Slashdot Mirror
3dfx/Gigapixel: Where Did it Go Wrong?
nvidia3dfxatibobby writes "According to this interview & 3dfx tribute, gambling upon buying Gigapixel and then hoping to book a spot in Microsoft's Xbox is the theory or rather last gamble that failed and thus brought the curtain down on 3dfx interactive. Of course since then, NVIDIA booked their spot in the Xbox and 3dfx were left in the dust. The interview also looks at a former 3dfx employee's perspective now working for NVIDIA and hence talking for NVIDIA."
The detail currently achieved by 3d accelerator cards is good, but nowhere near "real". There is also a huge difference in realism between one generation of graphics cards and the next (not the case with adding color depth).
Let's look at a few examples of just how wimpy today's cards are, compared to what we would like them to be.
How many poly's do you need for a good tree? Here's an order of magnitude example: You might need 10 polygons per leaf times 10,000 leaves plus another 100,000 for the trunk if you want to get the bark right. That's 200 thousand polygons. If you want 30 frames per second, that's 6 million polygons per second. Of course you'll also want them to be lit and shaded. Most of today's graphics cards would be hard pressed to maintain this, so trees in games at this point are usually done using billboards...a couple of textured polygons. Even given this, when was the last time you saw a good dense forest in a video game?. We need another factor of 1000 in power at least, just to have a forest.
Today's cards use a lighting model called Gourad. It is a smooth vertex shading model. Unfortunately, it doesn't do shadows, reflections, per-pixel lighting, or any of the other things you see every day in real life. Stepping up to a new lighting model (such as raytracing) is beyond today's graphics cards.
Ever notice that even though the cards render in 3d, you have to describe the world for them in 2d pieces? Polygons and texture maps are planar. Only recently have we been able to get cards powerful enough (consumer level like Radeon) to do some simple 3d texturemapping. This ability to represent volumetric data is another thing that we need so that we can do good fog/ghosts/fire/smoke/explosions and a hundred other things without trying to fake them with an animation played on top of a polygon. Think about how much memory this might take too. If I have a big texture map, say 512x512 at 32 bit color, that is a one megabyte texture. What if I want a texture that is 512x512x512? Suddenly I need 512 Mb of texture memory. How close are we to this on today's graphics cards? The best have 64 Mb.
Don't forget that these examples are for simplistic things like trees and fire that we can't even do right yet! Perhaps you are right that eventually we will reach a point of visual perfection where no further improvements are needed, but we are far, far from it right now.
Stereoscopic imagery. Film resolutions (4096x3072 and such). Multihead cards. 64 bpp, maybe even 128bpp - 16 bits per gun, the rest for highly elaborate alpha and reflectivity tricks. Raytracing in realtime. There's plenty of room for expansion here.
And never, ever ignore the possibility of the Next Cool Thing coming along.
~ radiographite: art by john shepard
Physics is starting to be integrated in graphics hardware.
Accurately resolving collisions between several dozen rigid bodies in real time is a tax on even the most advanced hardware. Most games fudge physics horribly, and real physics just plain doesn't happen in real time for any kind of a complex scene with numerous non-uniform, variable elasticity and independently moving objects.
So long as graphics cards continue to take on real time physics, you can be sure that graphic simulation hardware has barely entered its infancy.
When 2D cards basically reached their maximum ability (Although I'm sure there could be additional improvement, but not too many people care these days) what you're talking about basically happened. It wasn't about features or speed anymore, it was about price. Why buy an expensive card when you could get one just about as good for $30 or so. But, graphics card developers went to 3D.
Once they can no longer advance that field (which won't be for a long time I'd wager), they'll find something else to work on, such as something worthy of being called virtual reality.. devices that implement touch as well as sight, etc. The advances could go on.
Generally, graphics cards are aimed towards the hardcore gamer. There are those cards for 3D graphics artists, engineers and architechs who use CAD/CAE programs, etc., but those are a completely different thing. As long as there are gamers who want the latest technology to further immerse themselves in the game, there will be planty of new advances to sell to them. If the technology stops progressing, they won't be able to sell many cards, because everyone will have one.
At least that's what I think.
Odd, that there are as many people out there, including /. who mourn the loss of 3Dfx...
The truth of the matter is, 3Dfx was, for a goodly 2 years, one of the first big name companies to start the IP onslaught... Less than 2 years ago, they attempted to sue Nvidia, claiming patent infringement, when Nvidia was just starting to make a name for itself, for providing low cost 3D accelerators, which at the time, was just a notch above ATI for preferred hardware...
Then Creative Labs came out with a Glide wrapper, custom made for their TNT/TNT2 cards... 3Dfx tried suing them as well...
Then people in the emulation scene started coming out with Glide wrappers as well... 3Dfx threatened them with cease and desist letters...
Meanwhile, while 3Dfx was running around threatening to sue everyone who could possibly compete with them, Nvidia continued to develop their technology...
Are we noticing a trend here?
Nvidia made one error in their PR dept, when they threatened to pull their sponsorship for hardware review websites, if any competing 3D accelerator ads/logos were displayed on said sites... They admitted the mistake, and for the most part, apologised...
So really, the biggest irony here is, with the general uproar regarding big companies stomping on the little guy for IP "infringement", that just such a company got bought out by one of the very little guys they were trying to stomp on...
And yet the majority of responses online (other than Nvidia's, of course) has been negative, attacking Nvidia rather than 3Dfx, making 3Dfx into the martyr, nay, the sacrificial lamb, gobbled up by the slathering wolves of Nvidia...
Just because you can mod me down, doesn't mean you're right. Shoes for industry!
