Posted by
Hemos
on from the rolling-out-the-pretty-stuff dept.
Jonas writes, "I spotted that ATI has announced their next generation intentions where the 3D industry is concerned with their "Charisma Engine" and "Pixel Tapestry" technologies at this year's GDC. There's also an interesting article discussing the technology involved on their next gen 3D part. "
Moore's Law on Graphic Cards?
by
Anonymous Coward
·
· Score: 2
Do the graphic cards follow Moore's law? Or are they faster, or, slower in their evolution?
I sure need more power!
Re:Moore's Law on Graphic Cards?
by
be-fan
·
· Score: 2
Actually, the reason that graphics cards don't run at a GHz is not because the technology isn't there, (I think S3 had.18u fabs before AMD and Intel) but that the proc is a lot more complex in terms of pipeline. Its easy to pipeline a proc and make each step go very fast, thus you can complete that primative operation in x nanoseconds and drive the clock rate up to 1 GHz. For graphics cards, however, these primative ops take a lot more time thus you can only clock them at 100 or 200 MHz. Even though the GeForce uses something like a.2u process, it still can only clear 120 MHz, when a Pentium 120 did that witha.6u process.
-- A deep unwavering belief is a sure sign you're missing something...
Drivers
by
Anonymous Coward
·
· Score: 2
I'll say the same thing I've said since the ATi RagePro came out: cool, but will it take them less than a year to refine the drivers such that they'll actually take advantage of this?
Re:3dfx is actually a fraction of nVidia's size
by
Herbmaster
·
· Score: 2
If you're suggesting that the comparitive market caps have anything to do with the "size" of the companies as you refer to them in the first paragraph, you're mistaken. As a games player/developer, you don't especially care how much a company is worth on paper. The fact that nVidia is worth approximately half as much as ATI doesn't tell me anything about how many people actually use ATI or nVidia cards (I have no statistics about these amounts). Here's to hoping that 3dfx dies a miserable death.
I, like most gamers, don't care how many companies are making what they call "3D graphics cards". I think it was the S3 Virge chip that the company called a "3D graphics card" but the industry called a "3D decelerator", since actually using the card's hardware was slower than software rendering on a PII. I care how many companies are making good 3D graphics cards.
And that number's only going up. First there was the Voodoo, the Voodoo, and only the Voodoo. If you wanted a choice, you shopped between different cards with the same Voodoo chipset. If you wanted a high end card, it had TV out.
There was much rejoicing when the TNT came out and started heating up the graphics card competition... and even Matrox seemed to want to do 3D, although it took them until the G200 before they even had an OpenGL driver using their card... but now they have the G400, and their card's actually good at it. For people buying a new system, it's generally worthwhile to look at the Voodoo 3, TNT 2, G400, GeForce(of course), and ATI cards... and that's a good thing.
ATI sure is talking a good game, but do they have what is takes to back it up. NVidia did three things that changed the video card industry:
#1 Pruduced the fastest, high quality consumer oriented video chipsets
#2 Consistently met demand in the OEM and retail market
#3 Produced better reference drivers than chipset implementers could produce, and updated these drivers for advancements like Direct X very quickly.
ATI has shown it can produce a good video solution, but lacks in meeting retail market demand and driver support. Matrox builds awesome chipsets and cards (and excellent driver support) but doesn't give a damn about meeting demand. 3DFx kept their 3d spec closed therby limiting potential developer support, lost momentum, and didn't provide good reference drivers.
NVidia has proven they can do all three consistantly. And let's not forget #4 -Support from Software developers. Metting the first three criteria directly impact the fourth. Developers don't waste their time developing for hardware no-one owns.
-- "God fights on the side with the best artillery." - Napoleon, Marshal of France
- speaking truth to power
And how does it compare to the Voodoo5? It's announced on 3DFX's website but I don't remember seeing any review yet. Anyone knows the status of this baby?
"We have placed orders for production silicon already. Our software development is right on track. We are on the same release schedule as when the VSA-100 product was introduced at Comdex, which we stated would be in the Spring. That product will include all the features that have been promised. It will deliver real time, full scene anti aliasing. It will support dazzling cinematic effects via our t-buffer. It will feature 32-bit color depth, SLI implementations and astronomical fill rates. Despite the outstanding state of this first silicon, the boards used in the Cebit demonstrations do not represent production silicon. Shortly after GDC, we expect to be demonstrating Voodoo4 and Voodoo5 boards that are much closer to production quality."
The GDC is being held right now, March 8-12, so we should be getting some reports soon. Right now it looks like 3dfx is shooting for late April or May.
3D rendering is an easily parallelisable process. I remember, early Voodoo cards could use 2 Voodoo chips, one would render even lines, the other odd scan lines. There are a few bottlenecks to parallelisation, and mostly it's the texture access, but with cheap memory you should be able to cache that.
Depth of Field effects in 3dfx's new Voodoo5
by
Guppy
·
· Score: 2
Depth of Field effects will be supported in hardware by 3dfx's Voodoo5 series, scheduled to be be released sometime this spring (most likely in late April/early May). The Voodoo5 has a feature 3dfx calls the "T-Buffer", basically an accumulation buffer which can also be used for Full Screen Anti-Aliasing, Motion Blur, Soft Shadows, and Soft Reflections. All effects require the software to support the feature, except for anti-aliasing, which can be done automatically.
Re:Depth of Field effects in 3dfx's new Voodoo5
by
Guppy
·
· Score: 2
This is all fine and dandy, but this is hardly hardware support for things like depth of field. It requires that you render the scene n-times just to achieve a single frame. Real hardware support (for depth of field) would allow you to specify the focal length and the hardware would automatically blur non-focussed areas automatically.
I'm not too familiar with the details of 3dfx's approach (not being a 3d programmer myself), but my understanding is that what the VSA-100 does is use hardware method for producing "jittered" samples slightly off from the target pixel, which are then blended together in the accumulation buffer.
The "T-Buffer" effects can then be specified to be applied to certain objects, but not others. If a sub-pixel jitter is specified, then you get anti-aliasing. A larger jitter gives softening, and a very large jitter gives blurring. So far as I can tell, a program does not need to perform"n-renderings". The program still needs to specify the object (Or it can specify a mask and just do an area, I think), and the degree and quality of jittered samples, but from their I believe the chip does the rest automatically.
