Slashdot Mirror
New Graphics Firm Promises Real-Time Ray Tracing
arcticstoat writes "A new graphics company called Caustic Graphics reckons it's uncovered the secret of real-time ray tracing with a chip that 'enables your CPU/GPU to shade with rasterization-like efficiency.' The new chip basically off-loads ray tracing calculations and then sends the data to your GPU and CPU, enabling your PC to shade a ray-traced scene much more quickly. Caustic's management team isn't afraid to rubbish the efforts of other graphics companies when it comes to ray tracing. 'Some technology vendors claim to have solved the accelerated ray tracing problem by using traditional algorithms along with GPU hardware,' says Caustic. However, the company adds that 'if you've ever seen them demo their solutions you'll notice that while results may be fast — the image quality is underwhelming, far below the quality that ray tracing is known for.' According to Caustic, this is because the advanced shading and lighting effects usually seen in ray-traced scenes, such as caustics and refraction, can't be accelerated on a standard GPU because it can't process incoherent rays in hardware. Conversely, Caustic claims that the CausticOne 'thrives in incoherent ray tracing situations: encouraging the use of multiple secondary rays per pixel.' The company is also introducing its own API, called CausticGL, which is based on OpenGL/GLSL, which will feature Caustic's unique ray tracing extensions."
Being a 3D artist (mostly just a modeler and texture artist, but sometimes a generalist), I'm happy to see work like this being done. It seems like only yesterday I was waiting hours or all night for simple ray traced scenes.
While it may be underwhelming to some, I'm more than happy to see people working on this kind of tech. Sure, we've moved on from just "simple" ray tracing to using things like GI, etc, but in time we'll have that in real time as well. Some apps are already doing some tricks to enable real time GI and other tricks. (the key word being tricks, since they're not totally physically accurate). Obviously real time will always lag behind, but I look forward to the future.
WWJD.... for a Klondike bar?
2009 is the year of the ray traced desktop.
You can't take the sky from me.
Stop copying and pasting the article to generate almost the entire summary, especially when you don't do it right. The However, the company adds that 'if you've ever seen them demo their solutions you'll notice that while results may be fast -- the image quality is underwhelming, far below the quality that ray tracing is known for.' makes it look like you're talking about the Image quality of Caustic's new solution, which is obviously wrong. Here's the real paragraph:
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Do they get their chips from Flammable Systems, and their capacitors from Toxic Components Inc?
Table-ized A.I.
They've advertised Linux support too, but I haven't heard anything from these guys. Unless they're like nVidia and sit around killing kittens all day, it would be a good idea for them to actually do some research and figure out how GLX and DRI work. Even the ATI closed-source drivers still respect the GLX way of life.
(nVidia replaces the entire DRI stack. DDX, GLX, DRI, DRM, all custom. fglrx doesn't replace GLX. Just in case you were wondering.)
~ C.
Caustics are light reflected and/or refracted by curved surfaces. The pattern of light lines on the bottom of a pool is one of the more common types of caustic. The company chose a graphics term. The graphics people chose a term that has another, more understood meaning.
Not a sentence!
Juggler was very impressive for the time, but it was "only" real time high-color-depth animation playback (although even the compression method used was probably impressive back then). It was not real-time raytracing. Yes, Amigas were famously one of the first computers that made raytracing possible for home (or even pro movie/TV) users back then, but I remember that rendering a simple raytraced scene (a couple of primitives) in apps like Imagine 3D would have to run for a few hours, if not overnight. That might have been on an Amiga 1200, rather than my older 500, too.
Yeah. _Not_ in real time. I admit the article is confusing, but that Amiga anim was not done in real time.
The rendered images were encoded in the Amiga's HAM display mode and then assembled into a single data file using a lossless delta compression scheme similar to the method that would later be adopted as the standard in the Amiga's ANIM file format.
You must have misread the article... it reads "20x", not "20%".
I.e. a 1900% increase. Or however one would put that. 20 times faster.. much easier. Still within the margin of error? :)
( also per the article, they're actually pondering 200x faster down the line. )
Like with anything, I call vaporware until they show real silicon. Not because I think they are lying, most companies don't. However there are plenty of overly ambitious companies out there. They think they have figured out some amazing way to leap ahead and get funding to start work... only to realize it's way harder than they believed.
A great example was the Elbrus E2K chip. Dunno if you remember that, it was back in 2000. A Russian group said they were going to make the Next Big Thing(tm) in processors. It'd kick the crap out of Intel. Well obviously this didn't come to pass. The reason wasn't that they were scammers, in fact Elbrus is a line of supercomputers made in Russia. The problem was they didn't know what they were doing with regards to this chip.
Their idea was more or less to put their Elbrus 3 supercomputer on to a chip... Ok fine but the things that you can do on that scale, don't always work on on the microscale. There are all sorts of new considerations. So while their thing was all nice in theory on a simulator, it was impossible to fab.
