same competitor under FurMark. The reason for that discrepancy can likely be tracked to the voltage regulator itself, since Sparkle uses Nvidia’s reference design while Gigabyte has its own. Looking at the game leaves little room for interpretation. The Calibre X560’s slightly-higher power use in Just Cause 2 is most likely due to its memory overclock.
Asus appears most-miserly of the three middle cards, coming in 20 W lower than the same-speed Palit product.
The same scenario that played out in the power tests is also seen in the GPU temperature chart, at least when automatic fan control is used. Gigabyte’s “full fan” results appear less-impressive because the card was already pushing 89% fan speed under “auto fan” mode with FurMark.
Looking at the fan capacity of the middle cards, MSI barely edges out Asus with a card that’s clocked 20 MHz slower, while Palit falls behind.
Now that we know that FurMark results apply to every card except the Calibre X560, the only question was how we could accurately test full-load fan noise. The only application we have that can maintain full GPU load until temperature remains constant is FurMark, so we pressed on, knowing that only the Calibre X560’s results would be invalid.
The rest of the chart reveals that Asus beats Palit in noise as well as cooling, and that Gigabyte’s fans have a relatively-low maximum speed.