How can Cell help RSX *changed

NERO

Regular
Greetings B3D members.

In the latest issue of Electronic Gaming Monthly Julian Eggbrecht (President of Factor 5) mentions some benefits of using the PS3's hardware. He claims, "For Lair, the Cell is used to govern A.I. and generate all the polygonal models in the game, from soldiers to creatures to terrain." (pg. 40, issue 209)

Governing A.I. is expected of a CPU but generating polygonal models? Is this propaganda (as in every CPU functions like such but Julian won't admit)? Or does this explain anything new about Cell? (as in alleviating RSX from strain)
 
There is nothing stopping the Cell from generating geometry.. indeed any cpu can do it. (but Cell may be very good at it)

In Factor 5's case, I think what he is talking about with the "[Cell] generate all the polygonal models in the game" is that Cell is used to implement their (previously mentioned) LOD system. So basically, they have some high-poly (or perhaps even mathematically reprensented?) base model, and the Cell is used to scale it according to scene complexity demands every frame. RSX then reads in the results and lights/shades/renders them.

Just my extrapolations, of course.
 
Last edited by a moderator:
Thank you for a concise reply Bohdy. With all of the talk of now gen GPUs and numerous attempts to guess at the "secret" of some specially powered G70 (RSX), I wondered whether the talk of the number of triangles drawn - a number that the Xenos (released earlier) trumps - would have been rendered mute in regards to the Cell's assistance of RSX. Ignorance of the relationship between a CPU and GPU's share of drawing, scaling and shading leads to the question of; Is the console business' latest piece of hardware truly special (As Factor 5 implies) or not? We know of the main components of equipment in this piece of hardware and how these components link to each other, but is there any relevance, any newness to the ability of this particular CPU's assistance to its GPU?
 
Thank you for a concise reply Bohdy. With all of the talk of now gen GPUs and numerous attempts to guess at the "secret" of some specially powered G70 (RSX), I wondered whether the talk of the number of triangles drawn - a number that the Xenos (released earlier) trumps - would have been rendered mute in regards to the Cell's assistance of RSX. Ignorance of the relationship between a CPU and GPU's share of drawing, scaling and shading leads to the question of; Is the console business' latest piece of hardware truly special (As Factor 5 implies) or not? We know of the main components of equipment in this piece of hardware and how these components link to each other, but is there any relevance, any newness to the ability of this particular CPU's assistance to its GPU?

The PS3 hardware is definitely special, like the 360 hardware and like a lot of other hardware too. They all have their strengths and they are all special in their own way. And they are all inevitably useless at some things too of course.
 
Nero, most of your question has been answered before. Just use the search function. E.g.: This thread here contains a good interview on Cell. This thread contains some good information in the last few pages.
Etc.

Bottom line is, use the search function first, usually there's already lots of stuff on a particular topic.
 
Thanks much hupfinsgack. The final few pages of the second thread provided are very informative, especially the three-way interchange between Arwin, nAo and PiNky.

As for london-boy's response: Agreed. It is understood that all systems are special in their own way (like children and three-legged pets), but I had still wondered whether Cell's functionality with the gpu is different from other past cpu to gpu relationships to the point that it is a selling point for a "next generation" system. From what I have read it seems that this is not yet the case if ever.

Cheers
 
I had still wondered whether Cell's functionality with the gpu is different from other past cpu to gpu relationships to the point that it is a selling point for a "next generation" system.
I would say 'yes' due to the fast interconnect between the two processors, and the suitablity of Cell in many graphics tasks where traditional processors would have to sacrifice a lot of their available resources to provide the same contributions from the CPU.
 
Personally, what i expect, is low polygon count stuff, as imagine, deformable pipes, metalic objects, where whole parts is able to be ripped apart, and the places where the whole parts MEET, could be deformed, and made run time by the Cell, and fed to RSX. the result, would be, lets take a example:

- a plasma gun fires, hit a pipe(several places). the pipe explode (effects), and a part of it falls partly down. the end is still conected, but deforms, and is still a whole connected object, even when its bending. the whole object is made on site, by Cell, and would be different EACH time, but is made of 2 mechanical parts fused by mechanics and graphics. It would look as a pipe, with a half melted section, with a whole part angled down, who is held dangling from the part bent by the heat, while the other end, is melted straight off.

I also guess, it could be possible to deforme landscape, enemies, and other stuff.
 
