Ageia Physx

[karka]

You know Agea Physx is a some kind of unit for processing physical calculations. Think your GPU to imagine it. But this card is only dedicated to physical calculations. Now go to http://www.ageia.com/physxInAction/games.html and scroll down the page to the end.

 

Yes, Obsidian is using Physx. I think this is a good thing...

Edited January 14, 2006 by karka

[Meshugger]

They're implementing the SDK into their games, meaning that you'll see some nice physic effects like games that use Havoc, but that's about it. I would like to see a game that would require the physics-chip in order to run.

[karka]

Yes, it's a SDK but AFAIK it runs even better with a PPU. Also it is taking the load of physics calculation from the CPU which is a good thing too

Edited January 14, 2006 by karka

[Meshugger]

Yes, it's a SDK but AFAIK it runs even better with a PPU. Also it is taking the load of physics calculation from the CPU which is a good thing too

<{POST_SNAPBACK}>

 

Theoretically, yes. Since a single PPU would do the physics, while the other's handle A.I. and so on. But it's difficult to implement, so i believe it when i see it

[@\NightandtheShape/@]

<_< this isn't new to me, I've been looking at the SDK and API for a while. I'm pretty excited about the whole PPU thing.

 

The API is to the PPU, what DirectX and OpenGL is to the GPU.

 

Physics processing & Graphics processing arn't that far apart, you can for example currently hi-jack the GPU to perform certain collision detection computation.

 

Naturally they're not the same beast...

 

As for AI, I believe that'll stay on the CPU, AI is a RAM hog mostly, by taking the load for collisions and physics to a seperate card, along with the load for graphics being outsourced, aswell as sound... well you get my point here, itr leaves the CPU a little freer, so we should see beter AI as a result aswell .

[karka]

<_< this isn't new to me, I've been looking at the SDK and API for a while. I'm pretty excited about the whole PPU thing.

 

The API is to the PPU, what DirectX and OpenGL is to the GPU.

 

Physics processing & Graphics processing arn't that far apart, you can for example currently hi-jack the GPU to perform certain collision detection computation.

 

Naturally they're not the same beast...

 

As for AI, I believe that'll stay on the CPU, AI is a RAM hog mostly, by taking the load for collisions and physics to a seperate card, along with the load for graphics being outsourced, aswell as sound... well you get my point here, itr leaves the CPU a little freer, so we should see beter AI as a result aswell .

<{POST_SNAPBACK}>

 

AFAIK, on next gen. consoles some developers are dedicating one of cores, to the physical calculations. This is the way of implementing Ageia Physics on consoles.

[WITHTEETH]

I see this as a big performance booster also, but... It does not support any current games, thus why its not out yet. The Ageia needs support for their card through developers. ATI is currently working on building physics processor into their card. I wonder what support would be like with ATI, if it would need special support like AGEIA.

[@\NightandtheShape/@]

<_< this isn't new to me, I've been looking at the SDK and API for a while. I'm pretty excited about the whole PPU thing.

 

The API is to the PPU, what DirectX and OpenGL is to the GPU.

 

Physics processing & Graphics processing arn't that far apart, you can for example currently hi-jack the GPU to perform certain collision detection computation.

 

Naturally they're not the same beast...

 

As for AI, I believe that'll stay on the CPU, AI is a RAM hog mostly, by taking the load for collisions and physics to a seperate card, along with the load for graphics being outsourced, aswell as sound... well you get my point here, itr leaves the CPU a little freer, so we should see beter AI as a result aswell .

<{POST_SNAPBACK}>

 

AFAIK, on next gen. consoles some developers are dedicating one of cores, to the physical calculations. This is the way of implementing Ageia Physics on consoles.

<{POST_SNAPBACK}>

 

 

I don't know if the performance gain is the same.

[@\NightandtheShape/@]

I see this as a big performance booster also, but... It does not support any current games, thus why its not out yet. The Ageia needs support for their card through developers. ATI is currently working on building physics processor into their card. I wonder what support would be like with ATI, if it would need special support like AGEIA.

<{POST_SNAPBACK}>

 

ewwwwwwww

[WITHTEETH]

I see this as a big performance booster also, but... It does not support any current games, thus why its not out yet. The Ageia needs support for their card through developers. ATI is currently working on building physics processor into their card. I wonder what support would be like with ATI, if it would need special support like AGEIA.

<{POST_SNAPBACK}>

 

ewwwwwwww

<{POST_SNAPBACK}>

Sorry, my psychic abilities are failing me, can you put your feelings into words?

Edited January 15, 2006 by WITHTEETH

[karka]

I don't think so. Because, they are doing calculations with software. For example, in DSP cards there are some modules for calculating complex functions like sin, log, etc.. So, this makes these DSP cards very expensive and very fast.

 

I think, PPUs have similar architectures. "Emulating" the cards functions with a processor is always slower.

[WITHTEETH]

Wow, you've managed to go over my head! :D

 

What does DSP mean?

[karka]

Discrete/Digital Signal Processing. For example, your GPU is a DSP card actually, or in MP3 players there are DSP chips.

[WITHTEETH]

Discrete/Digital Signal Processing. For example, your GPU is a DSP card actually, or in MP3 players there are DSP chips.

