PhysX: PS3 can handle it; 360 features limited

pjbliverpool said:
I don't think its been said anywhere that you need a PPU plus dual core CPU to match cells abilities. All thats been said is that it takes a PPU enabled PC or cell to support all of the Novadex features. I don't think thats dependant on dual core and I don't think its saying that they support those features to the same degree.

They may both support the same features while the PPU can handle them twice as fast when coupled with a modest single core CPU.

SenatorMonkey said:
On a dual-core system with the PhysX chip, it ran at about 40fps, and took only 25% of one core.

A poster already stated that a dual-core CPU with a PPU still uses up 25% of one of the cores....
 
Last edited by a moderator:
BlueTsunami said:
A poster already stated that a dual-core CPU with a PPU still uses up 25% of one of the cores....

which surely means it would take up 25% of a single core CPU aswell?
 
Alpha_Spartan said:
I can imagine the sychophants blowing this way out of proportion. "OMG!! Xbox 360 is teh too weak to run PhysX!" All this states is that the Novonyx SDK in its current form is better suited for the PS3's architechture and PhysX processors.

The PhysX chip is very similar to CELL, so architecturally portability between the Novodex SDK between the two should be a big plus. From a design perspective Novodex sees PC+PhysX, PS3, and Xbox 360. From an architecture standpoint the Xbox 360 is the odd man out.

While I expect the PS3 to do physics better when all things are said and done at the end of this gen, I will wait to see if other physics developers have the same issue. This could be a limiting factor on Xbox 360, or it could be the Novodex methods for dealing with physics. I tend to think it is the later considering their own chip.

Where this is relevant is UE3 is using Novodex, although I think Havok plugs in as well.
 
pjbliverpool said:
which surely means it would take up 25% of a single core CPU aswell?

Probably (not to sure if one core of a dual-core CPU would translate to the same exact performance of a modest single core CPU]...but where do you get the...

pjbliverpool said:
They may both support the same features while the PPU can handle them twice as fast when coupled with a modest single core CPU.

what are you comparing?
 
I think Deano and the Ninja Theory team is using Havok's physics engine right? I wonder what the Havok team will say later this week at the second conference? What is the date of that thing anyway?
 
mckmas8808 said:
I think Deano and the Ninja Theory team is using Havok's physics engine right? I wonder what the Havok team will say later this week at the second conference? What is the date of that thing anyway?

The Eurographics conference actually finishes tomorrow. The site is here:

http://isg.cs.tcd.ie/eg2005/

The keynote was this:

Keynote Talk 2: Steven Collins, co-founder of Havok
Kinematics, Dynamics, Biomechanics: Evolution of Autonomy in Game Animation

Thursday, 1st September, 2005. 16:00-17:00.
VENUE: Burke Theatre
SESSION CHAIR: Carol O'Sullivan



The believable portrayal of character performances is critical in engaging the immersed player in interactive entertainment. The story, the emotion and the relationship between the player and the world they are interacting within are hugely dependent on how appropriately the world's characters look, move and behave. We're concerned here with the character's motion; with next generation game consoles like Xbox360™ and Playstation®3 the graphical representation of characters will take a major step forward which places even more emphasis on the motion of the character. The behavior of the character is driven by story and design which are adapted to game context by the game's AI system. The motion of the characters populating the game's world, however, is evolving to an interesting blend of kinematics, dynamics, biomechanics and AI driven motion planning.

Our goal here is to present the technologies involved in creating what are essentially character automata, emotionless and largely brainless character shells that nevertheless exhibit enough "behavior" to move as directed while adapting to the environment through sensing and actuating responses. This abstracts the complexities of low level motion control, dynamics, collision detection etc. and allows the game's artificial intelligence system to direct these characters at a higher level.

While much research has already been conducted in this area and some great results have been published, we will present the particular issues that face game developers working on current and next generation consoles, and how these technologies may be integrated into game production pipelines so facilitate the creation of character performances in games. The challenges posed by the limited memory and CPU bandwidth (though this is changing somewhat with next generation) and the challenges of integrating these solutions with current game design approaches leads to some interesting problems, some of which the industry has solutions for and some others which still remain largely unsolved..

I've yet to see any reporting on eurographics unfortunately :(

Other presentations of interest include:

COLLADA 1.4: Mastering next-generation 3D game assets (Sony) (http://isg.cs.tcd.ie/eg2005/IS3.html)
New games console architectures (ATI, IBM & NVIDIA) (http://isg.cs.tcd.ie/eg2005/IS4.html)
Next generation game technology (talks + panel) (http://isg.cs.tcd.ie/eg2005/IS5.html)

I kind of doubt we'll hear much of anything about it now.
 
pjbliverpool said:
which surely means it would take up 25% of a single core CPU aswell?
Good point, but I think what Aegia were saying is that 75% of a single core isn't enough for all the other game stuff the processor needs to do. So though single CPU+PhysX would run the simulation at a decent speed it'd potentially slow the rest of the game down. Of course I guess that's maxing out the physics content which might well be overkill. Would anyone mind dropping 6000 rocks down to 3000? Still a lot of rocks!
 
mckmas8808 said:
Well it does kinda matters. Games like Splinter Cell: CT had some advantages on the Xbox over the PS2. The biggest one to me being the graphics. The lighting in the Xbox version was WAAAY better than on the PS2. Its the small stuff like that, that can change a game IMHO.
So what if we hear that Havok runs better on Xenon? What then. My point is that this is just one isolated SDK.

I tried looking it up, but I couldn't find it, but I'd bet that the Cell implementation of the SDK is similar to that of the actual PPU, thus the advantage. I don't think they would go out of their way to optimize the SDK for the Xenon when they have their own hardware.
 