While the 3dfx "T-Buffer" affects are nice, I can't imagine developers using them, since it would lock the developers into certain hardware (again!). This is something that I thought was insane when Glide first came around. These type of proprietary extensions are exactly why I have never and will never buy 3dfx-only games or 3dfx video cards. You said it above, "basically an accumulation buffer", instead of using this type of thing, why not make an open extension to OpenGL/Direct3D to support these type of effects.
When the Voodoo 1 was first introduced, competing accelerators included the S3 Virge (The most numerous by a wide margin), the Rendition Verite, the nVidia NV1, etc... Good, fast OpenGL support was still the providence of ultra-high end professional accelerators, while Direct3D (Still below version 3 at that time, I think) was slow and glitchy. As a result, in addition to OpenGL and D3D, *everyone* at the time everyone had their own proprietary API which matched their own hardware closely, thus giving a sometimes substantial performance boost. nVidia had one for their NV1, PowerVR was pushing their PowerSGL, and I sort of remember S3 and ATI had their own as well. You still see PowerSGL support in Unreal, and I think S3 still pushes programmers to use Metal for the Savage2000, but of all the cards from that era only the Voodoo1 is still somewhat fast enough to be useable today, and only Glide is still programmed for.
Glide survived because it appeared at a time when consumer-level 3D was first starting to appear, plus the hardware was good, plus it was fast, and doubleplus because it was easier to learn and program for than D3D at the time. I think 3dfx held onto Glide far too tightly for too long, but it's existence is due more to history than any monopoly power on 3dfx's part (In fact, ATI and S3, both then and now, are each several times larger than 3dfx, both in terms of market cap and number of card shipped. nVidia has something like 5x the marketshare that 3dfx has.)
Re:3dfx is actually a fraction of nVidia's size
by
Guppy
·
· Score: 2
"If you're suggesting that the comparitive market caps have anything to do with the "size" of the companies as you refer to them in the first paragraph, you're mistaken."
I'm well aware of that, but I didn't have any hard statistics on hand for the more interesting data, like marketshare. I was researching this a few days ago, but I can't seem to find my source again.
Anyway, skipping the hard statistics, in terms of 3D accelerator market share (IE, who sells the most $$ worth of cards/chipsets), nVidia is number one. I forget who comes next, but I believe it's ATI, then S3, then Intel (Big with OEMs). Then way in the back comes everyone else, including 3dfx and Matrox. 3dfx has a pretty strong retail presence, but that's a small slice compared with OEM pie.
Here's to hoping that 3dfx dies a miserable death.
I have a hard time understanding the anger directed towards 3dfx. They don't have any monopoly power over the market (Never did), have released just about all of their source code, and their cards offer a pretty good bang for the buck.
At any rate, you could very well get your wish. 3dfx has been experiencing severe and accelerating losses for the last few quarters. They just had a layoff a few weeks ago. At the current rate, they only have enough cash to last for a year or two.
And one less 3D company means less competition in the marketplace. In the past few years we've seen a huge number of companies leave the field--Tseng Labs (Out of Business), Cirrus (Now doing audio/modem chips instead), Trident (Still around but miniscule), Real3D (Remains bought by Intel), Rendition (Remains bought by Micron), Hercules (Remains bought by Gullimot), Number9 (Still around, but just a brand that sells S3/nVidia chips), and Chromatics (Bought by ATI). I think Permedia might be out too.
That's a pretty big number of companies that used to design chips, but no longer. Now everbody else like Diamond and Creative just slaps a label and an S3/nVidia chip onto a board. A lot of industry analysts think the consolidation is going to continue.
Think nVidia wouldn't try to establish a lock on the market if they get big enough? Intel, ATI, and nVidia have all been looking at integrated chipsets--in the short term as a low-cost part, but in the long term as a possible way to get that lock. And their investors seem to like the idea.
ArtX has *serious* ethical problems (Article Link)
by
Guppy
·
· Score: 2
The article details what happened when Jon "Hannibal" Stokes, a writer for Ars Techica, posted a negative article on an ArtX trade show appearance. Afterwards, a number of Anonymous posts appeared on the Ars Technica forum which appeared to support ArtX, but which turned out to be from an ArtX's Director of Marketing.
Re:What happened to pixel volume rendering?
by
Guppy
·
· Score: 2
There are persistent rumors on investing boards that several companies are working with Voxel acceleration. One particularly interesting rumor concerns 3dfx's Rampage chip, scheduled for the end of this year. In one interview with 3dfx's European Product Manager, Luciano Alibrandi, the interviewer asked if 3dfx was working on Voxel technology. Mr. Alibrandi replied "No"--but several days later the interview was updated at 3dfx's request, with the "No" struck out and replaced with a "Can't Comment".
Anyway, we may find out if any of the rumors are true at the Game Developer's Conference that is taking place March 8-12.
Stupid Idea; Port Linux to a Graphics card?
by
Anonymous+Shepherd
·
· Score: 2
Is this even possible or feasible? What with cards with 64mb+ of memory, 'GPU', etc.
IE, framebuffer and display is no problem. Data would be loaded through a simple, stupid, microprocessor across the AGP bus; all you'd need. I'm sure there are Linux distros that could fit themselves comfortably within 64mb ^^
I agree with the curved surface rendering, and collision detection, but not the rest.
Depth of field is not appropriate for interactive games. In RL you refocus your eyes to look at different things, if you can't refocus just by controlling your eyes, you'd be half blind. It'd drive people crazy.
Integrated physics would lock the programmer into a certain physics model. Physics is not terribly CPU intense, and the demands vary a lot from game to game. Having specialized physics hardware on the video card is about as appropriate as having specialized AI hardware (IOW, it's not).
Voxels are either huge memory pigs or butt ugly. They might make nice 3d texture maps (if you're okay with fuzzy interpolation), but I wouldn't want to bother with them for whole 3d models.
Chromatics are a waste. They are so rarely useful that it would be better to special case the lighting effects when needed.
Radiosity would be nice, but it's not something you can just pipeline in (ditto for casting rays). However, there might be cheaper ways to get the same effect.
Collision detection is a no-go too. It would require a feed-back mechanism (the card says: Ok, there was a collision, now what?) which would cause the card to stall. 3D hardware is good as output devices only - you get everything set up and then send it off to the card to render. Collision detection sounds nice because it is geometry based and 3D cards seem so good at doing things with geometry, but, the app is also doing a lot with geometry so it makes sense to do it in the app as well. Plus, collision detection can mean different things in different situations. Safer to leave it on the app side. Let the card worry about gettings out to the display as quickly and as nicely as possible
To look into the future of consumer 3D one might want to look at high-end companies like SGI. Their machines can do cool stuff with the various buffers (i.e. render into texture memory, mutliple, independently controllable paths, etc.)