<{POST_SNAPBACK}>

I don't see how a Added Physics core to a Graphic card can go to slow. An added graphic core to a GPU making 2 graphic cores all together helps run at Near SLI speed. What im trying to say is, is that it seems as though to me that the physics core on a graphic card would run very solid.

 

I don't see the huge advantage as long as the pipelines (pci Express) aren't bottlenecking the 2 cores.

[karka]

 

Actually, i am not saying embeded physics to a GPU makes it slow. I am saying, dedicating the cores of a multiple cored CPU is somehow slower than a PPU.

 

But, somehow, if these cores have some complex functions which are using for calculating the physics, dedicating the core can be much more faster. Because, capacitance and propagation delay can be decreased with a dedicated core.

[WITHTEETH]

Thats the thing is that there is a dedicated core. I am not talking about a dual core but a completley seperate core that is on the same graphic card. Take a look at this graphic card, there are 2 cores, one to the left and right.

 

Lets say the one on the right is the dedicated physics chip, I con't see ohw it could be slower. Maybe I'm reading you wrong and you already understood this.

Edited January 15, 2006 by WITHTEETH

[karka]

I think we are saying the same things. For exampe, if these two chips are capable to calculate complex functions(sin, log, integrate etc..) by "hardware only" it is best to use them like you said.

 

But, if they are calculating complex funcitons by "software only" this increase calculation times radically. So it is always good to use some extra hardware modules to calculate these complex functions.

 

I entered my messages for next.gen consoles. They have GPU's. But not PPU's. So, developers must use CPU cores. If these cores are not capable(hardware wise) enough physics calculations may take very long time.

Edited January 15, 2006 by karka

[WITHTEETH]

I think we are saying the same things. For exampe, if these two chips are capable to calculate complex functions(sin, log, integrate etc..) by "hardware only" it is best to use them like you said.

 

But, if they are calculating complex funcitons by "software only" this increase calculation times radically. So it is always good to use some extra hardware modules to calculate these complex functions.

 

<{POST_SNAPBACK}>

Word shizzle. :cool:

[alanschu]

As for AI, I believe that'll stay on the CPU, AI is a RAM hog mostly, by taking the load for collisions and physics to a seperate card, along with the load for graphics being outsourced, aswell as sound... well you get my point here, itr leaves the CPU a little freer, so we should see beter AI as a result aswell .

<{POST_SNAPBACK}>

 

I disagree. Why would AI be a RAM hog? The two big concerns are always time and space, and many algorithms have been created that can store AI searches in linear space. They still have the exponential factor with respect to time though. There are many ways to trim branches of trees so that they are not stored in memory, but you'll still eventually need to traverse them, so the time consideration comes into play.

 

I'll admit I'm no expert, but algorithms such as IDA* were created because running out of memory is much, much worse than running out of time. A standard DFS will keep things linear in terms of memory usage, but runs the potential of hitting a black hole to which it goes on forever without finding a solution. IDA* ensures that other branches get checked, but has a drawback of going over nodes multiple times. For most useful algorithms, it seems as though the time explodes much faster than space requirements. Space requirements can also be reduced by an appropriate heuristic. However, depending on the type of heuristic, that calculation could be expensive, time wise.

 

 

I actually emailed a professor at my University, Jonathan Schaeffer (A quick bio). According to him, he feels that AI will continue to be pushed to the back burner unless someone makes an add on card for AI purposes.

[angshuman]

What's nice is that a CPU is great for running control flow-intensive DFS-like algorithms.

 

In fact, as far as I knew, even "physics" as used in games primarily involves collision detection, and collision detection algorithms are also typically control-heavy and thus are well suited to execution on standard processors with good branch predictors. What's weird is that the Cell's SPU's are completely non-speculative, which seems to render them completely useless for this task, so what the hell are those SPU's good for anyway?

[WITHTEETH]

What is a SPU?

[alanschu]

What's nice is that a CPU is great for running control flow-intensive DFS-like algorithms.

 

In fact, as far as I knew, even "physics" as used in games primarily involves collision detection, and collision detection algorithms are also typically control-heavy and thus are well suited to execution on standard processors with good branch predictors. What's weird is that the Cell's SPU's are completely non-speculative, which seems to render them completely useless for this task, so what the hell are those SPU's good for anyway?

<{POST_SNAPBACK}>

 

 

An excellent point. IMO the term "software mode" is often misused.

[angshuman]

What is a SPU?

A "Synergistic Processing Unit". It's a processor with a large amount of brute force but a lot less intelligence (that is why it needs an army of programmers to extract its full performance). The PS3's Cell is made up of of 1 PPU (Principal Processing Unit?) and 7 of these SPUs.

[Pidesco]

What is a SPU?

A "Synergistic Processing Unit". It's a processor with a large amount of brute force but a lot less intelligence (that is why it needs an army of programmers to extract its full performance). The PS3's Cell is made up of of 1 PPU (Principal Processing Unit?) and 7 of these SPUs.

 

I think it's Primary Processing Unit.

[karka]

Geneally it means Physics P.U.

 

http://en.wikipedia.org/wiki/Physics_processing_unit

Edited January 17, 2006 by karka