unnamed file 5385


Here’s the first review of a GeForce GTX 680 at Geeks3D. The GTX 680 has been launched months ago (April 2012, see here: NVIDIA GeForce GTX 680 Officially Launched) and even if I own an EVGA GTX 680 for months, I never had the time to write a review about the GK104-based card. I took the time this morning with a GTX 680 from ASUS.

unnamed file 5386

1 – GeForce GTX 680 Overview

The GTX 680 has a new clocking schema. In GTX 400 or GTX 500, there are different GPU clock speeds represented by PState or Performance State: the state P0 being the maximal performance state (for 3D gaming and maximal power consumption) and P12 being the state with minimal power consumption, usually in 2D mode under Windows desktop. You can use a tool like GPU Shark that shows you all performance states. With the GTX 680, things are slightly different, especially about the maximal performance clock speed. NVIDIA has introduced two types of 3D clock speeds that replace the P0 clock speed: the base clock and the boost clock (it’s the dynamic clocking, GPU Boost). For NVIDIA’s reference board, the base clock is 1006MHz and the boost clock is 1058MHz. Both clock speeds are actually related to the power consumption of the card (or TDP: Thermal Design Power). In short, the GPU runs at base clock when the power consumption is maximal, read 100% of the TDP. That’s the case with stress test utilities like FurMark. But in gaming scenarios, where the power consumption is lower (for example 60 or 70% TDP) the GPU can run at a higher clock speed: it runs at the boost clock speed. The GPU clock speed evolves between the base clock and boost clock with steps of 13MHz. Monitoring hardware on the PCB reads the power consumption at the level of power connectors and sends the information to the graphics driver which can decide to increase or decrease the GPU clock speed.

NVIDIA also introduced another feature: the power target. By default the power target is set to 100%. But that 100% is not equal to the real TDP of the graphics card. The real TDP corresponds to a power consumption of around 115% (the TDP of the GTX 680 is 195W). The GTX 680 will reduce (or increase) the clock speed to match the power target. And like for AMD’s Radeon HD 6900 / HD 7900 with PowerTune, you can manually set a value for the power target greater or lesser than 100%. For NVIDIA’s reference board, you can set the power target from 70% TDP up to 132 % TDP. This is useful for overclocking, since overclocking increases the power consumption.

In the following screenshot, you can see the different clock speeds and power target of a reference GTX 680:

unnamed file 5387
GPU Shark in detailed mode


作者 frank