Intel and AMD aren't amazing because of the chips they design, they are amazing because they can then actually fab those chips economically. You can design something that'll smoke a Core i7 in simulations. However you probably can't make it a real chip.
This smells of the same sort of thing to me. Notice that they have press releases and some shiny demo pictures, but it was clearly done on a software simulator. Ok well shit, I can raytrace pretty pictures. That doesn't prove anything. Their card? Apparently not real yet, the picture of it is, well, just a raytrace.
So who knows? Maybe they really do have some amazing shit in the pipeline. Doesn't matter though, they've gotta make it real before it matters. nVidia releases pretty pictures too. Difference is the pictures of the cards are of actual cards, and the pictures rendered are done on the actual hardware.
I am just never impressed by sites heavy on the press releases and marketing, and light on the technical details, SDKs, engineering hardware pics, and so on.
I shall remain skeptical until more information is forthcoming.
I'm fully aware of that. Notice I didn't say we've moved on from ray tracing to GI. I said we've moved on from "SIMPLE" ray tracing - the operative word being "simple". Perhaps I should have been more clear and said "we've moved on from just basic raytracing to more advanced and accurate methods of ray tracing", but I figured my point was clear enough.
WWJD.... for a Klondike bar?
http://www.youtube.com/watch?v=B3qtq27J_rQ
( no, not a realdoll advert - it's a vid of their current test card being twirled around in a human's hands. then again, maybe they raytraced that )
"Global Illumination"
It's a bit of a not-so-well-defined term, really, but within the context of current generation renderers, global illumination involves calculating not just direct lighting (i.e. a spot lighting a wall), but also diffuse indirect lighting (the light hitting the wall (dimly) illuminating the rest of the room) and even specular indirect lighting (such as caustics - like the light patterns you see in pools).
I know! This is totally going to solve the problem of the utter lack of glass spheres and infinite checkerboards in today's games!
At some point (not too far away), the average size of a polygon in a scene will drop to one pixel or smaller. It seems like the different rendering techniques will merge together... a bit like the strong, weak, and electromagnetic forces merged.
For something as ambitious as they have, it's very strange that their web site has no demos, absolutely nothing, of their products. No pictures, no videos, nothing.
That's how just about all ray-tracers work. The problem is when you want to avoid aliasing effects. The easiest solution is to use multi-sampling, but having a nice square grid of primary rays per pixel still creates some aliasing effect. Randomizing the directions of these rays using a statistical distribution is one way of improving things. But then, at every reflection and refraction the secondary rays converge and diverge even further, so they will not all hit the same triangle/object/texture which causes all sorts of texture caching problems.
This company seems to have found a solution with their "incoherent ray" solution.
Vintage computer adverts: http://www.vintageadbrowser.com/computers-and-software-ads
Nice try, though.
No existe.
Thanks for the explanation. However, names that require explanation are not good choices for the names of companies.
I remember when I first saw a very poorly drawn, shaky image of an animal and read that it was a Gnu, and read how clever the name was considered to be since it was, they said, "recursive": GNU is Not Unix.
It didn't bother the enthusiasts that most people in the world can't pronounce the name and have never seen a Gnu.
They found someone with artistic ability to make a better image of a GNU, but I've seen no evidence that anyone with technical knowledge realizes the depth of the self-defeat in choosing an obscure reference to an obscure animal.
To most people the word "caustic" means only "capable of burning, corroding, or dissolving".
To raytrace a soft shadow you have to send out at least 16 rays per shadow calculation, for each light and even then your gonna suffer from nasty artefacts. Compared to the raster solution which involves rendering the zbuffer of any given light source and merely doing some blurring. same quality, much reduced cost.
It seems to me that the algorithmic complexity grows just as fast for both rendering techniques in the case of many lightsources. Both are accomplished in steps linear to the number of lights.
Its all well and good that rasterization is "fast" for what we use it for today. But, its growth is linear to the number of primitives while there are other methods that are sublinear. For a large enough number of primitives the sublinear algorithm must be superior in performance.
"His name was James Damore."
He obviously likes kittens.
The revolution will not be televised... but it will have a page on Wikipedia
Didn't OpenGL always support real-time raytracing if you throw it enough hardware? Unfortunately "enough hardware" to render complex scenes in real time has not existed yet.
Can you magine what kind of 3D modeling rig god has? Somehow I don't think it's based on an ATI or nVidia chipset. ;)
The Christian Right is Neither (Christian nor right). See: Matthew 23, Matthew 25, Ezekiel 16:48-50
Even my MSX computer did real time raytracing like a champ, providing that all the pixels were produced from a 2D non-reflective surface using a 90 degree angle. Of course you had a limited color space, but otherwise everything run just smoothly.
Kidding aside, I suppose it's how far you want to take it. Amiga or MSX are not interesting anymore for about 90% of the things they did. The one exception is probably playing retro games.