AGEIA PPU Preview

[ironique]

Came across this preview. Sounds interesting. But would you be willing to spend an extra $250 to support better physics in-game? That's on top of the cost of you gfx card....

 

[imported_Kensai]

I will pay that much if there were games that actually used it.

 

[dug777]

Originally posted by: Kensai
I will pay that much if there were games that actually used it.

same, looks like it might be a while before we see developers supporting it...

 

[ironique]

Doesn't this step on dual-core toes... specifically in game development? I read somewhere that Epic was excited about dual-core coz they'd be able to transfer physics calculations to one core while the rest of the game code could be executed by the other core. Though initially developers may still depend on CPU for these physics calculations, eventually when (or if) the PPU picks up, wouldn't they have to change their code again?

 

[Acanthus]

Originally posted by: ironique
Doesn't this step on dual-core toes... specifically in game development? I read somewhere that Epic was excited about dual-core coz they'd be able to transfer physics calculations to one core while the rest of the game code could be executed by the other core. Though initially developers may still depend on CPU for these physics calculations, eventually when (or if) the PPU picks up, wouldn't they have to change their code again?

Unreal Engine 3 supports the Ageia PPU.

 

[biostud]

As the generation cycle of PPU will probably be a lot slower than video cards it could be a good investment.......that is if it's well supported in software.

 

[Kalessian]

I think it's awesome. I really hope it catches on because current in game physics are grievously lacking.

 

[sbuckler]

Originally posted by: ironique
Doesn't this step on dual-core toes... specifically in game development? I read somewhere that Epic was excited about dual-core coz they'd be able to transfer physics calculations to one core while the rest of the game code could be executed by the other core. Though initially developers may still depend on CPU for these physics calculations, eventually when (or if) the PPU picks up, wouldn't they have to change their code again?

The same as graphics cards I think you need lots of cpu grunt to just keep your physics card busy. Hence the more cpu's the better.

 

[ryanv12]

The Unreal 3 engine seems to support everything. I wonder how they're going to work it so that it still runs on computers that are single core and without physics cards (although I don't know who will be caught with such primitive technologies ). I suppose that depends on the game developer to take advantage of these features? Either way, I'd be interested in getting a physics card when the games are actually supporting it. For a long time I had always thought that the reason games can still be considered unrealistic is because of the lack of environment interaction. I'm definitely excited for enhanced physics.

 

[Drayvn]

Originally posted by: ironique
Doesn't this step on dual-core toes... specifically in game development? I read somewhere that Epic was excited about dual-core coz they'd be able to transfer physics calculations to one core while the rest of the game code could be executed by the other core. Though initially developers may still depend on CPU for these physics calculations, eventually when (or if) the PPU picks up, wouldn't they have to change their code again?

Well it doesnt seem to as Unreal 3 is using the Novodex Physics system which can utilize both the PPU and a CPU if necessary.

Also with the PPU, as its a dedicated Physics unit it can do 20 or more thousands of physics calcualtions compared to a CPU or multicore CPU where it can do a max of 100 or so.

 

[Insomniak]

If implemented, it will be revolutionary. If not, it will disappear.

I sincerely hope it gets used. Physics are the new graphics.

From what I've seen, the PPU is basically a very high throughput cluster of FPU units that's very parallel and slides through the core logic via the PCI/PCIe bus - it's good for processing physics, but little else. However, seeing a house come apart brick by brick and shingle by shingle when exploding would be an amazing experience I'd really like to see.

 

[reever]

It uses 2 slots?

 

[Insomniak]

Originally posted by: reever
It uses 2 slots?

No - there will be a PCI version and a PCIe version.

 

[xtknight]

for all of those saying that dual-core CPUs could be used instead--absolutely not...
this is a specialized processor, just like a GPU and some people are missing the point...can you run any game with a dual-core system and no graphics card? not even close...

Current games running on PC systems with high-end desktop processors, such as the Intel Pentium 4 or AMD Athlon 64, can support roughly 30 to 40 "active bodies," or physical objects that can interact with each other in-game. This limitation doesn't give developers much to work with in terms of physics simulation. Simulating a building blowing up in real time is impossible with such a small number of fragments, but increase the active body count to 32,000 or 40,000, which the Ageia PhysX PPU can handle, and then you'll have an explosion to talk about.

A dual-core CPU would best 80 "active bodies" while the PPU will still be much farther ahead. Do the math.

 

[Cheesetogo]

The only way I see this really working is if it gets put onto graphics cards, I really don't see many people buying another component At 300 dollars, not many people will buy it, and thus not many developers are going to have this implemented in games. However, if it gets implemented into lots of graphics cards, I thnk it will work.

 

[StrangerGuy]

Originally posted by: xtknight

A dual-core CPU would best 80 "active bodies" while the PPU will still be much farther ahead. Do the math.