Finally, in reference to nVidia vs. ATI. It seems that ATI has always scrambled to get competitive products to market (good marketing and channels though), whereas nVidia has been following a well-controlled technology curve and introducing innovative products (for the consumer market) that are well-rounded and work. Following this trend, I'll bet that in 6 months nVidia will have a good solid product with usefull features, but ATI and 3dfx will have products with quirky features and will be of questionable quality (how 3dfx could get away with saying that 16bpp is "good enough" and all we really need or want for so long is beyond me). Example: hardware T&L at the consumer level is truly useful (placing a major part of the rendering pipeline onto the card!!) whereas FSAA, which is very cool, is nothing more than oversampling. Technically, it can be done on any general-purpose 3D graphics which supports an accumulation buffer. I hope that 3dfx can do something useful besides pushing fill-rate, and I hope that ATI can come up with a truly powerful and timely product, but history doesn't bode well for these two. I'd love to be proven wrong by either company.
If collision detection was done on a separate chip on the card and not in the main CPU, it could be very helpful. You send the object you want to test and its potential positon relative to a "reference point" in the object and the card replies with a 1-bit yes or no, maybe more info, but I think that'd be enough. This way the card would never stall. Also, the collision detection wouldn't be one frame behind. The separate chip would do a lookup into the same memory the main CPU is using but wouldn't actually be testing based on that but the coordinates you send. Already matrices are accelerated (ATI is doing it in their new card with their bump mapping) in hardware, and that's all this would be would be a transformation matrix being applied and then a test of collision. I think it could work...
because I don't understand how putting a video card on a remote server could possibly speed up the rendering on a local machine?
-----------
"You can't shake the Devil's hand and say you're only kidding."
Re:waiting for the iMac comment, Carmack?
by
imac.usr
·
· Score: 2
I remember reading recently (on MacOSRumors; make of that what you will) that Apple was so pissed off at ATi for delays in the mobile Rage128 in the new PowerBooks that they're taking another look at companies like nVidia for OEM support.
IIRC Apple did rewrite parts of the RagePro driver library for the Mac, although I don't know if they're working on Linux versions as well. I'm guessing they're working very hard on BSD versions, though.
The G4 and even the newer iMacs make a quite decent Q3A platform thanks to the Rage128, but they still lag behind the latest PC video cards. Hopefully Apple will persuade developers to write the appropriate drivers for OS X so you can stick any AGP card in a G4 and have it work out of the box.
--
I use Macs for work, Linux for education, and Windows for cardplaying.
Re:Where next for high-end graphic cards?
by
hedgehog_uk
·
· Score: 2
There must be others
How about Phong instead of Gourard shading? Fast Phong algorithms for hardware implementation have been about for years. They're still more computationally intensive than Gourard but remove the need for specular texture maps and reduce mach-banding artifacts.
Real-time radiosity? Not for a very long time, methinks. Radiosity is usually pre-computed. I remember reading one of John Carmack's.plans where he said that ID had bought a supercomputer (a Sequent?) to perform the Quake radiosity calculations.
HH
Yellow tigers crouched in jungles in her dark eyes.
-- Yellow tigers crouched in jungles in her dark eyes. She's just dressing, goodbye windows, tired starlings.
Re:3D Texture mapping comments
by
hedgehog_uk
·
· Score: 2
generate the appropriate texture volume and then put that in the image cache with the standard 2D versions
As I mentioned elsewhere in this discussion, precomputed 3D texture maps would take up vast amounts of memory on your video card. IMHO it would probably be better to let the CPU compute the procedural textures and transfer them to the card using AGP.
Better still, provide a texture compiler that produces bytecode that can be executed directly by the card. Now that would be cool. Procedural displacement mapping (like RenderMan uses) would be ultra-cool.
So the first 3D card that can execute RenderMan shader bytecode will get my money:-)
HH
Yellow tigers crouched in jungles in her dark eyes.
-- Yellow tigers crouched in jungles in her dark eyes. She's just dressing, goodbye windows, tired starlings.
Nice to see support for 3D textures. These are very cool. The article says:
Traditionally, polygons are used to represent 3D objects. However, with 3D textures, volumes of texels (textured pixels) may also be used. In a 2D texture map (the kind that we see "glued" to a wall for instance) indexing occurs via two texture coordinates, whereas in a 3D texture, there are three coordinates.
One good example of 3D texture use would be that of a marble cube. If the corner of the cube were to be chipped off, any veins running through the marble would already be defined and visible without any additional textures being generated.
This means that you could chop a block of wood up, and have the wood grain on the cut surfaces rendered correcly. However, the article then goes on to say:
Unfortunately, we feel 3D textures will have to be used incredibly sparingly because in order to implement the marble cube example explained above, an artist would have to draw the entire 3D surface (including the veins inside the cube which may never be seen!).
This is incorrect. How can an artist draw the inside of a solid cube of marble or wood? I've never heard of a 3D texture being created in this way. They are normally generated procedurally, where you have a function that mathematically calculates the texture colour given the x,y,z coordinates within the texture. This does mean that you can't store 3D textures on the card, unless you pre-calculate an array of texels, but this would require vast amounts of texture memory on the card.
HH
Yellow tigers crouched in jungles in her dark eyes.
-- Yellow tigers crouched in jungles in her dark eyes. She's just dressing, goodbye windows, tired starlings.
Charisma Engine? Emotion Engine?
by
AugstWest
·
· Score: 2
Is this hardware, or an election? Honestly, this is just ridiculous. Somebody tell ATI that hardware can't have charisma. Somebody tell Sony that hardware can't have emotions.
Yeah, that's me, trendspotter extraordinaire. Takes a genius these days, eh?
Re:Charisma Engine? Emotion Engine?
by
crivens
·
· Score: 2
Why not? The hardware has to have emotions and charisma, cos most game players and geeks don't.
I'm kidding. It was a joke.
Re:Where next for high-end graphic cards?
by
Esperandi
·
· Score: 2
Wile all your suggestions are good, i think a rel photo-realistic engine will be quite different. it will basically be POVRay on a chip or perhaps renderman since I think the most appropriate thing would be to forget bump-mapping and go with a shading language that actually deforms the geometry of the object and then does true ray-tracing.
