7 Years of 3D Graphics

xtra writes "At Accelenation they are running a nice timeline about 7 years of pc 3d graphics contains much info and even talks about some of the not so well known players anyone still remember rendition? or BitBoys?" How many cards on their timeline chart have you used?

Comments

Infinity and beyond...

[ackthpt]

At the going rate, the board with CPU and chipset will be a daughterboard of the graphics motherboard. :]

How fast do we really need to go?

[TrollMan 5000]

Can graphics technology possibly get any faster? Well the GeForce2 GTS chip ran Quake3 at 80fps in May of 2000. Just twenty-two months later a GeForce4 Ti4600 can run Quake3 over three times faster. On that reckoning the GeForce6 in two years time should be running Quake3 at over 700fps. Is that fast enough for you!

Is there really much visual difference between 700 fps and 135 fps? I'm not really sure if the human eye can make the distinction. They're sure pretty-looking numbers, but do the results show for it?

And how long before video cards can render essentially photo-realistic graphics? Soon games will be more like interactive movies.

Re:How fast do we really need to go?

[Hieronymus Howard]

Ray tracing isn't the only way to achieve high-quality graphics. Renderman (used for Toy Story, Monsters Inc. and plenty others) isn't a ray tracer, but a sub-pixel renderer (if I remember my computer graphics M.Sc. correctly).

We've still got a very long way to go until we get Monsters Inc. quality real-time games. As you say, current cards render triangles. Curved surface rendering (e.g. NURBS) may come next. Anti-aliasing takes a lot of power. I think that current cards are still using Gourard shading, which is the most primitive shading model there is (correct me if I'm wrong here). The next step is Phong shading for highlight effects (there are hardware-optimised Phong shading algorithms, but they're still slower than Gourard). Then there's deformation mapping (Renderman again), etc. etc.

I believe that Quake 3 etc. does use radiosity algorithms, but that doesn't need to be done in real time, just when the level is compiled.

HH

Re:How fast do we really need to go?

[bribecka]

Renderman (used for Toy Story, Monsters Inc. and plenty others) isn't a ray tracer, but a sub-pixel renderer (if I remember my computer graphics M.Sc. correctly).

You're referring to PhotoRealistic Renderman (PRMan), the actual product developed by Pixar. It uses the REYES algorithm.

RenderMan is a specification for defining 3D scenes, much as PostScript is a specification for defining 2D documents/images. There are many renderers that are RenderMan compatible, including raytracers such as BMRT.

Only 7 Years?

[green pizza]

Heck, I still remember the "which is better, Silicon Graphics Reality Engine or Ferrari Testarossa?" threads in the USENET from the summer of 1992. Even the dual pipe / dual head SGI VGXT "Skywriter" from 1989 was pretty damned impressive. Even many, many years later.

Re:Number Nine

[Jeffrey Baker]

Uh? I don't remember any #9 cards with 128MB of memory. In fact, #9 was out of business two years ago. You may remember the #9 i128 series of cards, but those are very old and do not have 128MB memory.

You only need 16MB to handle th highest resolution computer graphics displays ever made.

Re:Number Nine

[Milalwi]

You only need 16MB to handle the highest resolution computer graphics displays ever made.

This is true for 2D displays, but when you start having double and triple buffering plus z-buffers it starts to add up. Then add the texture requirement and you can see why most newer cards have 64-128MB of memory on the cards.

Milalwi

More memory

[Traa]

> You only need 16MB to handle the highest
> resolution computer graphics displays ever made

you will allways need more memory (in 3D graphics accelerators), even if the display resolutions don't increase. Lets say we settle for a nice 2000x2000 ish display. Thats 4M pixels, at 32-bit is 16MB for the display.
At least double (32MB) but preferably triple (48MB) buffer this so you can create a new frame while the old one is being displayed. Then we need a Z-buffer (or W-buffer) to hold the depth values (24bit values) for each pixel, so we know what is in front of what, typically you might want to do some stencil effects to (8-bits, can be packed with the Z-buffer) that would be another 16MB. Now we have the basics for a 3D graphics display and are at 48-64MB.

But we are not done yet, now for some more interesting effects:
- Texture memory. Typically use the leftover graphics memory and swap the rest from host memory (but we don't like swapping, so preferably all textures should be in onboard mem) 2-64MB
- 2x Antialiasing (1 Backbuffer + 1 Z-buffer 2*2*size of display buffer) = 64MB (4x antialiasing = 256MB)
- Shadowbuffer (rendering into a kind of Z-buffer from the lightsource to create realistic shadows) 16MB
- Accumulation buffer effects like motion blur (very expensive, a good blur could take 4 to 32 frames) or depth of view could make us want another 4-32*16=64-512MB

I for one could easily use more then 1GB of onboard graphics memory.