RSS Feed

HCW Tech Blog

For the latest info on computer hardware, tech, news, video games, software tips, and Linux, check out our new improved front page: HCW Tech Blog

Reviewed by: Bryan Pizzuti [04.22.02]
Manufactured by: Chaintech


FSAA Performance

FSAA is the hottest new thing in video cards, and is designed to make this:

Into this:

to make scenes better looking. And yes, it's a simplistic explanation, but it's basically accurate. The idea is to smooth out blank pixels that might be located in a curve or diagonal line and reduce or eliminate the noticeable "stairstep" effect. The thing is, the more anti-aliasing you do, the more of a toll it can take on your video processor, depending on the method it uses. The old GeForce2 method, called Supersampling, is basically a brute-force approach, rendering a scene in a higher resolution, and then dithering it down to a lower one. But more sophisticated methods are appearing that use multisampling to blendpixels with their neighbors without so much of a performance hit. The Kyro uses a variation of this, and the GeForce3 supports it in their Quincunx FSAA mode, which is exclusive to NVIDIA GPUs.

We took Unreal Tournament and Thunder for another spin, this time to compare the various modes supported by both GPUs. Though 2X and 4X mode names are fairly common to video cards, in this case the methods are very different. The Kyro2 uses multisampling, but in these modes the TI200 uses supersampling. In the Kyro's case, the horizontal 2X multisampling was selected. We only ran this test in 1280x1024 so that all of the results could be compared on the same graph, and also included the normal mode figure for that resolution and bit depth for comparison:

FSAA is still a developing technology, as you can see. The Kyro2's 2X multisampling FSAA mode is very competitive with the TI200's supersampling, which simply can't keep up at 32 bit color. At 4X FSAA, neither card is especially playable, and they trade places between 16 and 32 bit color. NVIDIA's Quincunx FSAA mode, which is comparable to 4x supersampling, does quite well, being only slightly slower than 2x supersampling FSAA.

Next Page: (8)