Slashdot Mirror
New ATi 3D Chip
Cooper writes "Saw a piece on Sharky Extreme about a new ATI chip dubbed the Rage 6, which they say is going to be used on Microsoft's x-box as well as for PCs. It's got an on-board geometry processor like the NVIDIA GeForce. " Wow. 2 gigapixels per second? Wow. The graphics market is starting to really heat up - check the earlier story about the Voodoo 4 &5.
Seriously, though, sheer pixels per second is kind-of meaningless. (Actually, it'll be state changes per second. :) In most vector-based or polygon-based product, the bottle-neck is in calculating the outer perimiter of the shapes.
Personally, I think pixel-based displays are a dead-end, anyway. Aliasing is horrible, and the techniques to get round it do so by making the picture too blurred to tell.
What's needed is a pure analogue polygon-based display, capable of area fill and non-trivial shapes. 3D would be nice, too. Anyone got a holographic projector they could lend me?
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
Of course, larger textures, more polygons and features like full-screen antialiasing and enviornmental bump mapping and true color are great, but it is hard to get worked up over any particular video card when each of them only has bits and pieces of the really cool features, and when it comes down to it, all they really do is allow some minor improvements in image quality and allow you to up your screen resolution and get more frames. Great for the obsessed gamer, but game developers still have to make games with the low end in mind, so the games themselves aren't tremendously more impressive from a graphical perspective.
For those reasons I'm a lot more excited about the Dolphin and PSX2 than I could ever be about a mere video card, or even a new class of video cards (like those with on-board T&L, which is meeting with mixed enthusiasm from developers btw.)
But all these incremental advances and all the competition is great. It means that maybe in 3 or 4 years, there'll be PC games where the game developers aren't limited by hardware anymore, but rather the "developmer's bottleneck" will be the designer's creativity, effort and resources. That would be cool. :)
I call it the Eyeball(tm) and I've patented it and GPLed it. It can do over 3 quadrillion equivalent pixels (I say 'equivalent' because it deals in quantum elements, of which there are an uncountable number in a pixel) per millisecond and when combined with another Eyeball(tm) in a dual setup, can actually create realistic stereovisual effects. It can take input from the real world and give it to you in astonishing 3D quality, and with the new additions 'LSD', 'X' and 'Louisville Slugger', you can make it generate colors you may never have seen before. Unfortunately, such advanced technology comes at a price. The incredibly complex nature of the Eyeball(tm) is such that it requires a proprietary socket, the EyeSocket(tm) to interface with your system properly, and of course you'll need a brain (we realize that this excludes a vast majority of the world's game players, but the ones who do have one will greatly appreciate this invention).
.. TODAY!
Don't settle for wussy video cards that are limited to only 'pixels' and may or may not be out after Christmas. Use the Eyeball(tm) (or two) and enjoy true reality
(please note that overclocking the Eyeball(tm) or removing the Eyeball(tm) from the EyeSocket(tm) in any way voids the warranty and may damage said Eyeball(tm))
- The former ("application processor") tends to involve grabbing bits of memory from here and there, comparing them to other bits of memory, jumping, often adding something, and sometimes calling subroutines.
- The latter ("graphics processor") will be doing a whole lot of operations involving XORs, filling regions with values, and even doing some tight loops oriented towards filling regions with "shading."
The graphics processor is rather more likely to find useful some operations that do "mass updates," which is rather like what a DSP does; that is quite different from the "lots-of-control-statements" that you'll get with a "conventional"/"application-oriented" CPU.That may be a gross oversimplification, but there you go...
The patents that Transmeta has been granted somewhat confirm this point of view; the patents represent ways of optimizing the emulation of those "lots-of-control-statements."
There may be a real killer graphics chip right around the corner, although it is easy to argue that the last three years have involved the continuous release of successive generations of "more-and-even-more-killer" graphics processors.
I'm not sure that there is a "Transmeta" of the graphics world; it's probably not appropriate to talk about such until next February when you might conceivably be able to buy some Transmeta product...
If you're not part of the solution, you're part of the precipitate.
We're entering a time when the only way to program 3D cards is manipulating the registers directly. There's no way you can get these cards supported by a library in time for the next generation of cards. We're leaving the age of abstraction and it's becoming more important for software to access the graphics hardware directly.