The problem is that Dual-core CPUs will have a far larger share penetration than a discrete PPU, as Intel is aggressively pushing dual-cores now and within the next few years or so. Unless there is a viable way push the PPU into the mainstream, it's going to be a flop.

 

[Acanthus]

Originally posted by: StrangerGuy

Originally posted by: xtknight

A dual-core CPU would best 80 "active bodies" while the PPU will still be much farther ahead. Do the math.

The problem is that Dual-core CPUs will have a far larger share penetration than a discrete PPU, as Intel is aggressively pushing dual-cores now and within the next few years or so. Unless there is a viable way push the PPU into the mainstream, it's going to be a flop.

If they get havok, they will get half of the games on the market with high end physics.

 

[bunnyfubbles]

It is all up to game developers in order for something like this to succeed. Parallelism is obvious the solution for the immediate future, this just brings us a step closer. I'd always wondered why someone hadn't come along soon after the GPU to produce dedicated parts for game physics and/or AI. It has taken a while but here we are, on the virge of seeing something that could be really cool if embraced. If it just adds some ho-hum effects and doesn't offer some drastically different and wild game play (that is also very fun), then this product could end up flopping on the PC in head to multicores. However it is definately much faster for physics than even a dedicated CPU core could be, and would be a very welcome addition. But I've got a feeling, that if there are games that are produced to both support it and function without it, we won't be getting the best the PPU would have to offer.

 

[ddogg]

Originally posted by: Kensai
I will pay that much if there were games that actually used it.

:beer:

 

[MetalStorm]

I think CPUs can actually do many many more physics calculations than they mention, probably a couple of thousand quite easily, it's just that number is massively reduced when the CPU also has to work on running the game engine and all the bits associated with that.

 

[nitromullet]

I've got a feeling, that if there are games that are produced to both support it and function without it, we won't be getting the best the PPU would have to offer.

I agree with that, but that is going to have to be the way it is in the beginning. However, I think that the PPU could also take off if it creates an overall increase in game performance. For instance, say due to CPU limitations a game runs at given fps, but with the PPU offloading the physics portion of the processing from the CPU you all of a sudden get a 20-30% increase... I think you would have a marketable item at that point for a segment of the market.

I think the real way to market this thing is to try to get Asus to include the chip on their motherboards or even graphics cards (it might justify the high cost of Asus graphics cards). IMO, if Ageia wants this thing to spread, they are most likley going to have to sell them at a loss to get them out there. I think that a discrete card for $300 that doesn't do anything tangible will be a hard sell. To make matters worse for them is that unlike video cards, non-gamers simply won't buy these EVER. Whereas, you can sell a non-gamer a cheap video card and make a few bucks.

Well it doesnt seem to as Unreal 3 is using the Novodex Physics system which can utilize both the PPU and a CPU if necessary.

from: http://www.megagames.com/news/html/hardware/physicsdedicatedhardwaresoon.shtml

Tim Sweeney, founder and lead programmer of Epic Games comments, We've been using the NovodeX Physics SDK with the Unreal Engine 3 for the past year and it has added some awesome effects. It's going to be exciting to see what NovodeX can do once the PhysX chip hits the market.

Either way, I think they have a good idea on their hands, and if Novodex starts to show up in a few games I might just get a PPU to support people with good ideas.

 

[Drayvn]

Supposedly i read that they are doing different versions of this PPU. Cheaper ones i mean, which of course wont be able to do as many calculations but would be more affordable.

And they also said in some interview a few months back that they do want their PPU intergrated on motherboards of GPUs as that is their long term goal. And they are talking with everyone they can to try and use it.

Also if other physics companies produce programs, i would have thought they would be able to encode it to use this PPU. I didnt think Ageia would only make it work on their programs, that would be suicide, its like onyl certain games can work on certain GPUs.

 

[MetalStorm]

Originally posted by: Drayvn
Supposedly i read that they are doing different versions of this PPU. Cheaper ones i mean, which of course wont be able to do as many calculations but would be more affordable.

And they also said in some interview a few months back that they do want their PPU intergrated on motherboards of GPUs as that is their long term goal. And they are talking with everyone they can to try and use it.

Also if other physics companies produce programs, i would have thought they would be able to encode it to use this PPU. I didnt think Ageia would only make it work on their programs, that would be suicide, its like onyl certain games can work on certain GPUs.

It would make sense to have different levels of the card - afterall if you have a more complicated scene, then you're going to need a graphics card with sufficient rendering power to display it at a good FPS, so people with low end graphics cards won't get overwhelmed by 40,000 objects going everywhere.

I think the chance of the PPU being put on graphics cards is very low, for a start it will first have to become a compulsary item, and secondly you've got to try and fit a whole card on to the graphics cards... The PPU already has it's own heatsink and fan, and 128mb of memory, I think it's unreasonable to expect it on graphics cards in the next 3-4 years at least!