What's missing in soundcards is quality 3D sound. The best was Aureal A3D, but that's vapour now that Creative succeeded in bankrupting them.
Creative offers EAX, which is predefined reverb effects. That's fine for ambient sound, but it doesn't provide good three-dimensional sound location.
QSound, Sensura and A3D all provide some form of 3D sound, well beyond the quad-speaker setup; they are particularly effective while wearing headphones.
What they do is simulate the delay, volume and "sound wrapping"/pitch shift effects experienced in natural sound. Part of these are obvious: it takes a fraction longer for sound to reach the distant ear, and the volume will be lower; others are less obvious, caused by the bending of the soundwave as it wraps around the head -- or, at higher frequencies, is blocked by the head.
Anyway, point is that it is difficult to simulate 3D sound. A3D had it: you could swear that the rocket went within an inch of your head; or hear the click of a trap behind you. It added a lot to the game experience -- it's like the leap from Wolfenstein to Quake III. An order or two of magnitude difference in realism.
And now that A3D is stomped, I'm very doubtful that we'll be hearing good game sound any time soon. It's like going back to Wolfenstein; sure, things will still be fun... but they won't be anywhere near as sweet.
--
--
Don't like it? Respond with words, not karma.
is what will happen to ATI and Matrox.
ATI is currently a much more profitable company. It has outsold NVidia by nearly double: ATI chipsets are spec'd in nearly every laptop and many OEM boxes. But it has recently lost Apple's contracts, and NVidia is making inroads into the laptop market.
Matrox ruled the world with its 2D cards. They were fast and, every bit as importantly, incredibly stable. If you were using specialty software, you could rely on a Matrox card to work with it. The same could not be said of *many* other cards: driver incompatibilities were assured with them.
But both ATI and Matrox appear to have dropped the ball. I don't see any truly kick-ass cards coming out of them. And while Matrox 2D is still top-of-the-heap, it's not enough any more: fast 2D can be done by anyone, and not enough people require driver stability to make that Matrox's saving grace.
What happens if either, or both, of those companies fold? We'll be stuck with the same sort of abysmal situation we have with soundcards: a complete lack of innovation or advancement during the past ten years. Creative Labs owns the soundcard market, and to this day we do not have advanced sound capabilities of any sort of respectable nature.
I'd hate to see that happen, particularly with 3D. The visual presentation is nearly as important as the aural presentation, when it comes to fooling the mind into believing in the virtual environment. If NVidia becomes near-monopolistic, we'll end up with the same mediocre performance and features as the SoundBlaster.
That'd be a crying shame.
We need competition. We might not get it.
--
--
Don't like it? Respond with words, not karma.
...and articles such as this one make me sick to my stomach. Why? Because most of them are full of ill-researched speculation, that couldn't be further from the truth, but nevertheless seem plausible-- so people believe them.
Now, this particular article on Sharky isn't so bad. I've seen worse. Brian Burke was with us almost to the end, so he knows what he's talking about.
Anyway, my point to this post is to clarify a few details that people tend to get wrong in these articles:
* nvidia did NOT buy 3dfx. Rather, 3dfx became insolvent, and so asked nvidia to buy their assets, so they could afford to dig their own grave. That is why my TDFX stock is work pennies today. If nvidia had acquired 3dfx, the stock would be going up, not down (because eventually, the stocks would be one-and-the-same.)
* 3dfx did not "refuse" to let OEMs sell their products. 3dfx WANTED to sell to OEMs (such as Dell, Compaq, etc.) That was the whole reason they changed their logo-- to look more professional! But Napalm and Rampage were woefully late, and 3dfx fell so far behind on the performance curve, that OEMs weren't all that interested. 3dfx's inability to meet the OEM's product schedules didn't help, either. In the end, all that was left for 3dfx was the retail side-- something 3dfx had wanted to de-emphasize, with the STB merger.
(In defense of some of the people who made the above claim I just refuted... by "OEM" some people might have meant board companies, like CREAF-- in which case, they are absolutely right; 3dfx did stop selling to those companies intentionally. But that's not what killed 3dfx; retail sales only accounts for a very, very small portion of the 3D graphics market. By far, OEM sales (Dell, Compaq, etc.) is where the money is at. This becomes more evident if you consider the fact that 3dfx became insolvent, despite having the top-selling products in the retail channel. Reason? Because retail is just a trickle, compared to OEM sales.
* The gigapixel merger is NOT what killed 3dfx. The gigapixel purchase was the smartest thing 3dfx did in a long time. But it was too little, too late. The purchase of STB is ultimately what killed them. The problem was, the business model changed on them. Prior to the STB acquisition, companies such as Dell and Compaq bought boards from board companies. But by the time the acquisition took place, the OEMs started buying chips directly from the chip suppliers, and then contracting companies to build boards overseas. STB was a middleman; our days were numbered, but we didn't realize it at the time. And neither did 3dfx, and they paid dearly for it. What we should have done (hindsight being 20/20 and all,) was sell or lease Juarez to someone like Solectron, and return to being a chip company. 3dfx realized this, but we realized it too late; Mexican labor is considerably more expensive than that of the pacific rim sweatshops, so Juarez's market value declined before we could sell it.
HSJ$$*&#^!#+++ATH0
NO CARRIER
I would guess that a similar future awaits graphics card technology. So on what criteria will Graphics cards of the future be judged upon? What will be the defining factors that will give one card an edge over an other once this graphical end of history has been reached? The only one I can think of is price.
KTB:Lover, Poet, Artiste, Aesthete, Programmer.
KTB:Lover, Poet, Artiste, Aesthete, Programmer.
There is no