What comes after that? Well, I have to fight hard to keep my eretion down when thinking of this... Hardware-based realtime radiosity. *uNF*
BTW, the idea of hardware-based collision detection... I hope to GAWD that the hardware manufacturers out there, nVidia in particular (cause I haven't read what they're doing after the geForce256, everyone else has something in the public works) read your post. It would be most gorgeous and possible to have such a thing...
Esperandi
Full-screen AA - the geForce does it
by
Esperandi
·
· Score: 2
This was queitly done recently, but in the most recent geForce256 drivers, Fullscreen AA can be enabled with a little registry twiddling. From what I read there is a "performance hit" but I'm not sure how significant it is...
Esperandi
Re:3D Texture mapping comments
by
Esperandi
·
· Score: 2
Another problem I had with this article along the same lines is his view (as is POVRays current view, but this is fixed in Megapov and will hopefully make it into the next release) that bump-mapping is the same as surface-deforming textures. Sure, it might look the same on a screenshot of a bumpy pear sitting on some blistering wood (the screenshot on sharkyextreme), but when you get into a game, you gotta keep the bumps really small. Imagine you've got the wall of a dungeon. You think it'd be easier to make a plasma-fractal generated bump map instead of breaking the triangles on every single section of the wall (which I'm planning on doing in an upcoming game;)... well, when your users get into the game and start walking down this really kickass-looking craggy hallway, they're going to notice something. The crags aren't there. They can walk right through them. Which would be laughed at and viewed as very annoying by gameplayers. (it obscures your view and its ethereal?!) If REAL geometry-deforming maps were able to be applied, collision detection could work, it would keep the polygon count down and the game would be wickedly fast... but POVRay help off for such a long time, I fear if every card supports this fake bump-mapping (it takes the light and just moves the point and texture maps it a little different based on the new virtual position) that game developers will get happy with it and never realize how versatile real surface deforming textures can be... Combine it with a separate collision detection engine and we could have some amazingly realistic games...
This card that ATI is making is wondrous. I mean, stunning. However, until a few days ago I owned an ATI All In Wonder Pro. I didn't get it for its 3D because I knew it sucked, i got it for the video capture capabilities which it had in abundance. Only a dedicated Videum card I used to have performed better. I soon found out that ATI has absolutely horrid drivers. I mean, this card never worked right even when I got super bleeding edge beta drivers (they were beter than the released ones and more stable I don't know why they weren't public, they gave me a secret URL, forbade me from dispensing it, etc). I would capture a video using their proprietry VCR2 codec (more efficient codec than I had ever seen or have ever yet seen!) and then go to edit it. Playing the video would work, but seeking in any sense would crash the drivers. To this day (and the all in wonder pro is 2 or 3 years old) they have not fixed the problem.
I honestly hope that they support this card well and do a good job with the drivers because even if you've got the best card, its only as good as its drivers.
Although a lot of people I talk to think Matrox or 3Dfx are to be the biggest competitors for nVidia, I think ATI is a more worthy opponent: they're huge, since they're selling about a billion of chipsets daily to a lot of large OEMs (Gateway, Dell) so they've got the financial strength to take on nVidia's roadmap. And besides, their previous efforts (like the dual Rage Pro Maxx) were already quite good.
I hope they'll use this chipset to target the hardcore gamers and start a good battle against nVidia's supremacy. Like the AMD vs. Intel thing, us consumers will only benefit:)
3D Texture mapping comments
by
Mithrandir
·
· Score: 3
Unfortunately, this article highlight's the author's shortcomings in understand what a lot of high-end 3D graphics is about and how it is implemented.
The one major thing the author misses about 3D texture maps is that rarely are they hand drawn by an illustrator. A typical map is a procedural texture (think of rendering a marble texture using POVRay) so generating a lump of marble is not that difficult a thing to do.
For games, the programmer just needs to fire up Povray, 3DS etc and get it to generate the appropriate texture volume and then put that in the image cache with the standard 2D versions. I'm sure a lot of game engines will handle this pretty quickly.
--
Life is complete only for brief intervals in between toys or projects -- John Dalton
Charisma and Emotions explained.
by
doomy
·
· Score: 3
OTOH,
I feel that those two names come real close to describing these two very excellent egines as best describle on earth:) (this was before the destruction of earth by the vogons).
Lets see..
On Charisma, David Jenson wrote
When scientists and technicians hear the word charisma, they may first think of sales reps or politicians. But you'd be hard pressed to find a person in any influential biotechnology position who doesn't have some measure of charisma. Those on the scientific track are not exempt from this need.
Charisma is derived from the ancient Greek word kharis, meaning "to cause to strive or desire." The ability to motivate others to strive and succeed is a major building block of successful management, whether in a QC lab or in a corner office. Charismatic people describe goals by painting word pictures, thereby motivating others to a particular end. They have an exceptional ability to win the devotion and support of others. They have no fear of presenting their ideas to anyone who may be able to help them. And they have excellent persuasion and negotiation skills.
But more to the effect, I see charisma here derived from the Indian (as in South Asia) word. It too has a similar meaning to the regular charasmatic word. In this way, the word comes closer to a powerful healing force that is being ispearsed around the subject than anything else. This is a very visual word. A very charasmatic word. The word conjures a halo around it's subject and renders it in a light that leaves a very strong impression on anyone who hears it uttered. Thus, it is fitting a name for this chip (Which I believe would live up to this name). As would, the emotion engine in PlayStation 2. Which also conjures strong vibes and powerfully drawn meanings to the word and what the chip can do. Human emotions are powerful, machines, from the start of time (execpt for Marvin) are known to lack them. The very thought of a machine having these very emotions drive a very strong stab at anyone who looks at the PS2. The engine was made for it's artistic quality, it's ability to render something beautiful, so beautiful that it is almost real, that is the emotion, the charisma. --
-- ...free your source and the rest would follow...
Another article, plus ATI's Charisma White Paper
by
Guppy
·
· Score: 3
3dfx is actually a fraction of nVidia's size
by
Guppy
·
· Score: 3
Everybody likes to compare nVidia and 3dfx as the two top companies, but in reality 3dfx is a small fraction of nVidia's size. I don't have exact numbers offhand, but nVidia currently has about 45-50% of the graphics market while 3dfx has something like 10-15%, and I believe Matrox is even smaller than 3dfx.
Here's a comparision of some market caps (data from The Motley Fool).
