AMD finally gets back to Physics on the GPU

But GRAW did just that.
Physics on the PPU, with debris damage...not on the CPU...if you enabled PhysX.

GRAW 2 did it with a GPU.

But it's seems some here claim that is not possible...I guess that dosn't make me the fool.

 

not in the case of graw it just didnt do the enhanced physics if no ppu or nv gpu was present

 

Yes, your amazing graphics improvement with PhysX in GRAW

http://forums.overclockers.co.uk/showthread.php?t=17568825

http://www.anandtech.com/show/2001/5

And that was one of the more kind reviews of the card. Anandtech are a lot more harsh in their update and followup.

Also, PhysX in GRAW didn't do anything differently than CPU physics. Just added more graphics particles which had no impact on gameplay.

Additionally

http://www.firingsquad.com/features/ageia_physx_response/page2.asp

Ummm, yes, physics really is done on CPU in GRAW. And with Havok to boot. Ageia didn't dispute that, and agreed with it.

But at least unlike GPU PhysX

Note how they deliberately avoid mentioning the player in any of this. Anything that affects the player is modeled in Havok. So yes, Ageia PPU can do stuff that GPU PhysX cannot do, thank you for pointing that out.

Regards,
SB

 

hasn't rigid body physics just been added to physx ?

 

http://www.thefreelibrary.com/AGEIA...ic+Cross-Platform+Physics+Game...-a0143483949

2006...

 

rigid-body simulation was available on the PPU, it hasn't been done on the GPU until now (Fermi and the latest SDK) before that it ran on the CPU.

But I don't think this topic was about what nVidia can and can't do with its acquired library.

 

Ah, good info to have, do you know if Rigid body physics is limited to Fermi or if it's also being ported down the line to pre-fermi hardware. It wouldn't be all that surprising if some of the architechtural changes in Fermi are what is allowing it. And yeah, probably wrong thread for this.

Regards,
SB

 

After the acquiring of Ageia, the lead engineer said in an interview that rigid body simulation would never be done on the GPU so I was actually surprised to see it announced in the latest SDK for Fermi. Trying to calculate the posibility of rigid body simulation on G8x/G9x/GT2xx turned up a "Divide by Zero" error on my calculator.

I think the whole point of the rocket sled demo was "See, we can do rigid body simulation too"(now, asteriks)

 

So there's a possibility of PPU level physics, Pre-Fermi level Physics, and Fermi level physics in PhysX? With Fermi and PPU appearing at first glance to be pretty similar.

Makes me wonder if companies using PhysX will now have to have a code path for CPU, GPU (pre-fermi) and GPU (post fermi). Or possibly ignore rigid body in order to support as many cards as possible.

Regards,
SB

 

nvidia were quick to drop support for the PPU, wouldn't be surprised if that went for pre-GF100 physics as well. I can already see how they'll proclaim that it's up to the developers to support anything other than nvidia's prefered physics paltform.

 

Actually, they continued new software releases for the PPU hardware for 3 more years after buying Ageia. Even though the hardware was never all that capable to begin with. But yeah, you would see it like that.

 

How's the latency if they're going to do all physics on the GPU now? GPU pipelines are vastly deeper than CPUs after all, and it's an external chip you need to communicate with across buses by DMAing packets back and forth etc... Does it take an appreciably large portion of frametime before you get physics results back from the GPU?

Perhaps too early to tell, considering fermi boards are hardly available for sale still...

 

Ofcourse they don't officially stop supporting it, but like multi-core CPU PhysX, they certainly aren't going to entice using PPU PhysX. You're really getting the hang of NV's marketing Quack.

I'd love to see benchmarks on B:AA for instance with GPU or PPU PhysX, before I would claim the PPU wasn't that powerful.

 

Here I thought this thread was resurrected with good news of AMD actually doing some GPU physics calculations. I should have realized it would just be another ridiculous replay of the Nvidia PhysX whining.

 

I don't know about the latest announcement on the PhysX SDK but the rocket sled demo still does the rigid body on the CPU, that was confirmed by a developer on the demo GDC 2010 video.

 

Blame AMD...nothing to show.

 

You can thank Lonbjerg for the thread resurrection with some odd claims and then proceeding to exacerbate it with claiming no physics calculations done on CPU in GRAW.

It certainly wasn't the PhysX sceptics that did it.

Regards,
SB

 

Ah you are right SB. Weeks after the prior post the non-sequiter flies in.

Still no need to take the bait. Does anyone happen to know if there is any progress on physics form the AMD side though? OpenCl based bullet or whatever the consensus was last time? Is AMD working on this stuff or not? I heard they were advising or something before, but nothing recent.

 

According to AMD, there are two games planned for the holiday season that will use GPU Physics through Bullet/DMM. I can only assume that one of them is the recently announced Star Wars, The Force Unleashed 2.
It seems AMD is paying Pixelux to provide the Bullet SDK to developers and nVidia is also supporting Bullet GPU physics, or at least have demoed it back at the first Fermi launch.

It also seems AMD have written the OpenCL layer with Havok to run on the GPU (As seen last year), but intel isn't allowing its release(yet.)