Trianges, Pixels and Texels. Oh My!

Gimmie Fill Rate!

So the GF256 has more pixels than any other video card has texels, right? Sure. But unfortunately, it is not a balanced video card. I absolutely love the idea of hardware T&L and think that NVIDIA definitely made the right move by including it. However, when I see the disappointing fill rate (keep in mind that fill rate has basically doubled between every true generation), I can't help but shudder. A game like HALO offers super-detailed lifelike models with so many polygons on them that they look better than most CGI sequences, seemingly making the GeForce256 *the* card to use due to the insane triangle throughput of the card.

However, HALO also has incredible lighting effects, water effects, explosions, etc. which all require several passes and tonnes of fillrate if you want to run at a high resolution. This is what the GF256 will not be able to provide in sufficient quantity, IMHO. HALO will probably be run at 640x480, maybe 800x600 resolution with this card, to provide a decent framerate with full effects (and who wouldn't want them?!) Now, if you have a big monitor (17", 19" +), switch your Quake resolution to 640x480, then 1024x768. Considerable difference, no? You can't notice the pixels as much since they are a lot smaller now, especially not the "jaggies". Jaggies are the boundaries between two objects, one overlapping the other. It's where the pixels of one texture or 3D object end, and the other begins - wherever there isn't a straight horizontal or vertical line, there are little "steps" where suddenly the object in the foreground gets wider, taller, lower, etc. You can see them pretty clearly and whether or not you are aware of it, it definitely takes away from image quality significantly - at least at lower resolutions.

Guess it's 640x480 for me…

Do you see the problem yet? Perfectly smooth circles and faces with tonnes of curves ... displayed by bigass pixels, or you cut out the special effects in favor of a higher resolution. Definitely not a pleasant choice considering you spent so much money on a supposedly next generation video card. However, while the GF256 has the next-gen hardware T&L and the quad pixels per clock, it doesn't have the next-generation fill rate.

Now we come to the obvious question/flame that will be thrown in my face. "But with 4 pixels/clock, you can get TWICE the fillrate for every MHz you overclock that you could get with a V3 or TNT2!!!" True, but here's the catch: you have to overclock it. With 23 million transistors (more than even the Athlon!!!), a high voltage (from the AGP bus) and .25 micron build process, the GF256 is very, very hot. This has been admitted by NVIDIA already, and do you honestly think they'd be running at only 120MHz if they had a choice (as compared to the 175Mhz reached by some TNT2 Ultras or 183MHz found on the V3)?

Second catch about overclocking: would you really risk it? The GF256 costs as much as mid/lower-end P3s. That is a lot of money you are possibly throwing away by trying to get performance that the card already isn't providing (face it, if you overclock, you're not happy with what you get, or you have little to lose - like Celerons :)