Overclocking
As you can see from the picture above we ran the cards using the older Forceware drivers which have PCIe Gen 3 enabled.
For an overclocking comparison we will compare 1 card versus another. There are so many configurations we could use when comparing multiple setups so in this section we will just compare mano-a-mano, 1 vs. 1.
Overclocking the GTX680 will primarily feel a little odd as you don't directly overclock the frequencies, rather you set offsets/limits the card can works toward depending on the power target you also set. In effect you are setting the overclocking limits of the card, not what is actually achieved as the GPU will decide exactly how much boost is required and adjust the clockspeed accordingly. This is is what NVIDIA call GPU boost.
We managed to hit 1240MHz on the KFA2 card with the power limit set to 132% which is a good achievement considering the stock speed is already breaches the 1GHz barrier at 1006MHz. As expected, the memory also reacted very well to overclocking too reaching 1550MHz (6200MHz effective).
Our OEM AMD card was also a tidy overclocker hitting the milti with AMD Overdrive at 1125MHz from a stock clocked (above) 925MHz. Not satisfied with this we went a step further with the Sapphire TRIXX utility which allowed us to clock the HD7970 to 1255MHz on the core and an amazing 1720MHz (6880MHz effective) on the memory with the voltage slider set to maximum.
For giggles, we ran both 4x GPU setups in their max overclocked state but our poor 1200W Antec High Current Pro model didn't appreciate the 1340W load we placed upon it(!) and decided to take it's ball home. Thankfully, it was simply the load protection of the PSU kicking in and so no serious damage was done, other than a toasted windows install and my own cardiac arrest so if you do plan on running four of these cards in a heavily overclocked state I would recommend a 1500W PSU as an absolute minimum. Thankfully, as you will see below, there were no such problems running the setups at their stock clocked levels which is how we will be comparing the configurations today.
Power Consumption
To test for power consumption we do not use anything too technical, just a wall meter which measures the total system power draw minus the monitor. For load tests we use FurMark (mutli-GPU) which fully loads all of the GPU's. We then take a measurement when the temperatures of the cards have levelled out to ensure an accurate (where possible) power consumption reading. While Furmark stresses graphics cards to their limit, normal gaming scenarios do not add the same level of stress so the figures below are a 'worst case' scenario.
As you can see from the chart above, in single card configuration, it is pretty much even however when multiple cards are added, the NVIDIA setup becomes the less hungry, a trend which continues in 2x, 3x and 4x configurations.
Temperatures
To test temperatures we measured idle temperatures after booting windows, letting all applications finish loading and ran a few benchmarks. Once the benchmarks were complete we left the card to reach a cooling plateau where we then took the idle temperatures. For the load tests, we ran Furmark for 20 minutes, taking the absolute maximum temperature attained.
Another victory was had by NVIDIA with the card(s) running quite a lot cooler than their AMD counterparts. The difference in cards can be summarised simply by one AMD HD7970 hitting the same temperature as four GTX680's together. Again, in 'normal' gaming circumstances, the temperatures will likely be lower than this. One point to note however is that the NVIDIA cards clock frequencies began to throttle back slightly when temperatures exceeded 70c which no doubt assisted NVIDIA in controlling the temperature of the GPU. NVIDIA would call this a 'feature' of GPU boost and indeed it is good that the card has some form of thermal protection rather than simply crashing. Comparing the two cards temperatures however becomes difficult because on testing I did not notice any such throttling of the AMD cards.
Let's move on to what we all want to see - just how well the different setups perform and compare in our suite of benchmarks...
