It’s safe to say that Nvidia is really competing with itself at this point in time. The current GeForce GTX 680 is pretty much even in performance to AMD’s Radeon HD 7970 GHz Edition, but much quieter and uses less power. The GeForce Titan outperforms AMD’s single-GPU flagship by a wide margin, but costs a cool grand, so it’s out of reach of most users.
Enter the GeForce GTX 780. At first blush, it seems like a “Baby Titan”, but that would be inaccurate. Let’s look at the base specs, compared to both the Titan and the GTX 680.
|Feature||GTX 680||GTX 780||GTX Titan|
|Memory Type||GDDR5 (6gbps)||GDDR5 (7gpbs)||GDDR5 (6gpbs)|
|Transistors||3.5 billion||7.1 billion||7.1 billion|
|Core Clock Speed (ref)||1006 MHz||863 MHz||836 MHz|
|Boost Clock||1058 MHz||900 MHz||876 MHz|
|Noise Under Load (ref)||46 dBA||43 dBA||46 dBA|
Given that the GTX 780 uses the same GPU chip as the GTX Titan, but with roughly 15% fewer shader cores and half the memory, the GTX 780 offers about 80% of the gaming performance of a Titan, as we’ll see shortly. Take a look at that memory speed, too: 7000 MHz (effective), or 1gpbs faster throughput than the Titan or GTX 680. There’s no lack of memory bandwidth with the GTX 780. However, Nvidia told us that the GTX 780 would only have about a quarter of the double precision floating point performance of Titan. In other words, the GTX 780 will be a great gaming card, but won’t come close to Titan for high end GPU compute.
Digging a little deeper into the features of the GTX 780 card itself, Nvidia’s made some interesting design decisions in the reference design. The cooling subsystem is tweaked from Titan to run even quieter. Nvidia accomplished this by managing fan speeds to run closer to a steady state, rather than ramping the fan speeds up and down rapidly.
The GTX 780 will cost substantially less than a Titan, at about $649 for reference grade cards, but that's nearly $200 more than a 2GB GTX 680. However, 4GB GTX 680s still cost nearly $600, so the price differential between a GTX 780 and GTX 680 4GB card isn’t as large, while new new card offers quite a bit more performance. Still, $649 is a pretty steep price for a video card, and it’s partly a result of AMD’s inability to compete on single GPU performance. The lack of competition puts Nvidia in the enviable position of being able to set higher prices than they might have if competition had been stiffer. I included a GTX 680 4GB card for comparison, but it’s likely that performance differences with a 2GB card will be minor.
With this sobering thought in mind, let’s take a look at performance.
Let's Get Straight to Performance
Canned, in-game benchmarks and synthetic tests have fallen out of favor, but they can still reveal a lot about comparative performance. Single GPU performance is most affected by average and minimum frame rates, but these high end cards tend to perform pretty smoothly, even in demanding titles. That doesn’t mean you’ll see high average frame rates, though. It all depends on the game, resolution and detail settings. Micro-stuttering and similar hitching issues are more prevalent in multi-GPU setups, which I tend to avoid in general.
The test system is based on the current high end of Intel’s Ivy Bridge line, the Core i7 3770K, clocked at its reference 3.5GHz / 3.9GHz turbo. This is the high end of current mainstream PCs. While Intel’s new Haswell CPU is just around the corner, its impact when running with these GPUs will likely be small. Intel is still offering the extreme high end LGA 2011 CPUs, but I wanted to stick with something that came close to a mainstream system.
I also want to explore the upgrade side of the equation, so I’m including an Nvidia GTX 580 reference card as part of the mix. Is the GTX 780 a viable upgrade from the GTX 580? What about an upgrade from the GTX 680?
First, let’s take a look at 3DMark Fire Strike, at the Extreme setting, just to calibrate some expectations.
The GTX 580 is far, far behind.
3DMark isn’t a game, however. I ran the built-in benchmarks for three recent games: Bioshock: Infinite, Tomb Raider and Metro Last Light. Each test was run at two resolutions, 2560 x 1440 and 1920 x 1080. The first resolution represents the new generation of 27-inch, 16:9 displays, while 1080p is the mainstream resolution for most users today. I also ran each benchmark at two different detail levels – completely maxed out, and a general “high” setting, usually by selecting a “high” preset from the game’s menu.