BlueTsunami said:
Probably (not to sure if one core of a dual-core CPU would translate to the same exact performance of a modest single core CPU]...but where do you get the...



what are you comparing?


Im not saying thats the case, im just saying that from what we know so far, it could be. The reverse could also be true.

A more relevant question IMO is that when Cell is matching the PPU's peak performance (if it can) what does that leave over for everything else like Game control, vertice generation, AI etc... when compared to say a PPU coupled with a dual core Athlon?
 
Alpha_Spartan said:
I don't think they would go out of their way to optimize the SDK for the Xenon when they have their own hardware.
If they don't optimize their engine for XB360 they'll lose out custom to those engines that ARE optimized, such as Havok. Unless they think the XB360 isn't going to be big enough to worry about, and if they think their PhysX chip for PC will outsell it I think they're being extremely optimisitc, it would be foolish to release a product without making it the best they could within their level of expertise.
 
"The comparison was that on a single-core system, it ran at about 5fps and occupied the full processor. On a dual-core system with the PhysX chip, it ran at about 40fps, and took only 25% of one core."

image004.gif


http://playstation3.cz/redirect.php?url=http://www.overclockers.cl/news/ageia/ageia_en_02.htm

"Before publishing, we were able to get our hands on the PhysX SDK demo (recently named NovodeX Rocket) and we were able to record a video to show a bit more about AGEIA’s technology. This demo called “Jenga Tallâ€￾ is composed of 360 bodies and the animation is quite smooth even though it is being done by the CPU. The bodies you will see in this video were manipulated in real time and no scripting of any sort was done. As a final note, one of the most impressive demos we found was called “Building Explodeâ€￾. This one handles more than 4000 bodies and the animation slows to a crawl, showing the need to have dedicated hardware for physics calculations as proposed by AGEIA."



... w/out the PhysX chip is PC in deep shit,speaking of PS3 games ports to PCs,and the Physx chip will become part of mainstream in horizont of 3 yrs,meanwhile will PS3 continue to maximalize its effectivity.
... good luck.
 
Shifty Geezer said:
Good point, but I think what Aegia were saying is that 75% of a single core isn't enough for all the other game stuff the processor needs to do. So though single CPU+PhysX would run the simulation at a decent speed it'd potentially slow the rest of the game down. Of course I guess that's maxing out the physics content which might well be overkill. Would anyone mind dropping 6000 rocks down to 3000? Still a lot of rocks!

True, but 75% of what CPU? Are they talking an AthlonXP 2800+ or an FX57? If you need 90% of the AXP you probably only need about 60% of the FX57 for the same result.
 
Acert93 said:
The PhysX chip is very similar to CELL ...
I'd love to see more evidence for this, in the form of an architectural overview of PPU...

The diagram I remember seeing (and then losing, sigh) looked nothing like Cell. It consisted of lots of fixed-function blocks, 10-15 of them in total. Of course that might have been a logical architecture, rather than a physical one...

Jawed
 
pjbliverpool said:
Im not saying thats the case, im just saying that from what we know so far, it could be. The reverse could also be true.

A more relevant question IMO is that when Cell is matching the PPU's peak performance (if it can) what does that leave over for everything else like Game control, vertice generation, AI etc... when compared to say a PPU coupled with a dual core Athlon?

Yeah...and I guess thats the problem with these comments about PhysX being good on the Cell compared to other settings. It would have been cool if they release a diagram of the exact numbers for each setup (so we could see exactly how much of Cell is devoting to the physics calculations, then compare it to the others)....but I guess we'll have to wait for that.
 
This was impressive from a capability aspect:
MH: Bet On Soldier, from Digital Jesters, is using the Ageia PhysX SDK and will ship with hardware optimized features. For instance, the Gas Launcher weapon will be used interactively to “shootâ€￾ acid, hallucinogen gas grenades and incendiary “flameâ€￾ grenades. It is obviously the capacity of PhysX to manage the fluids in real time, which makes the Gas Launcher so interesting and impressive.
At least it provides an example of the importance of good liquid physics. I wouldn't mind seeing a demo of that. :smile:

-aldo
 
aldo said:
This was impressive from a capability aspect:
At least it provides an example of the importance of good liquid physics. I wouldn't mind seeing a demo of that. :smile:

-aldo

Neither would I. Their "programmer art" tech demos don't do them much favours ;)
 
Titanio said:
He didn't indulge in much tech talk - doesn't sound like he was as in the loop with regard to relative PS3 performance, for example, as the other guys doing these presentations. The " From what we understand, the Cell processor will have physics capabilities" bit is kind of funny from that perspective :D

GS: Finally, what features would you like improve on your PPU design for the next version?

MH: Next generation PhysX processors will add additional processing capacity to support features such as hair and clothing simulation. Our goal is to simultaneously simulate many different physics effects so that the entire experience becomes richer. However, we are not in a position to announce specific details of a pre-released product at this time.

... PPU1,PPU2,PPU3 .. the begining of the end,total mess,each game has 5 versions /for single core,dual core,for PPU1 - no hairs,for PPU2 ... LOL/
 
Last edited by a moderator:
Jawed said:
I'd love to see more evidence for this, in the form of an architectural overview of PPU...

The diagram I remember seeing (and then losing, sigh) looked nothing like Cell. It consisted of lots of fixed-function blocks, 10-15 of them in total. Of course that might have been a logical architecture, rather than a physical one...

Jawed
And if that's true, it explains the results and we know it was relatively easy to port PhysX code to the Cell. I don't believe they hard coded three different sets of code for the platforms. They just ported between platforms and the PC and the PS3 ran the fastest.
 
Back
Top