Memory Bandwidth: Testing The Limits
Just as we saw in our coverage of Nvidia's GeForce GTX 660 Ti (and the subsequent round-up, Seven GeForce GTX 660 Ti Cards: Exploring Memory Bandwidth), we face the challenge of testing the 660's pared-down memory interface. To be fair, we need to compare Nvidia's OEM and retail versions of the GeForce GTX 660, which center on different GPUs (the former employs GK104, while the latter is GK106-based, as we already know). The direct competition includes Nvidia's GeForce GTX 660 Ti and 670, along with AMD's Radeon HD 7950, 7870, and 7850.
After the first GeForce GTX 660 Ti reviews went live, many readers correctly pointed out that different sites came to a wide range of conclusions. And, depending on the game, quality setting, and resolution, AMD or Nvidia hardware might have ended up on top.
Today, we see this same back and forth between the GeForce GTX 660 and Radeon HD 7870. Nvidia's GeForce GTX 660 and 660 Ti find themselves at a disadvantage when we run tests that aren't GPU-intensive. In those cases, Nvidia's powerful architecture isn't being fully utilized because the narrower memory interface bottlenecks performance.
As in yesterday's GeForce GTX 660 Ti round-up, we're using Batman: Arkham City to illustrate our case. We disabled tessellation, horizon-based ambient occlusion (HBAO), and multi-view soft shadows (MVSS) in order to not slow down the GPU too much. Running the game like this, without anti-aliasing, yielded the benchmark results we were expecting at both tested resolutions based on each card's specs. Nvidia's GeForce GTX 670 inched out AMD's Radeon HD 7950, followed by the GeForce GTX 660 Ti and Radeon HD 7870. Both the OEM and retail GeForce GTX 660s keep the Radeon HD 7850 at bay. The real question is this: what happens when we increase the resolution and turn on anti-aliasing?
Benchmarks At 1920x1080 and 2560x1440
Let’s begin our analysis with the results we obtain without any anti-aliasing, aside from FXAA. As expected, our contenders place exactly in the order we detailed above. Bringing up the rear are two passively-cooled Radeon HD 7750s running in CrossFire. Why did we bother including them? Consuming less than 100 W, their combined performance is more than compelling, especially considering their price.
Next, we drop the entry-level CrossFire setup and re-run this test at a higher resolution.
The GeForce cards manage to keep up, even at higher resolutions, despite their 192-bit memory interface and, in some cases, smaller capacity. However, this only applies when anti-aliasing is turned off. Once you start turning on higher MSAA levels, the performance situation turns ugly for the Nvidia's mainstream line-up, and limited bandwidth becomes more obvious. Interestingly, the GK106-equipped GeForce GTX 660 actually fares better at higher resolutions than the theoretically-superior GeForce GTX 660 Ti, based on GK104. Meanwhile, the OEM GTX 660 built on GK104 doesn’t stand a chance when more demanding MSAA modes are selected.
AMD's Radeon HD 7850 does unexpectedly well, too, becoming more competitive starting at 4x MSAA. Looking at these numbers, we can safely say that factory overclocked models with core clocks at 1 GHz are a good match-up to the GeForce GTX 660.
Why is this benchmark even relevant? Simple: it represents any game with a less demanding graphics engine, such as those ported over from gaming consoles and many older DirectX 9 or 10 titles still popular today.