well i should say that the above is true more often than not, but it isn't true 100% of the time. it varies from project to project, and some projects may not even make use of X-fire/SLI (even when it is enabled) due to the coding of the application and/or the nature of the data being crunched. those projects, which i believe make up only a small fraction of all DC projects out there, are obviously the few that don't seem to be affected by X-fire or SLI.
i would say though that the vast majority of DC projects that allow GPU computing do benefit from not X-firing or SLIing. the premise is that, when gaming, neither X-fire nor SLI scale perfectly...that is to say, two 7970s in X-fire does not equal twice the performance of a single 7970 - rather its a bit less than twice the performance of a single 7970 (and sometimes substantially less depending on the game being played). tri-fire/SLI performance scaling is even worse than dual card X-fire/SLI, and quad-fire/SLI scaling is even worse than tri-fire/SLI scaling...
...but take those GPUs out of X-fire/SLI, and all of the sudden each video card has the potential to reach its maximum individual performance, provided there is no PCIe bandwidth bottleneck (which lies in selecting the right mobo). in this "no X-fire/SLI" configuration, some dual card setups and some triple card setups can truly achieve the performance of two and three individual cards (in other words, 100% performance scaling...unheard of in the X-fire/SLI world). most mobos these days have at least 2 PCIe x16 slots, one of which has full x16 bandwidth, while the other has only x8 bandwidth. the first slot operates at full x16 bandwidth when the other is not populated. but when the 2nd x16 slot is populated, the first slot drops down to x8 operation to match that of the 2nd occupied slot. this is the most common inherent (built into the hardware) PCIe bandwidth bottleneck - 2 cards simply cannot double the performance of a single card, X-fire/SLI or not, b/c PCIe bandwidth to the top PCIe x16 slot is cut in half when running 2 cards.
some of the newer generation intel chipset mobos (thought i'm not sure about AMD chipset mobos) have at least 2 PCIe x16 slots that can operate at full x16 bandwidth all the time. obviously those mobos will allow 100% performance scaling w/ 2 cards.
mobos that have 3 PCIe x16 slots can't operate all 3 at full x16 bandwidth, and are usually auto-arranged to something like x16 x16 x0 w/ 2 cards installed, and x16 x8 x8 when 3 cards are installed. obviously these mobos are capable of 100% performance scaling w/ 2 cards, but not 3 cards.
earlier in the thread i mentioned some mobos w/ 4 PCIe x16 slots. the one purchased by the OP can achieve 100% scaling w/ 2 cards, but not 3 cards b/c the slots arrange to x16 x8 x8 with 3 cards installed. neither my MSI 790FX-GD70 nor its successor, the MSI 890FXA-GD80, can achieve 100% performance scaling w/ 2 cards b/c once the 3rd or 4th (or both) PCIe x16 slots are occupied, the 1st or 2nd (or both) PCIe x16 slots, which are the ones that are otherwise capable of full x16 bandwidth, drop down to x8 bandwidth. the same can be said about occupying all 4 slots - 100% performance scaling cannot be achieved
i believe Gigabyte had a mobo with 5 PCIe x16 slots on it (can't remember the exact model, though i believe it used an AMD 890FX chipset), but i can't recall anything really special about its PCIe lane or slot arrangement due to spacing constraints. ASUS had the Crosshair IV Extreme, which also was an 890FX chipset mobo that had 5 PCIe x16 slots. this board was actually capable of running 3 cards at full x16 bandwidth (and therefore at 100% performance scaling), and in 2 different slot configurations no less!
OP, hope i'm not derailing the thread, and despite this mostly being the answer to someone else's question, hopefully i was also able to provide some insight into the PCIe limits of your new mobo or, perhaps pointed out some things you might not have known about it so you can choose your future GPUs wisely.