The overall picture
And now, the moment we've all been waiting for. Well, unless you skipped straight to this page. We've taken frame rates from all of our games and averaged them in an effort to concoct an "overall" performance figure for each graphics card.
First, though, some caveats. We really must stress that these overall performance numbers represent overall performance only across our benchmark suite. As much as we think we did a decent job of representing what's on store shelves today, you might get different results with a different mix of games, or perhaps even different tests within each game. There's sadly no such thing as a numerical panacea to gauge a graphic card's overall performance.
As we move from comfortingly repeatable benchmarks to the wild and scary world of statistics, the math involved is also subject to different schools of thought. In our storage value roundup, for instance, we used harmonic means to smooth out the wide disparities between individual results. Here, we've gone with unweighted arithmetic means of the average frame rates obtained for each game. If we want to give all games equal importance, then the unweighted arithmetic mean gives us a mathematically correct average of performance across our test suite. But if we assumed the number of frames output was the constant, then it'd be proper to use an unweighted harmonic mean. Yep, "wild and scary" sounds about right.
Finally, remember that performance per dollar is only a small part of the equation. These numbers don't account for noise levels, power and thermal requirements, or your personal eye candy needs. Not everybody has to have antialiasing switched on in every game, and some of us might not mind sacrificing a little performance for a little extra quiet.
Now that we've completely destroyed the credibility of the numbers below, let's take a look at them! We started off by averaging our results minus exclusively DirectX 11 titles, since we wanted to compare current-gen cards to last-gen ones. We took numbers from Borderlands, Bad Company 2, DiRT 2 in DX9 mode, and Just Cause 2 only.
The previous-gen cards put up one heck of a fight, with the Radeon HD 4870 actually beating the Radeon HD 5770 in raw bang for buck and the "factory overlcocked" GTX 260 nipping at the heels of the GTX 460 768MB. We can take another look at this information by expressing it as a bar graph, with the key unit being overall frames per second per dollar:
Yes, DX11 GPUs can produce better graphics and, in the 5770's case, with lower noise levels and power consumption to boot. But raw performance per dollar at the ~$150 and $200 price points doesn't appear to have budged much over the past year. Interesting.
Next, let's forget about the previous-gen cards and look at our whole test suite. The average frame rates below encompass all of our games, minus DiRT 2 in DX9 mode. (We wouldn't want to include the same game twice.)
What do these graphs tell us? The bar chart tells a simple story of the cheaper cards generally being the best raw values. The scatter plot expresses more of the complexity here. Based on it, if raw performance per dollar is your chief concern, then we'd be inclined to recommend either the GeForce GTX 460 1GB or the GeForce GTX 470. The GTX 470's thermals and noise levels aren't terrific, though, so doling out a definite recommendation at $300 isn't as easy as it might seem.
One thing looks clear: AMD really needs to cut the prices of its 5800-series Radeons.
Full system pricing
Another way to think about performance per dollar is to look at the graphics card as part of a complete gaming system. Top-of-the-line cards obviously look like poorer deals in a vacuum, but does that change when we account for the rest of the build?
To answer that question, we took the overall DX11 performance numbers you saw above, and we factored the cost of build based on our latest system guide into our prices:
|Processor||AMD Phenom II X6 1055T||$199.99|
|Motherboard||Asus M4A89GTD PRO/USB3||$149.99|
|Memory||Crucial 4GB (2 x 2GB) DDR3-1333||$93.99|
|Storage||Western DIgital Caviar Black 1TB (6Gbps)||$94.99|
|Power supply||Corsair TX650W||$89.99|
In case you're wondering, we took the components from the system guide's Utility Player config with the enclosure and power supply of the Sweeter Spot. This build should do a decent job of representing what your typical PC gamer might want to buy: a fast $200 CPU, a good motherboard with next-generation I/O, four gigs of RAM, a terabyte hard drive, a reasonably beefy power supply, and a nice, stylish enclosure.
Work full system prices into our performance data, and the top-of-the-line cards end up looking like the best deals. Our "factory overclocked" GeForce GTX 480 looks particularly well positioned. Perhaps that's not such a big surpriseafter all, spending an extra $160 to go from a GTX 470 to a GTX 480 doesn't seem so crazy if you've already got over a thousand bucks set aside. Food for thought. Just don't go thinking you absolutely need an extra 15 FPS, especially if you have a 24" or smaller monitor.
|The TR Podcast 169 video: Win10, Elon's musk, and the gimpy GTX 970||0|
|In the lab: Dell's Venue 8 7000 tablet||3|
|Qualcomm posts record revenue, loses high-profile design||5|
|Intel refreshes high-endurance server SSDs with 20-nm NAND||15|
|The TR Podcast is live on Twitch right now||1|
|We'll be streaming the TR Podcast LIVE in one hour||1|
|Report: AMD plans Kaveri refresh for June-July time frame||50|
|TR subscribers get early access to Homeworld Remastered beta||19|