Well, since we're offering our own anecdotal evidence, I purchased the Sapphire 7970 "Dual-X" OC card for the following reasons:
- NVIDIA's "boost clock" is your method for overclock, but you cannot directly control when the card does or doesn't "boost". Thus, you can configure the boost to extend all the way up to, say, 1500Mhz, but if the card doesn't feel like going there (the internal boost mechanism doesn't like heat, or the power supply voltage, or whatever) then it simply will not do that speed. Thus, you are NOT in control of your overclock like you are on prior parts (AMD and NV.)
- Boost clock cannot be disabled. Meaning IF you're overclocking, you're not guaranteed to stay at a specific speed. Maybe that's cool from a stability standpoint, but it's concerning to me from an enthusiast overclocking standpoint. It means the strong potential for inconsistent performance when you aren't expecting it...
- 2GB framebuffer < 3GB framebuffer. With the kind of horsepower available on a 680 or 7970, I want to throw a fistfull of AA at pretty much every game. But at 2560x1440, I'm going to run out of framebuffer far sooner on the 680 than I am with the 7970 while playing the multitudes of games that have "high res" textures from the modding community. Skyrim at 4xMSAA + Adaptive Multisampling + high res textures + shadowmaps set at 8192 results in ~2.4Gb of framebuffer usage. I'd be paging a 680 to death at these settings; the 7970 shrugs and keeps moving.
- Given equal clocks, they perform equally. A 680 is clocked around 10% faster than a 7970 at default speeds, which is why it generally beats the 7970 by a similar percentage. As you overclock, the 7970 actually gets more benefit per-mhz than the 680 does. They also appear to have very similar max overclocks all things considered, but see also: my prior concerns regarding the overclocking on a 680.
Now, there are a list of CONS with the AMD board too. Drivers are an obvious one; I've been sticking to the 12.3 Betas (the REAL betas, not the 12.2's that were mistakenly labeled as 12.3 BETAs) and have not had issues. Nevertheless, I have indeed heard complaints about flashing textures in some games with newer drivers. Also, the 7000 series of video cards is a whole new architecture for AMD (GCN) which means game performance can be a bit varied for the time being; things should settle a bit with a few more driver revs.
Power consumption as a con? I dunno, people have loved NV cards for years and they've always been MONSTER power consumers. Now we're within 20W or so at peak load with the nod going to NV, and suddenly it's oh-so-important that we MUST have the lowest power consumption? Meh, maybe, but for me? An extra 20W stacked onto my overclocked 3930k, overclocked 32Gb of ram, and a 6-disk SSD RAID array is pretty much rounding error for my rig that pulls more than 500W from the wall. And I'm just gonna overclock my video card anyway, so is the slight increase in extra power really a large concern to me? Not really.
As for fan or coil noise? There aren't AMD's issues, these are issues for the manufacturer of your card. My Sapphire OC card has zero coil noise, and the fans are barely audible at their highest speed (which they never actually get that fast unless you force it via Afterburner or something.) I've also had no problems with TRIXX, AMD's PowerTune, or MSI Afterburner for my overclocking needs. I'm unaware of any DLL needs to make overclocking 'work', but perhaps that's also a card manufacturer issue (some cards may use 'non-reference' clock generators or something.)
All in all, I'm incredibly happy with my purchase. To be quite fair, I'd probably be similarly happy with a 680, albeit for slightly differing reasons. I do agree that NV has the upper hand on driver quality and initial compatibility with the newest games. I also agree that NV has better compute options in-game on older models, but keep in mind that Kepler is NOT a compute card like Fermi was. You shouldn't get into the 680 expecting awesome CUDA performance similar to it's standard gaming performance.
In fact, had the 680 not had this quirky Boost Clock business, I'd probably be rocking one right now. I did, in fact, wait until the 680 had been fully unveiled before making my purchase decision -- and boost clock was the one thing that finally swung my vote to AMD. Maybe after Boost Clock has another full generation to really get nailed down and fully understood by the masses will I be more interested in working with it. But I'm skeptical of any new power limiting technology on the first go-round; the same went for ATI when they introduced PowerTune, the same went for Intel when they introduced Turbo, and the same went for AMD when they first introduced TurboCore. All three had (at minimum) moderate issues that were resolved on the second and subsequent) generations.