Is PS3 able to preform ray tracing via cell

Why not? It's all up to someone to say "h3llz je4h!" to the idea.

Yeah, but the problem with that is that you have to keep buying packets of fairy dust in order to sustain the performance. They still haven't figured out how to link into the whole "think happy thoughts" thing yet.

 

Also, let's not forget that there's an RSX in there too. Going for a software renderer would mean that all those 300 million transistors would sit idle there...

 

I would think combine the RSX with Cell. RSX transforms and textures geometry, Cell lights it with raytracing. And on a puzzler with simple graphics (Tetris) those 300 million trannys would be going to waste anyway

 

Getting back the Z-buffer contents from RSX would be the most valuable piece as it would accelerate that 'first hit' raycast. Texturing would actually be more perspective correct using the raytracer to get texture coords because you are effectively doing perspective projection at the pixel level when casting those rays. However, the GPU gives you all nature of anisotropic filtering, texture compression, so if you're going to grab texture colors from the GPU output, might as well also throw out normal maps and displacement maps. Actually, with a raytracer, you wouldn't have to fake per-pixel displacement with a parallax effect -- you could go ahead and do the real thing.

 

How's about, create models in Cell, z-buffer from RSX, ray-cast for texture coords and lighting per pixel, use RSX to texture, beck to Cell for post-processing, and out to front buffer.

 

You'd need some serious AA in that case though, either by casting more rays per pixel (4-9), by applying some very high quality filtering to the displacement texture - or both...

 

any CPU can do raytracing - what matters is, how FAST can you do it in real-time, for games?

unfortunately, even with the large leap in peak performance that Cell offers, it will not be nearly enough to do true raytracing with complex scenes at high framerates.

the dream of realtime raytracing in games is still a ways off.

I believe we need two major technological advances so that we can have raytracing in games.

1.) hardware that is geared around doing raytracing from the ground up. not merely an increase in flops. there a couple of examples of small, low powered chips that have been designed to do realtime raytracing. this is a step in the right direction. eventually, these chips will become more refined and perfected. once that happens, then we can only hope that the process / processor can be scaled up to current high-end GPU / CPU levels of transistors and core frequency, so that we have GPUs or new types of PROCESSORS that are capable of running raytracing in highly complex scenes.

2.) I believe we need some new algorithms that dramaticly reduce the amount of raw processing power that is normally needed to pull off realtime raytracing. be that in the form of hacks, tricks, cheats, etc, so that even if the quality of the raytracing is not 100% exactly as good as offline prerendered raytracing seen in CGI, it is "good enough" and most importantly, FAST enough, for games.

these are simply my own thoughts about what is needed before we see raytracing in games. I have no background in graphics or technology in general. just my observations

 

I wonder what a game using this sort of method would look like?

 

By the way, why would it be better to replace scanline rendering with raytracing? So many people seems to be so interested in it, but I just don't see why?
All the really nice stuff like GI, area shadows, subsurface scattering etc. require bounced rays, multiple samples and so on which 1. take more time to calculate 2. mess up memory access. But these effects are usually performed in separate passes in offline rendering, and most of the geometry is processed with scanline algorithms - thus most renderers today are hybrid renderers. Exclusive raytracing is slow and old stuff...

 

You either trace that ray or not... Optimization is possible in the data management part, to speed up hit detection by using acceleration structures like octrees, voxels, BSP etc.

You can also play with the number of samples per pixel - both oversample or undersample - but that'll produce artifacting. Flickering shadows and noisy shading are inacceptable because they're simply disturbing. Like, no bi/trilinear filtering or MIP mapping on high contrast textures - totally ruins an otherwise highly realistic (models, textures) scene.

 

I would expect that we will see more & more advanced scanline / scan conversion renderers that encorporate many of the things we want out of realtime RT, GI, etc. but it will take some time. on the PC side, 4-5 years from either ATI or Nvidia or someone else that is brave.

on the console side, probably 5-7 years, with the arrival of Xenon's successor and PlayStation4.

mind, you PC graphics in 5 years and console graphics in 5-7 years will still not be capable of everything we want in realtime graphics, it just might be the first step towards getting away from only having polygons, textures and pixel shaders. real-time 3D graphics have not changed much, funamentallym, from the 80s and 90s (polygon + texture) We have seen the addition of realtime shaders in this decade, that is about it, other than ever-increasing speed/performance. This era were shaders is the only new thing will hopefully end by 2010-2012, if not sooner, and then we see a new era of 3D graphics with some of the currently impossible-to-do-in-realtime stuff, suddenly becoming possible.

even PS4 will not be able to do all that we want.... but since the conventional brute processing requirement for things like full-on true realtime raytracing in uber-complex games will NEVER be met within our lifetimes (going from EE to Cell, 35x at best, is a good lesson here), the industry needs to find unconventional ways of arriving at the type of graphics we want, without the need for sheer brute horsepower.

I feel optimistic about the 3D graphics industry, and that, given time, it will come up with some very interesting and impressive solutions that will at least bring some of the things that we all want, within a reasonable amount of time, 6 to 12 years, which seems like an eternity, but is better than waiting for brute force processing power to make it possible.

<----- me just rambling

 

I 'll be little OT but ,there is no necessity to create a thread for this:
It's sometime interesting to browse the job oppotunities on the developpers web page .
Naughty dog 's jobs page has some interesting hints.

http://www.naughtydog.com/corporate/index.html
clic on job opps.

egg:
these guys really love being close to the metal :