ATI: 4,141.51 million S3: 1,607.80 million nVidia: 1,808.46 million TDFX: 218.09 million
Re:What happened to pixel volume rendering?
by
Junks+Jerzey
·
· Score: 3
Any game that uses OpenGL for more than a rasterization-only API automatically uses geometry acceleration on cards/drivers that support it.
Realistically, the boost is less than you may think. An average game doesn't spend more than 15% of a frame doing transformation. So the Ultra-Fast-Geometry-Accelerator-of-the-Future is going to buy you a 15% speedup in that case. The other issue is that geometry acceleration is only useful when you pump the data straight to the card and don't want or need intermediate results. For example, you'll have to transform points (one way or another) to do collision detection against instanced objects. But you can't use the geometry acceleration in that case, because the CPU needs the results.
Geometry acceleration is good, but it's not the panacea that many people are expecting it to be.
Re:time to rethink for intel/AMD/etcetc
by
Microlith
·
· Score: 3
That card's core is a highly specialized chip, focusing on the math for the transformation of points, lights, textures, blurs, and several things that don't involve most of the core of an x86 chip. Linux will never run on it unless they make one with a big flash rom, and a general purpose controller chip (read: i960 or other).
What kind of use is the card? LOTS! Go check out the Intergraph Intense3D Wildcat 4110. It runs in most prebuilt p2/p3 graphics workstations (huge card), and takes so much of the processing off of it, the only thing the cpus are needed for is getting the software started, and other extended math calculations (we love those fcurves!), and rendering of the final image. By the way, this card does everything the geforce AND v5 do... but i'm not sure as to it's fillrate, but 21fps in a scene with 80000 polys is impressive. And game cards are catching up quite quickly to the power of the "industrial card".
Acelleration... What needs to be done is the accelleration of the front side bus. It's just poking along at roughly 133mhz now, maybe 200 on athlon systems (but that's only ram to cpu). It'd be better if that were 1/1 with the cpu speed, or if it were faster, leaving the cpu with a wide open pipe to the ram and the other system peripherals.
Totally not answering the question....
by
luckykaa
·
· Score: 4
The 3d card market seems to have an alternative version of Moore's law:
Every 18 months, the number of people making 3D graphics cards halves. There's only about 6 companies making 3D chips now.
More features for no one to support
by
Junks+Jerzey
·
· Score: 5
Speaking as a game programmer, these advances are coming so fast that there's no time to concentrate on (1) pushing the limits of what a current generation of cards can do, and (2) dealing with card-specific features.
On the first point, there's not enough time to sit down and focus on where all the rendering time is going in a complex game. Well, more like there are so many card and driver combos out there that the best we can do is try to write generic code and have it work across the board. If we could focus on one card, say a Voodoo 2, then we could push the limits of that chipset out beyond what people only expect from a GeForce. But there's no time for that, so we plow ahead using about 50% of each card's capabilities for the three month window until the next card comes along.
On the second point, 3dfx, Nvidia, Matrox and ATI (and S3, and...) are all branching out into odd and card specific feature sets. 3dfx has their T-buffer. Nvidia has "8 lights per triangle hardware lighting." Matrox has a certain kind of hardware bump-mapping. ATI has all sorts of wacky stuff. The bottom line becomes "Do we want to just focus on writing a great game, or do we want to spend an extra six months of development so we can support special features of all these cards that were considered hot eight months ago when we were still pre-beta?" And tacking in special Matrox-only support, for example, is hell on QA. It makes a lot of sense to ignore such features, unless we're getting a bundle from the card company to cover us for the trouble.
Where next for high-end graphic cards?
by
tjwhaynes
·
· Score: 5
It seems that while the push for ever increasing image quality is going on, we are getting much closer to realistic, real time rendering of scenes. I wonder just what else is needed to really be able to push the envelope of visualization and realism further. Here's my current wish list.
Proper curved surface rendering - not just pushing the polygon count ever higher but actually rendering, for example, bezier patches with multi-pass textures.
Depth of field - most graphic cards today blur the insides of polygons when they are close (tri-linear mapping) but do nothing to blur the edges of the polygon, breaking the realism. And everything in the far field is in clean focus. Having real depth of field, so that there is some defined focal distance would help.
Integrated collision detection - we pass the cards all these vertex coordinates, fans and strips. It must be possible to pass some of the collision detection from the CPU to the graphics card. Using something like Orientated-bounding boxes at various detail levels and then passing the final collision detection to the card for some arbitration at the polygon level might help.
Integrated physics engine - gravity, flexion, distortion both plastic and elastic, hinges, rotation and friction. And anything else:-)
Volume rendering - either voxels per se or some iso-surface rendering based on potential fields.
There must be others - it looks like ATI is going to finally give us proper bump mapping and range-based fogging. Do we also need a proper chromatic model so that we can get rainbows through glass objects? Should there be real-time ray-casting or radiosity support so that real lighting effects (say carrying a flaming torch down a corridor and having proper soft-edged shadows) can be achieved?
Cheers,
Toby Haynes
-- Anything I post is strictly my own thoughts and doesn't
necessarily have anything to do with the opinions of IBM.
Slashdot Mirror
Do the graphic cards follow Moore's law? Or are they faster, or, slower in their evolution?
I sure need more power!
I'll say the same thing I've said since the ATi RagePro came out: cool, but will it take them less than a year to refine the drivers such that they'll actually take advantage of this?
If you're suggesting that the comparitive market caps have anything to do with the "size" of the companies as you refer to them in the first paragraph, you're mistaken. As a games player/developer, you don't especially care how much a company is worth on paper. The fact that nVidia is worth approximately half as much as ATI doesn't tell me anything about how many people actually use ATI or nVidia cards (I have no statistics about these amounts).
Here's to hoping that 3dfx dies a miserable death.
I'm not a smorgasbord.
I, like most gamers, don't care how many companies are making what they call "3D graphics cards". I think it was the S3 Virge chip that the company called a "3D graphics card" but the industry called a "3D decelerator", since actually using the card's hardware was slower than software rendering on a PII. I care how many companies are making good 3D graphics cards.
And that number's only going up. First there was the Voodoo, the Voodoo, and only the Voodoo. If you wanted a choice, you shopped between different cards with the same Voodoo chipset. If you wanted a high end card, it had TV out.
There was much rejoicing when the TNT came out and started heating up the graphics card competition... and even Matrox seemed to want to do 3D, although it took them until the G200 before they even had an OpenGL driver using their card... but now they have the G400, and their card's actually good at it. For people buying a new system, it's generally worthwhile to look at the Voodoo 3, TNT 2, G400, GeForce(of course), and ATI cards... and that's a good thing.
