Sign in with
Sign up | Sign in

Benchmark Results: SLI

GeForce GTX 580 And GF110: The Way Nvidia Meant It To Be Played
By

We’ve established on more than one occasion that Nvidia’s scaling in SLI, employing two graphics cards, is superb. That remains the case with its GeForce GTX 580, and we ran each of the games in our benchmark suite just to demonstrate the firepower available through two of these cards.

Each comparison requires a step-by-step explanation because, while we were shooting for the highest settings in each title, AMD’s Radeon HD 5870 doesn’t support all of the same anti-aliasing modes.

In Just Cause 2, with Nvidia’s boards running 32x CSAA and AMD’s employing 8x MSAA, performance is actually fairly close at 2560x1600 with all of the detail settings maxed out. You don’t get a ton of extra performance by switching from GTX 480s to 580s. But the frame rates are generally pretty playable in both cases.

DiRT 2 makes for an easier comparison, as all three configurations are set to 4x MSAA. What you see is what you get. Regardless of whether you’re running a pair of Radeon HD 5870s or GeForce GTX 580s, though, 2560x1600 is easily accessible.

The Radeon cards seem to struggle with 8x MSAA in Battlefield: Bad Company 2 (switching to 4x MSAA yields faster frame rates than both of the GeForce pairs, though). Switching from GeForce GTX 480s to 580s is good for a roughly 17% performance increase here.

With all three combos utilizing 4x MSAA, the comparison is easy in Aliens Versus Predator. A 16% speed-up is about what you can expect after trading your 480s in for 580s.

Lost Planet 2 is almost untestable with 8x MSAA enabled on the Radeon cards, and even 4x MSAA suffers a huge slowdown. With that said, the GeForce boards do fine in this TWIMTBP title, jumping 23% when we switch from GeForce GTX 480s to 580s.

The highest you can crank Metro 2033 is 4x MSAA. But even then, this game brings AMD’s boards to their knees with its Very High quality setting. If we weren’t so suspect of the optimizations made in favor of Nvidia, we’d label this game the new Crysis and add it to our benchmark suite for the next three years (kidding). Here again, frame buffer limitations on the AMD cards are hurting performance, giving the larger 1.5 GB repository on each GeForce board a marked advantage at the resolutions you'd want to use if you had a pair of these pricey boards.

Evening Out The Score

It is worth noting that AMD's recent Catalyst Hotfix 10.10d does include some optimizations for CrossFire in a number of different games. Only one (Metro) is represented in our suite, though. Testing that title with the hotfix increases frame rates from 4.7 FPS to 14.0 FPS on our Radeon HD 5870 1 GB CrossFire configuration. The new number is still not playable, but it's a significant improvement.

You can further improve performance in frame buffer-limited situations by shifting away from the 1 GB 5870s and toward the 2 GB cards, working around the ceiling imposed by high resolutions and extreme multi-sample anti-aliasing. Using a pair of Radeon HD 5870 2 GB boards, performance in Metro 2033 increased even more to 23.67 frames per second. Of course, now you're looking at an additional ~$50 per board, increasing the cost of two 5870s considerably without exceeding the performance of two GeForce GTX 480s.

The only other title to see a significant gain here (compared to the first graph on this page) is Battlefield: Bad Company 2, though this seems to be a result of the CrossFire hotfix, not the extra 1 GB of memory per card. For the most part, spending more on the 2 GB boards won't really help out that much.

React To This Article