Nvidia following with Physics Simulation on GPUs

Nooooooooooooooooooooooooo! *echo and fade*

 

Yeah, that's good stuff. Anything that helps to make the game world more "live" i.e. less static is good in my books. Whether it's ultimately done with a GPU, MC-CPU or PPU, I don't care. Just bring on the physics explosion, thank you

Here's the AGEIA GDC presentation info: http://www.cmpevents.com/GD06/a.asp?option=C&V=11&SessID=2102

 

http://www.cmpevents.com/GD06/a.asp?option=C&V=11&SessID=2113

 

Not this physics crap again...

 

Ageia may be dead before they start. SLI Physics

I guess some here may have been right. It might make more sense to use the 2nd video card to do physics instead of an add in ppu. PPU needs to a) come out to market b) Be cheaper than they've announced so far.

http://www.tgdaily.com/2006/03/20/nvidia_sli_forphysics/

http://www.firingsquad.com/news/newsarticle.asp?searchid=9435

"So, how much faster is SLI Physics versus traditional CPU physics? We have to wait to get the driver - and supporting applications - into our hands to be able to judge the capability of the technology. Nvidia gave out some preliminary figures achieved in a demo that shows 15,000 boulders colliding with each other. The demo, aptly nicknamed "BoulderMark" was run by an SLI Physics enabled machine and GeForce 7900 GTX graphics cards with 64.5 frames per second while the CPU-only system topped out at out 6.2 frames per second."

That's great, new benchmark. BoulderMark. I think that's funny.

I actually like this the more i think about it. I don't have sli but a lot of people do already. This is like an added benefit they can already enjoy without buying new hardware.

 

http://www.beyond3d.com/forum/showthread.php?t=29299

http://www.havok.com/content/view/187/77/
Havok are targetting SM3.0 implementations because of the cross-platform benefits - PC NVIDIA/ATI, PS3 and XBOX 360. There's are greater math capability in all pixel pipelines so its actually more likely to use pixel shader than vertex shaders, with the geometry being deformed via render to vertex buffer like operations. It'll work both as a rendering pass in a frame, allowing for single graphics operation, or it can work by shuttling the ops off to a second board with the first doing the graphics rendering.

As explained by the Havok FAQ, this is not an "automatic" process accelerating current Havok implementations. This is a separate module for Havok, that is an additional cost over the rest of the Havok Middleware.

 

I think it is a good idea. I am interested how Nvidia and Havok got around the limitation of having users with single cards, running both physics and graphics code...wouldnt that result in a marked decrease in performance? Is this an SLI or Crossfire only feature?

 

There's not really that much difference between using one or two cards here. The only significant advantage for SLI would be for games that scale poorly when SLI is enabled. In both cases, if SLI scales well, you'd be dropping your resolution to run physics calculations on the GPU.

 

Chalnoth, I dont quite follow. See I was thinking if you are running physics and graphics calculations on for example the game FEAR (assuming it has Havok FX and everything implemented). Suppose the grphx card in question is running the game at 50 fps at 1024 x 768. This is the non-physics enabled card....now if I run the game on the physics enabled card, wouldnt the framerate drop rendering the game unplayable? I hope I am making sense.

 

That's how I interpreted it as well. But obviously hardware accelerated physics only makes sense if the CPU is the bottleneck, in which case you are going to have spare GPU resources.

If a game is running at 50fps at 1024x768 due to a CPU bottleneck then you might see performance at 1024x768 increase dramatically by offloading physics to the GPU. If you are already GPU bottlenecked at those settings then hardware accelerated physics will do nothing for you, whether it's an additional GPU or PPU.

 

As I said before, this isn't an automatic process, it has to be developed with the Havok FX extension in place. However, this being the case, think of this more as using the graphics processor as a large pool of math processing that can handle these type of cacluations better than the CPU is equipped to.

 

No. If the game is tuned to take up about half of the available GPU power, then instead of running the game slowly at 1024x768, you'd run it at 800x600 at about 40fps (assuming the game is entirely GPU-limited, which is not the situation in which you'd want to do this anyway).

 

Yeah I realize that Dave.

Thanks for the explanations guys.

 

Firingsquad have posted an interview with Havok FX.

 

On the one hand, what would you expect them to say?

On the other, I tend to believe that. It's also why I keep saying Creative might be a nice fit for both sides.

 

I'm not sure if it is possible but I suppose a nice split for Quad Sli would be 3 x graphics and 1 x physics so allowing a high level of graphics and taking some of the load off the cpu so you were not as cpu limited. If it works like that of course ?!

 

What?

 

Can anyone tell me how this is different from ATI's physic support with the x1xxx series please? the inq seems to think that ATI will need to play catch up to nv, but I thought ATI was there first?

http://www.theinquirer.net/?article=30434

 

Combining a discrete high-quality sound card with physics would keep Creative relevant (something they worry about as integrated sounds gets incrementally better), not add yet another card to the system, provide some economies of scale in production, marketing, and distribution, and greatly increase (tho still would take some time) the installed base that developers can rely on having the thing.

Just to be clear, I do tend to think that the most likely scenario is that the gpu boys swallow physics over time. . .but if an Ageia wants to avoid that, then Creative is the most likely looking alternative to me.