ATI sure is talking a good game, but do they have what is takes to back it up. NVidia did three things that changed the video card industry:
ATI has shown it can produce a good video solution, but lacks in meeting retail market demand and driver support. Matrox builds awesome chipsets and cards (and excellent driver support) but doesn't give a damn about meeting demand. 3DFx kept their 3d spec closed therby limiting potential developer support, lost momentum, and didn't provide good reference drivers.
NVidia has proven they can do all three consistantly. And let's not forget #4 -Support from Software developers. Metting the first three criteria directly impact the fourth. Developers don't waste their time developing for hardware no-one owns.
"God fights on the side with the best artillery." - Napoleon, Marshal of France - speaking truth to power
And how does it compare to the Voodoo5? It's announced on 3DFX's website but I don't remember seeing any review yet. Anyone knows the status of this baby?
3D rendering is an easily parallelisable process. I remember, early Voodoo cards could use 2 Voodoo chips, one would render even lines, the other odd scan lines. There are a few bottlenecks to parallelisation, and mostly it's the texture access, but with cheap memory you should be able to cache that.
Depth of Field effects will be supported in hardware by 3dfx's Voodoo5 series, scheduled to be be released sometime this spring (most likely in late April/early May). The Voodoo5 has a feature 3dfx calls the "T-Buffer", basically an accumulation buffer which can also be used for Full Screen Anti-Aliasing, Motion Blur, Soft Shadows, and Soft Reflections. All effects require the software to support the feature, except for anti-aliasing, which can be done automatically.
"If you're suggesting that the comparitive market caps have anything to do with the "size" of the companies as you refer to them in the first paragraph, you're mistaken."
I'm well aware of that, but I didn't have any hard statistics on hand for the more interesting data, like marketshare. I was researching this a few days ago, but I can't seem to find my source again.
Anyway, skipping the hard statistics, in terms of 3D accelerator market share (IE, who sells the most $$ worth of cards/chipsets), nVidia is number one. I forget who comes next, but I believe it's ATI, then S3, then Intel (Big with OEMs). Then way in the back comes everyone else, including 3dfx and Matrox. 3dfx has a pretty strong retail presence, but that's a small slice compared with OEM pie.
Here's to hoping that 3dfx dies a miserable death.
I have a hard time understanding the anger directed towards 3dfx. They don't have any monopoly power over the market (Never did), have released just about all of their source code, and their cards offer a pretty good bang for the buck.
At any rate, you could very well get your wish. 3dfx has been experiencing severe and accelerating losses for the last few quarters. They just had a layoff a few weeks ago. At the current rate, they only have enough cash to last for a year or two.
And one less 3D company means less competition in the marketplace. In the past few years we've seen a huge number of companies leave the field--Tseng Labs (Out of Business), Cirrus (Now doing audio/modem chips instead), Trident (Still around but miniscule), Real3D (Remains bought by Intel), Rendition (Remains bought by Micron), Hercules (Remains bought by Gullimot), Number9 (Still around, but just a brand that sells S3/nVidia chips), and Chromatics (Bought by ATI). I think Permedia might be out too.
That's a pretty big number of companies that used to design chips, but no longer. Now everbody else like Diamond and Creative just slaps a label and an S3/nVidia chip onto a board. A lot of industry analysts think the consolidation is going to continue.
Think nVidia wouldn't try to establish a lock on the market if they get big enough? Intel, ATI, and nVidia have all been looking at integrated chipsets--in the short term as a low-cost part, but in the long term as a possible way to get that lock. And their investors seem to like the idea.
Speaking of ArtX (Which ATI recently bought), there's an interesting article up at Ars Technica, "ArtX: Half-truths and Misrepresentation?".
The article details what happened when Jon "Hannibal" Stokes, a writer for Ars Techica, posted a negative article on an ArtX trade show appearance. Afterwards, a number of Anonymous posts appeared on the Ars Technica forum which appeared to support ArtX, but which turned out to be from an ArtX's Director of Marketing.
This incident appeared on Slashdot as ArtX, Hannibal and Consumer Fraud.
There are persistent rumors on investing boards that several companies are working with Voxel acceleration. One particularly interesting rumor concerns 3dfx's Rampage chip, scheduled for the end of this year. In one interview with 3dfx's European Product Manager, Luciano Alibrandi, the interviewer asked if 3dfx was working on Voxel technology. Mr. Alibrandi replied "No"--but several days later the interview was updated at 3dfx's request, with the "No" struck out and replaced with a "Can't Comment".
Anyway, we may find out if any of the rumors are true at the Game Developer's Conference that is taking place March 8-12.
Is this even possible or feasible?
What with cards with 64mb+ of memory, 'GPU', etc.
IE, framebuffer and display is no problem. Data would be loaded through a simple, stupid, microprocessor across the AGP bus; all you'd need. I'm sure there are Linux distros that could fit themselves comfortably within 64mb ^^
Anyone?
-AS
-AS
*Pikachu*
I agree with the curved surface rendering, and collision detection, but not the rest.
Depth of field is not appropriate for interactive games. In RL you refocus your eyes to look at different things, if you can't refocus just by controlling your eyes, you'd be half blind. It'd drive people crazy.
Integrated physics would lock the programmer into a certain physics model. Physics is not terribly CPU intense, and the demands vary a lot from game to game. Having specialized physics hardware on the video card is about as appropriate as having specialized AI hardware (IOW, it's not).
Voxels are either huge memory pigs or butt ugly. They might make nice 3d texture maps (if you're okay with fuzzy interpolation), but I wouldn't want to bother with them for whole 3d models.
Chromatics are a waste. They are so rarely useful that it would be better to special case the lighting effects when needed.
Radiosity would be nice, but it's not something you can just pipeline in (ditto for casting rays). However, there might be cheaper ways to get the same effect.
-----------
"You can't shake the Devil's hand and say you're only kidding."
I remember reading recently (on MacOSRumors; make of that what you will) that Apple was so pissed off at ATi for delays in the mobile Rage128 in the new PowerBooks that they're taking another look at companies like nVidia for OEM support.
IIRC Apple did rewrite parts of the RagePro driver library for the Mac, although I don't know if they're working on Linux versions as well. I'm guessing they're working very hard on BSD versions, though.
