We'll wrap things up, as we often do, with a couple of value scatter plots. These show average performance (computed as a geometric mean of all our test numbers) plotted against pricing. Our first plot shows 99th-percentile data converted to FPS for easier parsing, while our second plot shows plain-old average FPS numbers for comparison.
In both cases, the best solutions reside near the top left of the plot, where performance is highest and price is lowest. Cards that sit too close to the bottom-right corner, conversely, are slower and more expensive than the rest.
Note that the plots show performance data obtained at both the "Ultra" and "Very high" settings. The "Very high" data points are appended with "(VH)" to distinguish them from the others; their colors are also a little fainter.
According to our 99th-percentile metric, which is usually the most representative of real-world performance, the GeForces are quicker than the Radeons. The GTX 650 Ti 1GB outmatches the 7790 at both the "Very high" and "Ultra" settings, and the GTX 650 Ti Boost is far ahead of the 7850 2GB. The contest between the 7870 and the GTX 660 is closer, but it's still a win for Nvidia—especially since other reference-clocked GTX 660 cards similar to our Asus model are available for under $200 at Newegg right now.
So, the Nvidia cards are better options, right?
Well, not entirely. As we saw on the past few pages, the GTX 650 Ti Boost and GTX 660 both grapple with rare but substantial latency spikes that interrupt gameplay. Those spikes aren't represented in our 99th-percentile data, since they affect less than 1% of all frames. Because of those spikes, the GeForces spend more overall time working on longer-latency frames than their AMD rivals. Meanwhile, our frame-by-frame plots reveal that the Radeons suffer from timing issues of their own: short, frequent frame latency surges that can make gameplay feel less fluid than it ought to be. We noticed both the big slowdowns on the GeForces and the smaller, more frequent ones on the Radeons during our play-testing sessions.
This is a perfect example of the inadequacy of FPS averages, by the way. None of these issues are discernible from the FPS numbers alone.
Complicating matters further—especially since we're now bringing value into the picture—all Radeons from the 7790 upward come with free copies of BioShock Infinite. If you buy an Nvidia card, you'll need to purchase the game separately... and that 50 bucks could go toward a higher-tier card.
Yeah, this is a tricky one.
All things considered, I think the Radeon HD 7870 is the best option among the cards we tested for BioShock Infinite. It's competitive with the GTX 660 in our 99th-percentile metric, it spends less time working on high-latency frames, and it comes with a free copy of the game. Things are less cut-and-dried with the other cards, though, and I'm hesitant to make an unequivocal recommendation. The best option may come down to your own personal preference.
Before we sign off, a word about BioShock Infinite. Turns out this game really doesn't need that much horsepower to look good and run well. Cards like the GeForce GTX 660 and Radeon HD 7870 are fast enough to run this game quite smoothly at the highest detail level at 1080p. Lowering the detail level only slightly makes the game perfectly playable on even $140-150 GPUs, as well, which is great news for PC gamers without a ton of cash to spare.