Is PS3 able to preform ray tracing via cell

[Squeak]

Personally I would be a lot more exited about some sort of REYSlike rendering, at least for part of the scene.
If the rumours of a microtessellation unit within RSX turns out to be true that could very well happen.
As opposed to even the most efficient, fully fletched raytracer, the REYS approach was conceived from the beginning as a way of getting photorealistic (or near) rendering at the highest speed possible.

 

[blakjedi]

http://www.beyond3d.com/forum/viewtopic.php?p=532441&highlight=ray*#532441

to quote myself:

markvaid wrote:
are you guys sure the terrain demo was running of two cells not one. Is there any confirmation on that?


Heres some background on the most impressive one.. the terrain building demo

Taken from http://gear.ign.com/articles/615/615521p1.html?fromint=1

"The key to this demo was that the imagery was ray-traced in realtime solely via a _ pair of Cell processors_. Instead of using a video card to display the graphics (thereby limiting the "wow" factor), the Cell processors rendered the image to a series of JPG images, which were then streamed to a standard G5 tower over the network. Each image was rendered at 1280x720, creating a 720p high-definition scene of the area."

A "pair of cells" that render to jpgs then streamed together... hmmmmmm...

 

[AlexV]

A little besides the point, but doesn`t SM3.0 offer the possibility to at least send rays through the all of the pixels of a rendertarget, using vPos?Which may be a first step...

 

[Shifty Geezer]

Going with the opinion all the texturing was procedural (no textures, like the clouds were procedural as was identified) and the geometry was being produced in realtime from two sets of source data, and the Cells were presumably of the lower clocked variety, I wonder what options there are for RT of a sort?

If RSX can generate models and texture them, fill a 3D world, and use Cell to ray trace the scene, certainly a degree of complexity could be managed in realtime. Not in games, save maybe some reflections and refractions in 3rd+ gen games? And maybe some puzzlers could go way out with realtime raytracing if the graphics are simple enough?

Wishful thinking! But I do think there may be possibilities there for RT capabilities, maybe more to be seen in Linux programs than games.

 

[Titanio]

A "pair of cells" that render to jpgs then streamed together... hmmmmmm...

I know what you're thinking, but apparently most of the time they've demoed this they've had a joystick with which it can be interacted with in realtime.

 

[ralexand]

A "pair of cells" that render to jpgs then streamed together... hmmmmmm...

I know what you're thinking, but apparently most of the time they've demoed this they've had a joystick with which it can be interacted with in realtime.

Has a frames per second been mentioned for this demo?

 

[Titanio]

A "pair of cells" that render to jpgs then streamed together... hmmmmmm...

I know what you're thinking, but apparently most of the time they've demoed this they've had a joystick with which it can be interacted with in realtime.

Has a frames per second been mentioned for this demo?

No specific figure, but it was smooth to my eye, and the IGN article mentions a mode that lets you see the SPEs load balancing and so forth to maintain the framerate.

Anyway, I don't think it's clear what rendering techniques this demo was using, and in my mind I don't think it was even meant directly as a demo of purely its rendering capability.

 

[Farid]

Yes, but a caveat, it was a 2 Cell IBM blade server prototype. 400GFLOPS.

Also the ray tracing demo was running on two CELLs.

Note that the Cells were runnings at ~2.4GHz.

 

[cobragt]

that would be 4.8ghz of processing power right? 2 cells.

 

[AlgebraicRing]

that would be 4.8ghz of processing power right? 2 cells.

No, that would be 2 2.4 Ghz processors. They do not add linearly. I actually think the parallelism may afford greater processing over time. More pipe width implies more volume for flow. A hang in one processor is not a hang in the other processor, so it can keep chugging. Furthermore, the speeds of the rest of the components may have an effect, things like bus speed, mem speed, IO speed, etc. If they are the same in both situations, the two processors have an advantage. Who can drink more water, 2 people who can swallow at speed X from 2 hoses with speed Y, or one person who can swallow at speed 2X from a hose with speed Y? If Y is the bottleneck, then the two hoses become more powerful.

 

[PC-Engine]

I noticed slight stuttering of the terrain demo.

 

[_phil_]

about 60 % of what can do raytracing is raycasting.( the 40% rest is about special recursivity properties ? ).The voxel hint was just to point out that it's basically simple 3d pixels ,too.

As for Gpu not good at raytracing ,the last trend of Shader programming ,is all about implementing a raytracer at the pixelshader level.

http://fabio.policarpo.nom.br/relief/index.htm

Too bad ,the curved relief mapping demo seems to have been removed from the link.It had object silhouete correction.

http://graphics.cs.brown.edu/games/SteepParallax/mcguire-steepparallax-poster.pdf[/quote]

 

[one]

Reminded me of these:

http://www.beyond3d.com/forum/viewtopic.php?t=18849
http://www.beyond3d.com/forum/viewtopic.php?p=481714#481714
http://www.beyond3d.com/forum/viewtopic.php?t=20268

Now, if CELL + RSX can work as one big GPU...
Also this comment from David Kirk is interesting though not directly related.

 

[PiNkY]

Raycasting works by casting a (single) ray from your Center of Projection through each Pixel Center of your Projection Plane to your model to sample its color. It's quite fast and often used in Software rendering/collision detection. AFAIK Wolfenstein/Doom/Hexen etc. used a variation where only a single scanline would be raycasted per frame.

 

[_phil_]

The difference between raycasting and raytracing is that in raycasting the ray stops at "first hit ".Raytracing follows the ray ad infinitum if wanted.

 

[Intel17]

Will people be able to buy Cell based computers in the near future, or is the PS3 the only way to get your hands on them?

 

[Carl B]

Will people be able to buy Cell based computers in the near future, or is the PS3 the only way to get your hands on them?

I think if Kutaragi and crew follow-through this time all the way, they're thinking that PS3+hard drive WILL be a Cell based computer.

 

[Shifty Geezer]

Depends. By all accounts PS3 + HD = Cell computer. If software appears for general applications then Cell PCs might well appear, though PS3 will represent the best value for money. Does PS3 support monitors directly or is an adapter required?

 

[Farid]

Will people be able to buy Cell based computers in the near future, or is the PS3 the only way to get your hands on them?

Regardless the fact that this is slightly off-topic, I don't expect Cell based workstation to be available to a mainstream, or even enthusiast base, if you ask me.

IBM with the loss of Apple, no matter how you see it, is in a shit position when it comes to anything else but their xServers business. (I talking about Personal Computers here)

On the other hand, I could see Cell CPU computers replace SGi Stations in some fields, like medical imagery, basic VR simulations, etc...

In other words, Cell is definitely not an X86 contender.
Hell, I don't even think it's a PPC contender (And somehow, the Cell is a PPC + 8 "DSPs").

 

[c0_re]

IMO anyone that thinks there gong to see Realtime raytracing in any next gen game is smoking somthing bad, there has to be much better and cheaper solutions to accomplish what people use RT for.