The G4 and even the newer iMacs make a quite decent Q3A platform thanks to the Rage128, but they still lag behind the latest PC video cards. Hopefully Apple will persuade developers to write the appropriate drivers for OS X so you can stick any AGP card in a G4 and have it work out of the box.
I use Macs for work, Linux for education, and Windows for cardplaying.
There must be others
.plans where he said that ID had bought a supercomputer (a Sequent?) to perform the Quake radiosity calculations.
How about Phong instead of Gourard shading? Fast Phong algorithms for hardware implementation have been about for years. They're still more computationally intensive than Gourard but remove the need for specular texture maps and reduce mach-banding artifacts.
Real-time radiosity? Not for a very long time, methinks. Radiosity is usually pre-computed. I remember reading one of John Carmack's
HH
Yellow tigers crouched in jungles in her dark eyes.
Yellow tigers crouched in jungles in her dark eyes.
She's just dressing, goodbye windows, tired starlings.
generate the appropriate texture volume and then put that in the image cache with the standard 2D versions
:-)
As I mentioned elsewhere in this discussion, precomputed 3D texture maps would take up vast amounts of memory on your video card. IMHO it would probably be better to let the CPU compute the procedural textures and transfer them to the card using AGP.
Better still, provide a texture compiler that produces bytecode that can be executed directly by the card. Now that would be cool. Procedural displacement mapping (like RenderMan uses) would be ultra-cool.
So the first 3D card that can execute RenderMan shader bytecode will get my money
HH
Yellow tigers crouched in jungles in her dark eyes.
Yellow tigers crouched in jungles in her dark eyes.
She's just dressing, goodbye windows, tired starlings.
Nice to see support for 3D textures. These are very cool. The article says:
Traditionally, polygons are used to represent 3D objects. However, with 3D textures, volumes of texels (textured pixels) may also be used. In a 2D texture map (the kind that we see "glued" to a wall for instance) indexing occurs via two texture coordinates, whereas in a 3D texture, there are three coordinates.
One good example of 3D texture use would be that of a marble cube. If the corner of the cube were to be chipped off, any veins running through the marble would already be defined and visible without any additional textures being generated.
This means that you could chop a block of wood up, and have the wood grain on the cut surfaces rendered correcly. However, the article then goes on to say:
Unfortunately, we feel 3D textures will have to be used incredibly sparingly because in order to implement the marble cube example explained above, an artist would have to draw the entire 3D surface (including the veins inside the cube which may never be seen!).
This is incorrect. How can an artist draw the inside of a solid cube of marble or wood? I've never heard of a 3D texture being created in this way. They are normally generated procedurally, where you have a function that mathematically calculates the texture colour given the x,y,z coordinates within the texture. This does mean that you can't store 3D textures on the card, unless you pre-calculate an array of texels, but this would require vast amounts of texture memory on the card.
HH
Yellow tigers crouched in jungles in her dark eyes.
Yellow tigers crouched in jungles in her dark eyes.
She's just dressing, goodbye windows, tired starlings.
Is this hardware, or an election? Honestly, this is just ridiculous. Somebody tell ATI that hardware can't have charisma. Somebody tell Sony that hardware can't have emotions.
Yeah, that's me, trendspotter extraordinaire. Takes a genius these days, eh?
Wile all your suggestions are good, i think a rel photo-realistic engine will be quite different. it will basically be POVRay on a chip or perhaps renderman since I think the most appropriate thing would be to forget bump-mapping and go with a shading language that actually deforms the geometry of the object and then does true ray-tracing.
What comes after that? Well, I have to fight hard to keep my eretion down when thinking of this... Hardware-based realtime radiosity. *uNF*
BTW, the idea of hardware-based collision detection... I hope to GAWD that the hardware manufacturers out there, nVidia in particular (cause I haven't read what they're doing after the geForce256, everyone else has something in the public works) read your post. It would be most gorgeous and possible to have such a thing...
Esperandi
This was queitly done recently, but in the most recent geForce256 drivers, Fullscreen AA can be enabled with a little registry twiddling. From what I read there is a "performance hit" but I'm not sure how significant it is...
Esperandi
Another problem I had with this article along the same lines is his view (as is POVRays current view, but this is fixed in Megapov and will hopefully make it into the next release) that bump-mapping is the same as surface-deforming textures. Sure, it might look the same on a screenshot of a bumpy pear sitting on some blistering wood (the screenshot on sharkyextreme), but when you get into a game, you gotta keep the bumps really small. Imagine you've got the wall of a dungeon. You think it'd be easier to make a plasma-fractal generated bump map instead of breaking the triangles on every single section of the wall (which I'm planning on doing in an upcoming game ;)... well, when your users get into the game and start walking down this really kickass-looking craggy hallway, they're going to notice something. The crags aren't there. They can walk right through them. Which would be laughed at and viewed as very annoying by gameplayers. (it obscures your view and its ethereal?!) If REAL geometry-deforming maps were able to be applied, collision detection could work, it would keep the polygon count down and the game would be wickedly fast... but POVRay help off for such a long time, I fear if every card supports this fake bump-mapping (it takes the light and just moves the point and texture maps it a little different based on the new virtual position) that game developers will get happy with it and never realize how versatile real surface deforming textures can be... Combine it with a separate collision detection engine and we could have some amazingly realistic games...
Esperandi
This card that ATI is making is wondrous. I mean, stunning. However, until a few days ago I owned an ATI All In Wonder Pro. I didn't get it for its 3D because I knew it sucked, i got it for the video capture capabilities which it had in abundance. Only a dedicated Videum card I used to have performed better. I soon found out that ATI has absolutely horrid drivers. I mean, this card never worked right even when I got super bleeding edge beta drivers (they were beter than the released ones and more stable I don't know why they weren't public, they gave me a secret URL, forbade me from dispensing it, etc). I would capture a video using their proprietry VCR2 codec (more efficient codec than I had ever seen or have ever yet seen!) and then go to edit it. Playing the video would work, but seeking in any sense would crash the drivers. To this day (and the all in wonder pro is 2 or 3 years old) they have not fixed the problem.
I honestly hope that they support this card well and do a good job with the drivers because even if you've got the best card, its only as good as its drivers.
Esperandi
I hope they'll use this chipset to target the hardcore gamers and start a good battle against nVidia's supremacy. Like the AMD vs. Intel thing, us consumers will only benefit :)
(Another cool article on the charisma is here.)