Kewl thread :)

I've always thought that the CPU always starts out the T&L then passes it to the GPU to accelerate it...
Maybe, the cell is more efficient in pushing geometry than the RSX so the developers prefer to use this functionality in order to free some resources the GPU could dedicarte elsewhere.
I'm actually quite intrerested in how the cell aids in post processing... would love to hear more about that.
 
From the look of the flight tutorial video for Lair, anyone else getting the same impression about the terrain? It looks procedurally-generated. Not that it's something you can instantly tell, but I just noticed an uncanny resemblance to that old PS2 terrain demo where the point was to showcase fractal-generated terrain. I could be wrong, it just has that Bryce-ish look (better than Warhawk though) that I thought smacked of algorithmic origins. I thought it might be pertinent to this particular thread. PEACE.
 
From the look of the flight tutorial video for Lair, anyone else getting the same impression about the terrain? It looks procedurally-generated. Not that it's something you can instantly tell, but I just noticed an uncanny resemblance to that old PS2 terrain demo where the point was to showcase fractal-generated terrain. I could be wrong, it just has that Bryce-ish look (better than Warhawk though) that I thought smacked of algorithmic origins. I thought it might be pertinent to this particular thread. PEACE.


Yeah I'm thinking the samething too.
 
From the look of the flight tutorial video for Lair, anyone else getting the same impression about the terrain? It looks procedurally-generated...
I haven't seen that vid, but I don't imagine so. Eggbrecht has spoken up about use of technology (and how great they are!) and hasn't once mentioned 'we're creating terrain on the fly which just isn't possible on lesser hardware.' He has said that the terrain is a lot more detailed than previous efforts, with 'detail every 2 metres' that it's guessed means a vertex in the mesh every 2 metres, though that comment wasn't clarified. Still, if they're using procedural lanscape generation, why not tell everyone when you're spealing off all the features of your engine?
 
Do you think they might be trying to use rsx for basic terrain handling, while cell doing procedurally generated geometry details in close range. For example, rsx handling basic terrain form and structures(buildings, vehicles), while Cell handling procedural geometry details such as trees, rocks, boxes, etc... a mix of both techniques.

Could it be made or would dependency be bad issue?
 
Do you think they might be trying to use rsx for basic terrain handling, while cell doing procedurally generated geometry details in close range. For example, rsx handling basic terrain form and structures(buildings, vehicles), while Cell handling procedural geometry details such as trees, rocks, boxes, etc... a mix of both techniques.

Could it be made or would dependency be bad issue?

Or maybe they used a fractal to generate the basic "random" terrain, and didn't put enough randomness in it. I takes a lot of time and effort to generate that level detail if you are going to do it by hand, so guess they would have used some sort of fractal generator to do it. The only other option would be to take a digital terrain map of a real location on Earth and use that with fractalisation on top of that to give more detail.
 
Julian Eggebrecht:

The multicore processing powering th Ps3 is a turbo-powered leap over what Factor5 was used to working with on the Gamecube.
For Lair the Cell is used to govern A.I. and generate all polygonal models in the game, from soldiers to creatures to terrain.
 
Julian Eggebrecht:

The multicore processing powering th Ps3 is a turbo-powered leap over what Factor5 was used to working with on the Gamecube.
For Lair the Cell is used to govern A.I. and generate all polygonal models in the game, from soldiers to creatures to terrain.

Does that really deserve to be quoted? I'm sure teams with significant coding depth (and some HW exclusivity) find the Cell's internal bandwidth, flop and integer count, memory structure, and independent co-processor speed to be extraordinary. Maybe even revolutionary (I know I do) but that goes without saying... doesnt it?
 
Or maybe they used a fractal to generate the basic "random" terrain, and didn't put enough randomness in it. I takes a lot of time and effort to generate that level detail if you are going to do it by hand, so guess they would have used some sort of fractal generator to do it.


I did not think fractal generator after reading the article but it does make sense considering that from all I've read (and have tried to understand) CBE lends itself to the calculations entailed by fractal generation, ergo it is a fast math crunching machine. It once again goes back to whether the Cell, while in the PS3, can do this efficiently to the point that it can be sold as a generational leap in gaming system architecture that will cause a generational leap in creativity in games.
 
For Lair the Cell is used to govern A.I. and generate all polygonal models in the game, from soldiers to creatures to terrain.

As far as I know, generating polygonal models is meant in the context of their LOD method and not in the context of procedural creation.
 
Back
Top