Isn't the PS3's real power the union of the CELL & RSX?

sonycowboy

Newcomer
Sorry, if this is all basic stuff for you guys :idea:

I'm seeing lots of discussion about each part independantly with the knowledge that the 2 will work together, but isn't that really the focus of the system?

It seems like that union is what drove the design of the PS3.

From the tremendous floating point performance to the use of the XDR memory to the FlexIO pipeline to the RSX and the split of the 2 memory pools. It seems to all point to the CELL taking over ~some rendering from the GPU.

Now here's my question:

Are the some vertex applications that the CELL can do just as well as a GPU? That is, does the GPU sometimes get bogged down doing some vertex transformations that require a full cycle that the CELL can handle just as well, freeing the GPU to handle some of the more complex operations?

If so, is this one of the keys to unlocking the PS3's power or is the XBox 360's CPU just as capable of doing these same things? Or is the Smart EDRAM capable of doing many of the same things.

I'm just trying to understand the architectural choices that Sony has made and why so many developers are pointing to the CELL being the PS3's strongest feature, when in simple FLOP ratings, it's barely 10% of the GPU. (Understanding that CPU is required for actual computation)
 
It's quite hard for current GPUs to create geometry and to return the results from that. A Cell could potentially do most of the required transformations and collision detections that are done by the CPU at this moment. It would be really swell for physics calculations, ray-based effects like sub-surface scattering and for generating and modifying geometry.

In short: everything you would want a PPU to do, like waves in the water, contracting muscles, realistic skin, bouncing boobies, flowing clothes, deformable objects and a destructible environment with lots of explosions.
 
I don't think that you could raytrace with Cell fast enough to get artifact-free results...
 
Laa-Yosh said:
I don't think that you could raytrace with Cell fast enough to get artifact-free results...

I'm sure you can blur the hell out of the result. That would be defy the point somewhat, but at least PS3 can do raytracing. Imagine the forums around the net. ;)
 
london-boy said:
I'm sure you can blur the hell out of the result.

Bluring won't help you, because the view-dependent sampling changes as you move the camera. If you are undersampling too much, then you'll get flashing - this point on the model is now occluded 10%, now 35%, now 5%... same with subsurface scattering. Lots of noise are not something a developer would like introduce, especially after finally getting some AA and AF to get rid of texture/edge aliasing.
 
london-boy said:
Laa-Yosh said:
I don't think that you could raytrace with Cell fast enough to get artifact-free results...

I'm sure you can blur the hell out of the result. That would be defy the point somewhat, but at least PS3 can do raytracing. Imagine the forums around the net. ;)

I remember there was a small demo going around the demo scene that had ray tracing. I think the important thing is the quality of the RT not that it could be done.
 
They choosed to raytrace it because it 's probably the simpliest way to render it.Writing a raytracer is easy and straightforward.
I heard that the 2 cell raytraced height IGN datas ,then they streamed the result to a G5.
 
PC-Engine said:
london-boy said:
Laa-Yosh said:
I don't think that you could raytrace with Cell fast enough to get artifact-free results...

I'm sure you can blur the hell out of the result. That would be defy the point somewhat, but at least PS3 can do raytracing. Imagine the forums around the net. ;)

I remember there was a small demo going around the demo scene that had ray tracing. I think the important thing is the quality of the RT not that it could be done.

What i meant is that for the geeky fanboys on the net, the important is tah reeltyme raee-traycin' ;)
 
Laa-Yosh said:
I don't think that you could raytrace with Cell fast enough to get artifact-free results...

Probably not for a whole scene, but you could generate pre-lighted textures for things like skin. Like you probably couldn't run physics for each and every object, like a path covered with pebbles, or all the individual leafs in a forest. You just have to pick and choose what would benefit most, and upscale everything else. Like only moving whole branches, and calculating only the pebbles underneath the feet.

In that way, you can make the whole scene look very realistic and alive, which would be the whole point.

Edit: wrong quote.
 
As said many time before on this board: Raytracing is the ultimate solution to the shadow problem (at least ultimate in terms of available computingpower now and well into the future). Indeed, raytracing was originally invented to do shadows.
 
Fake solutions are faster, and have worked pretty well so far... I still don't get this raytracing-fetish... ;)
 
Raycasting != raytracing.

Basically, but "Raycasting" implies only the initial rays will be calculated, not recursions. "Raytracing" implies folowing each ray recursion usually, very deeply, so there is a difference.

Later
 
london-boy said:
PC-Engine said:
london-boy said:
Laa-Yosh said:
I don't think that you could raytrace with Cell fast enough to get artifact-free results...

I'm sure you can blur the hell out of the result. That would be defy the point somewhat, but at least PS3 can do raytracing. Imagine the forums around the net. ;)

I remember there was a small demo going around the demo scene that had ray tracing. I think the important thing is the quality of the RT not that it could be done.

What i meant is that for the geeky fanboys on the net, the important is tah reeltyme raee-traycin' ;)

Or teh reeltymz REYES. :LOL:
 
Back
Top