Mo' Data, Mo' Problems: Use Every Tool For A Better Conclusion
Before we come to any conclusions, here are the average frame rates and minimum frame rates, per FCAT. We aren't expecting performance to change dramatically based on AMD's frame pacing driver. However, because metering out frames helps eliminate drops and runts, effective frame rate should go up with Catalyst 13.8.
We're leaving the Metro: Last Light numbers out of this summary. While it's true that there's a very real problem affecting AMD in that game, and you should know about it, we don't want it throwing off our comparison of the other titles.
The minimum frame rates, especially, look a lot better under Catalyst 13.8 with frame pacing. If this was the only thing we were interested in evaluating, we'd congratulate AMD on a job well done and call it a day.
But as it turns out, things get a little more complicated when we add side-by-side comparisons of the videos used to generate our quantitative data. We had hoped that the results from FCAT would clearly correspond to the gaming output used to create them. This would have made FCAT a Holy Grail for objectively describing the experience you get from gaming on any combination of graphics cards.
The problem we encountered is that, in some cases, the video appears to conflict (or at least fail to confirm) the benchmark numbers generated by FCAT. I think we have an explanation for Far Cry 3, but the Tomb Raider and Battlefield 3 videos aren't readily complementary of the data. As we're sure you can imagine, any challenge in relating benchmarks to real experiences forcibly calls into question the value of the numbers, even if they're completely correct.
It would appear, then, and perhaps not surprisingly, that the only mechanism we can fully trust when it comes to evaluating performance is our eyes. To facilitate this, we have side-by-side video comparisons. Unfortunately, we don't get tidy numbers to encapsulate the results. Sometimes we don't even end up with agreeing opinions. So, benchmark numbers remain an important component of our reviews; they aren't going anywhere. But it's important to acknowledge that there's more to this story than average frame rates. Side-by-side video comparisons are important, too.
What conclusions can we draw about AMD's Catalyst 13.8 driver, then? First, we applaud the company for addressing frame pacing, and we look forward to the second phase of its driver release that'll add OpenGL, Eyefinity, and DirectX 9 support. It's encouraging that FCAT reports significantly lower frame time variance using the new driver. Conversely, it's also clear that AMD needs to fix a problem with Metro: Last Light. While it's a bummer that there are still outstanding issues, at least we know the company's working to fix the problems reported in CrossFire.
What about the Radeon HD 7990 specifically, now that it's selling for $700? The card is certainly more attractive after the price cut, and AMD's Never Settle eight-game bundle is pretty awesome for folks who don't already own its contents. Although the Radeon HD 7990 is more power-hungry than the GeForce GTX 690, and arguably nowhere near as elegant, a $300 premium on the dual-GK104 card makes it a bit easier to live with some of the 7990's persistent quirks. It seems clear that the GeForce GTX 690 delivers more consistent performance. But more so than ever before, AMD's dual-GPU card is at least a contender now.
At the end of the day, though, we'd still be most inclined to recommend two Radeon HD 7970s in CrossFire or two GeForce GTX 770s in SLI ahead of either dual-GPU board. With that said, now we need to determine if the rest of AMD's cards are equally affected by Catalyst 13.8 when we pair them up. Looks like it's time for some more benchmarking...