Introduction

Today's most discussed subject is image quality. There are several reasons for this, but the basis of it all is the fact that 3dfx's latest product, the Voodoo3, isn't supporting 24/32-bit color. Because of this many people have jumped to the conclusion that the Voodoo3 will have seriously inferior image quality. However Gary Tarolli from 3dfx jumped up in defense of his beloved product and wrote a little column explaining (well more like mentioning and stressing) that their output is more like 22-bit than 16-bits. Some sites called this marketing crap while other, more serious sites like this one (ahem) decided to investigate. As a result, most people now understand that Voodoo3 does indeed have some special tricks to improve its 16-bit output so that it approaches 22-bit output. But now the question remains; just how good is this 22-bit color output? More importantly, how good is Voodoo3 image quality in general? To answer this question, I will start with an introduction explaining the most important factors that influence image quality: color resolution, spatial resolution, texture resolution, and the technique to handle multilayer transparency. Before I get into that I want to stress a couple of factors that influence screenshot comparisons in general: the do's and don'ts of comparing image quality.

How can you do an honest image quality comparison?

This a very serious problem... there are a lot of sites out there telling you "this is good quality and that is bad quality"... but who determines what is good and bad? Let's take an example. A GIF format picture with 256 different colors is probably very good for the Internet but is that same picture format still ok for a professional publication? It's pretty clear that the acceptability of image quality depends on the situation; something might be ok for this application but not ok for another application. On another level, image quality and acceptability is something that is very personal, where I might like something you don't and visa-versa. So who is right? Who decides on the border between acceptable and unacceptable?

The logical conclusion is this: the acceptability of image quality is a standard you have to decide for yourself. I cannot tell you what is good and what is bad, it all depends on your personal taste and how biased you are. Maybe you like to believe that your are unbiased, but AFAIK everybody is always a little bit biased no matter how hard they try not to be, and this has an influence on whether others will like it or not. I mean, let's face it. If you spend $X on a product and a friend visits you and says that card sucks, how do you feel? How do you respond? If you're like me you'll kick him in the crotch and tell him to get out of your house. It's natural for people to feel defensive about their choices, even up to such levels that it becomes ridiculous. This is why sentences like "crushes the competition", "kicks ass", and "wipes the floor with" are so funny. I have seen benchmark comparisons where a 2 fps difference was called a "minor difference" and a "huge difference" depending on the card that had the lead. So how do we handle it? Well, simply put, you have to make your own decision while also respecting the opinions of others, because you are both right. You will judge what is acceptable based on different factors and backgrounds, so rarely will there be one absolute winner. Now with that out of the way, let's look at the screenshots so you can make your own personal biased decision =^)

A few comments on comparing:

I just told you to look for yourself, but you must consider the images with a grain of salt. Do they really represent the quality of that product? Like I said, everybody is biased. Heck, even the people taking those screenshots are biased. People will take screenshots where one board looks better than the other based on whom they want to win! Maybe not on purpose, but possibly unconsciously they do it. Even worse, some will knowingly use dirty little tricks to make something look better or worse. So when you are comparing shots, remember to have a look around at different sites, sites that are differently biased.

One often-used trick is zooming, and the underhanded version is zooming without telling. What's wrong with zooming? Well, think about how you play the game. Will you be playing a game with a zoom factor of 200 or 300%? I didn't think so. Zooming is useful to show you small differences, but it's not a good way to compare image quality. If you want to figure out if a screenshot produced by an certain accelerator is an acceptable image, then you have to look at it like you would see it if you had that accelerator in your machine and were playing the game shown in the shot. The best way to do that is to download the screenshot to your hard drive so you don't have to view it in a browser which can cause loss of quality. I advise using an image viewer like ACDSee, or any other program that allows you to view an image full screen. This full screen option is essential because you usually play a game full screen. Another thing to do is to match your screen resolution to the resolution of your screenshot. So by looking at the screenshot at the correct screen resolution in full screen you will have the best possible approximation of what it will look like on your system. Using that display, you should then decide for yourself whether the image is acceptable or not. Remember to sit at your normal playing distance. Don't move closer to the screen. Just judge the screen from how you would play.

Most online comparisons show zoomed shots because they want to show a difference even when the difference is minimal. There's no use to them showing you ten screenshots that look identical. These detail shots do have their uses, and can help show you the differences to look for, but these shots are not ideal for making a conclusion.

Even with all these precautions there is still one huge difference between seeing the real output and using a screenshot. The real output from a 3D accelerator is moving; it's not static when you play. Motion has a huge impact on the visual perception of an image that a single static screenshot can never convey. Some artifacts are only noticeable in moving images. One example of such an artifact is the problem with moiré patterns and aliasing. Basically both effects create moving artifacts and they are not so noticeable in single static shots. Unfortunately, there is no easy way to solve this problem. The storage space available to most online sites is limited and the download bandwidth is in short supply. For this reason the only true way to judge image quality acceptability is to look at the real thing. If that's impossible you will have to go for the screenshot technique described earlier.