Gestalt takes a look at NVIDIA's next generation graphics card, the GeForce 256, which was launched in Europe at a press conference during ECTS.
The GeForce 256 raised a storm when it was announced recently. NVIDIA are claiming it's a giant leap for computer graphics. 3dfx are trying to write it off as one small step for NVIDIA. So, which of them is right?
On paper the GeForce 256 certainly looks very impressive. Here's an eye opener for you - the GeForce's processor has 22 million transistors on it. That's more than twice as many as a Pentium III processor, and more than the next generation Merced CPU from Intel is expected to have even. And that's still two years down the line!
According to NVIDIA the graphics card industry is moving four times faster than the CPU industry, and we're only ten years from being able to render scenes like those in Toy Story and A Bug's Life in real time on a desktop computer. Which is downright scary.
The biggest feature of the GeForce 256 is its hardware acceleration for Transform & Lighting, or T&L for short (not to be confused with T&A). T&L acceleration basically means that the graphics card doesn't just set up the scene and then output it to your monitor anymore, it also takes over the task of lighting the world and moving objects (and your view) around it.
The bottom line is that for games that support it, this will take a big load off your CPU. Given that the GeForce 256 is far more complex than your CPU and designed from the ground up to do this kind of work, it should be able to do it a lot faster than your processor.
Money Does Grow On Trees
The question is, how much faster? Well, NVIDIA claim that their GeForce 256 will be able to throw around something in the region of 15 million polygons a second. Let's say that you're playing a game at 60 frames per second, that means that each frame can have up to a quarter of a million polygons in it.
Which probably means nothing to you, until I tell you that Quake III Arena averages 10,000 polygons a frame. In other words, the GeForce 256 should be able to handle around 25 times as much detail as Quake III Arena is using! id Software have apparently decided to add a whole new detail level to Quake III Arena for the GeForce 256 to take advantage of, and even then it will only be using a fraction of its power.
The true implications of this were shown off by NVIDIA during the press conference. In one demo a tree was shown. As the camera zoomed in on the tree you could see individual leaves hanging on its branches, and as you got even closer you could see veins on the leaves. The level of detail was startling. As a joke, the demonstrator then turned the leaves into dollar bills.
Ok, this is just one tree, but NVIDIA also showed Bungie's "Halo" in action, as well as scenes from a demo prepared by the designers of "Experience", to show that the technology does have real uses. Again, the level of detail was impressive.
Another of their demos was the Viper demo, which renders a Dodge Viper car on the screen. The same demo was used for the TNT2 launch, but for the GeForce 256 the Viper now has more polygons in one wheel than the RivaTNT2 demo had in the entire car!
Putting the demo into wireframe mode to show just how detailed the model was, the demonstrator zoomed the camera in on one of the wheels, showing that not only was the wheel round, but the tread marks in the tire were now modelled instead of just being painted on as a "skin", and even the nuts that hold the wheel on to the car were now fully modelled.
In fact, as NVIDIA pointed out this level of detail essentially makes bump mapping (as seen on the Matrox G400) obsolete. Bump mapping is designed to simulate detail that isn't there, but with transform acceleration you can actually put the detail in as real texture-mapped polygons instead.
To demonstrate this, another of the demos showed water rippling on a lake. On a current graphics card those ripples would have been simulated with bump mapping, but on the GeForce 256 they were all made up of real triangles being rendered in real time by the graphics card. To prove this, the demonstrator again switched into wireframe mode where you could see that the surface of the lake was made up of thousands of tiny triangles, all moving up and down.
The Tomorrow People
Of course, as 3dfx are quick to point out, the problem with this technology is that no game really takes advantage of it so far. The good news is that we should start seeing games start to push the card's abilities within a year. The bad news is that a year is an eternity when it comes to graphics cards, and both 3dfx and NVIDIA will have new chips by then that will outperform even the GeForce 256.
What 3dfx don't want you to know is that T&L acceleration will actually speed up any game using standard OpenGL transforms, and that DirectX7 is going to support it as well. The problem is that none of the games available today have anywhere near the number of polygons they would need to really make the GeForce 256 sweat.
Just how much effect adding T&L acceleration to your system will have on frame rates in games like Quake II that already support it remains to be seen... It's likely that any increase in frame rates on older games will be more down to increases in fill rate than anything to do with the T&L acceleration though.
King Of Fill Rate
Speaking of fill rates, this is something that NVIDIA have been very quiet about. They're claiming 480 million pixels per second, which is almost three times as much as a Voodoo 3. But the more important texel fill rate isn't being widely publicised.
It seems that it will be around 480 million texels per second though, and 3dfx are claiming that this "low" fill rate makes the GeForce 256 unbalanced. After all, however many polygons it can throw around in its silicon, unless it has the fill rate to display it all on the screen at a decent framerate it's pretty useless.
What 3dfx hope you won't notice though is that at "just" 480 million texels per second, the GeForce 256 should still be 30% quicker than even a top of the range Voodoo 3 3500, as well as being able to handle much higher polygon counts.
Of course 3dfx's next graphics card will have an even higher fill rate, but the fact is that whereas 3dfx are still showing shoddy tech demos running on an array of 8 overheated Voodoo 2 cards, NVIDIA are showing real games working on real GeForce 256 graphics cards. It now seems certain that, for several weeks at least, 3dfx are going to lose their fill rate crown.
It's rather telling that 3dfx have had to resort to FUD to counter the GeForce 256. FUD ("Fear, Uncertainty, Doubt") is a tactic much loved by Microsoft, whereby a company tries to convince its customers that there's no point in buying its competitors products because a) they're not as good as they look, and b) we have something better coming out soon anyway.
3dfx have followed this pattern perfectly, first announcing their next card (code-named Napalm) and claiming that it would be out this autumn (which obviously hasn't happened), and then when the GeForce 256 was announced they started picking holes in the numbers NVIDIA have been claiming for the card, and have tried to persuade people that it is unbalanced and underpowered.
Amusingly, one of 3dfx's staff had come along to the NVIDIA press conference. When questioned about 3dfx's claims about the GeForce 256 by a journalist, NVIDIA's representative simply replied "They can have their own opinions".
When asked about 3dfx's next card, the NVIDIA man said he hadn't seen it yet and wouldn't try to comment on it, at which point the 3dfx rep called out that they were having a demonstration in a couple of hours, and he was welcome to come along and see it for himself.
"I've got better things to do", replied NVIDIA.
Despite 3dfx's attempts to FUD the launch, it's clear that the GeForce 256 is head and shoulders above the best that 3dfx have to offer at the moment.
The problem is that in a few months time 3dfx will be releasing their new "Napalm" card. Although we don't have any figures on that yet, it's pretty much certain that it will outperform the GeForce 256 in terms of raw fill rate. Napalm will also have full scene anti-aliasing, but crucially it won't have onboard T&L acceleration. And of course there's also S3's Savage 2000, which is shaping up to be a strong contender as well.
NVIDIA have certainly stolen the lead with their GeForce 256, but until we have a review board of our own to check, it's hard to say how much better than today's cards it will be. What is certain is that six months to a year from now, just when the first games that really take full advantage of T&L acceleration are starting to appear, the card will be surpassed by even better cards from both 3dfx and NVIDIA.
But that's life.