Testing at Graphics Fully Maxed
PC gamers often dream of running at very high resolutions, with all the eye candy set to 11. With this new generation of cards, it’s possible in many games, though a few may still struggle.
Bioshock Infinite uses the Unreal Engine. It’s got a fair amount of eye candy, but overall polygon counts and texture resolution seem pretty modest. It’s a clear case of a game that was designed for a console, with minimal changes in art assets for the PC port. On a standard 1080p display, Bioshock Infinite runs very well on a GTX 680 and above. At the higher resolution, the GTX 680 and Radeon HD 7970 struggle a bit, but the GTX 670 hits a solid 60 fps.
At Ultra custom settings (with all the sliders maxed), Tomb Raider uses an enhanced set of hair algorithms called Tress FX, which was actually developed by AMD. At 1080p, the Radeon HD 7970 hits the magic 60 fps number, as does the GTX 780 and Titan. The GTX 680 falls short, and the GTX 580 is even further behind.
At 2560 x 1440, the Radeon HD 7970 ekes out 60 fps, but the GK110 based Nvidia cards perform very nicely.
The Metro Last Light benchmark is an intense combat scene with lots of spectacular action and postprocessing effects. Even at 1080p, nothing comes close to 60 fps. You will get additional performance by disabling anti-aliasing, but the Metro Last Light engine is still incredibly demanding.
Playing with Graphics On Simply High
Sometimes you need to calibrate your expectations. Let’s see what happens when we notch the performance level down a couple of levels.
Changing the setting in Bioshock Infinite to “High” uses a DirectX 10 code path. All the cards perform quite well on 1080p monitors. Jacking up the resolution, and only the GTX 580 falls short.
Tomb Raider’s high setting means it runs well on just about everything we tested. So that’s a good fallback for GTX 580 owners, but you do lose some of the HDR and other post-processing visuals.
Using a Watts Up Pro, I captured power settings while the system was idling, and also while running 3DMark’s Fire Strike with all the sliders dialed up to their maximum settings. Note that the power supply used is a Corsair TX850w.
The GTX 580 was somewhat infamous for power and heat, though its power draw under high load doesn’t look all that bad – until you realize that the GTX 680 is about 40% faster in Fire Strike than the GTX 580, while drawing less power. The GTX 780 uses more power at full load than a GTX 580, but its power efficiency is much better, on a performance per watt basis. It also uses 30W less at idle, which adds up over time, since the graphics card will spend a lot more time idling than running at full bore in most PCs.
Surprisingly, the Radeon HD 7970 GHz Edition is still relatively power efficient. However, it’s noise levels are much higher than the GTX 780 (or even the Titan or 680).
In addition to the GTX 780, Nvidia is shipping the final release of the GeForce Experience. GeForce Experience is a useful tool which will alter game settings for you, with supported games, so you don’t need to dig into the game’s own menu settings. On my system, GFE could optimize six games. However, it’s worth noting that GeForce Experience isn’t always perfect. Take Tomb Raider, for example.
My own settings mirror most of GeForce Experience’s suggestions, except for resolution and Shadows. I learned from bitter experience that Tomb Raider would become a slideshow at 2560 x 1600 in some areas; dialing back to 1920 x 1200 helped considerably – and this was with a GeForce Titan and recent drivers.
Is the Upgrade Worth It?
GTX 680 owners probably don’t need to upgrade. The GTX 680 still offers good performance on all but the highest resolutions, for most games. On the other hand, if you’re still running a GTX 580, it may be worth considering an upgrade, though that $649 price point will make most people pause. It’s well known that Nvidia is prepping the GTX 770, but that’s not out yet, so we don’t know the impact on buying decisions quite yet. But it’s likely the 770 will still outperform the GTX 580; we just don’t know the margins yet.
On the other hand, if you’re a user who buys high-end gear every other generation, and you’re still on a GTX 580, the GTX 780 offers a substantial boost in performance. The GTX 780’s initial price isn’t really much more than the GTX 580 at its initial launch. And considering the number of shader units and that incredibly high speed memory onboard, maybe the price isn’t so bad after all.