The one major thing the author misses about 3D texture maps is that rarely are they hand drawn by an illustrator. A typical map is a procedural texture (think of rendering a marble texture using POVRay) so generating a lump of marble is not that difficult a thing to do.
For games, the programmer just needs to fire up Povray, 3DS etc and get it to generate the appropriate texture volume and then put that in the image cache with the standard 2D versions. I'm sure a lot of game engines will handle this pretty quickly.
Life is complete only for brief intervals in between toys or projects -- John Dalton
OTOH,
:) (this was before the destruction of earth by the vogons).
I feel that those two names come real close to describing these two very excellent egines as best describle on earth
Lets see..
On Charisma, David Jenson wrote
When scientists and technicians hear the word charisma, they may first think of sales reps or politicians. But you'd be hard pressed to
find a person in any influential biotechnology position who doesn't have some measure of charisma. Those on the scientific track are
not exempt from this need.
Charisma is derived from the ancient Greek word kharis, meaning "to cause to strive or desire." The ability to motivate others to strive
and succeed is a major building block of successful management, whether in a QC lab or in a corner office. Charismatic people
describe goals by painting word pictures, thereby motivating others to a particular end. They have an exceptional ability to win the
devotion and support of others. They have no fear of presenting their ideas to anyone who may be able to help them. And they have
excellent persuasion and negotiation skills.
But more to the effect, I see charisma here derived from the Indian (as in South Asia) word. It too has a similar meaning to the regular charasmatic word. In this way, the word comes closer to a powerful healing force that is being ispearsed around the subject than anything else. This is a very visual word. A very charasmatic word. The word conjures a halo around it's subject and renders it in a light that leaves a very strong impression on anyone who hears it uttered. Thus, it is fitting a name for this chip (Which I believe would live up to this name). As would, the emotion engine in PlayStation 2. Which also conjures strong vibes and powerfully drawn meanings to the word and what the chip can do. Human emotions are powerful, machines, from the start of time (execpt for Marvin) are known to lack them. The very thought of a machine having these very emotions drive a very strong stab at anyone who looks at the PS2. The engine was made for it's artistic quality, it's ability to render something beautiful, so beautiful that it is almost real, that is the emotion, the charisma.
--
HotHardware has another article on the R6 "Charisma", as well as a copy of ATI's White Paper.
Everybody likes to compare nVidia and 3dfx as the two top companies, but in reality 3dfx is a small fraction of nVidia's size. I don't have exact numbers offhand, but nVidia currently has about 45-50% of the graphics market while 3dfx has something like 10-15%, and I believe Matrox is even smaller than 3dfx.
Here's a comparision of some market caps (data from The Motley Fool).
ATI: 4,141.51 million
S3: 1,607.80 million
nVidia: 1,808.46 million
TDFX: 218.09 million
Any game that uses OpenGL for more than a rasterization-only API automatically uses geometry acceleration on cards/drivers that support it.
Realistically, the boost is less than you may think. An average game doesn't spend more than 15% of a frame doing transformation. So the Ultra-Fast-Geometry-Accelerator-of-the-Future is going to buy you a 15% speedup in that case. The other issue is that geometry acceleration is only useful when you pump the data straight to the card and don't want or need intermediate results. For example, you'll have to transform points (one way or another) to do collision detection against instanced objects. But you can't use the geometry acceleration in that case, because the CPU needs the results.
Geometry acceleration is good, but it's not the panacea that many people are expecting it to be.
That card's core is a highly specialized chip, focusing on the math for the transformation of points, lights, textures, blurs, and several things that don't involve most of the core of an x86 chip. Linux will never run on it unless they make one with a big flash rom, and a general purpose controller chip (read: i960 or other).
What kind of use is the card? LOTS! Go check out the Intergraph Intense3D Wildcat 4110. It runs in most prebuilt p2/p3 graphics workstations (huge card), and takes so much of the processing off of it, the only thing the cpus are needed for is getting the software started, and other extended math calculations (we love those fcurves!), and rendering of the final image. By the way, this card does everything the geforce AND v5 do... but i'm not sure as to it's fillrate, but 21fps in a scene with 80000 polys is impressive. And game cards are catching up quite quickly to the power of the "industrial card".
Acelleration... What needs to be done is the accelleration of the front side bus. It's just poking along at roughly 133mhz now, maybe 200 on athlon systems (but that's only ram to cpu). It'd be better if that were 1/1 with the cpu speed, or if it were faster, leaving the cpu with a wide open pipe to the ram and the other system peripherals.
The 3d card market seems to have an alternative version of Moore's law:
Every 18 months, the number of people making 3D graphics cards halves. There's only about 6 companies making 3D chips now.
Speaking as a game programmer, these advances are coming so fast that there's no time to concentrate on (1) pushing the limits of what a current generation of cards can do, and (2) dealing with card-specific features.
On the first point, there's not enough time to sit down and focus on where all the rendering time is going in a complex game. Well, more like there are so many card and driver combos out there that the best we can do is try to write generic code and have it work across the board. If we could focus on one card, say a Voodoo 2, then we could push the limits of that chipset out beyond what people only expect from a GeForce. But there's no time for that, so we plow ahead using about 50% of each card's capabilities for the three month window until the next card comes along.
On the second point, 3dfx, Nvidia, Matrox and ATI (and S3, and...) are all branching out into odd and card specific feature sets. 3dfx has their T-buffer. Nvidia has "8 lights per triangle hardware lighting." Matrox has a certain kind of hardware bump-mapping. ATI has all sorts of wacky stuff. The bottom line becomes "Do we want to just focus on writing a great game, or do we want to spend an extra six months of development so we can support special features of all these cards that were considered hot eight months ago when we were still pre-beta?" And tacking in special Matrox-only support, for example, is hell on QA. It makes a lot of sense to ignore such features, unless we're getting a bundle from the card company to cover us for the trouble.
It seems that while the push for ever increasing image quality is going on, we are getting much closer to realistic, real time rendering of scenes. I wonder just what else is needed to really be able to push the envelope of visualization and realism further. Here's my current wish list.
There must be others - it looks like ATI is going to finally give us proper bump mapping and range-based fogging. Do we also need a proper chromatic model so that we can get rainbows through glass objects? Should there be real-time ray-casting or radiosity support so that real lighting effects (say carrying a flaming torch down a corridor and having proper soft-edged shadows) can be achieved?
Cheers,
Toby Haynes
Anything I post is strictly my own thoughts and doesn't necessarily have anything to do with the opinions of IBM.