Agiea PhysX Shennannigans

[seanp789]

I for one have been drooling at the mouth for nearly a year at the mention of a dedicated PPU. So what do we get with hardware physics simulation?

http://www.ageia.com/products/release_notes.html
This is the current version of Agiea's physX drivers. If you take the time to read it you will notice some misleading contradictions to http://physx.ageia.com/



+fluids (The maximum number of capsules, boxes, and are limited to 1024 in a scene.
+rigid body (There is a limit of 2048 D6 joints per scene, joint limits must be symetrical)
-soft body (nope)
-cloth (software only, no hardware support)


Algeia and Nvidia both bragged about their "boulder scenes" 50,000 and 15,000 respectively. This seems to have little bearing on what we can hope for from games. When a video cards states it can process 10M triangles per second that really offers so little information since there are so many other things a video card must do. Similarly, PPUs "boulders" seem to be the most basic performance measure when nothing else is running. Throwing in joints, fluids, etc... is much like turning on shaders, AA, AF, HDR all which drop overall performance.

I think the first generation of physics is pretty much going to be a one trick pony and not half of what these companies are claiming. So far no Vista or 64-bit support is is in the works but that can certainly change. The good news is Agiea continues to fix and update their drivers on a regular basis. No waiting 2 years for a driver fix (That means you Creative!).

As far as Nvidia and ATI the information is still very limited with no currently working solution. Nvidia seems to be at the largest disadvantage so far according to a few website articles.

My prediction is that Agiea will dominate the market early as they have many promising titles already lined up. The market will eventually move to a common API, which may take a few years, and Agiea will either come up with something incredible to stay in the propietary game or die the way 3dfx did only to have its remains sucked up by nvidia or ATI.

I would not be surprised to see future games supporting multiple physics renderers until one gains a dominating foothold. The promise of fully destructable environments is still a long way off, just more wooden crates and bright barrels.


That said I cant wait to see Ghost Recon PC in action. Agiea has set its release date for May same as Ghsot Recon. I can only guess that the launch delay has something to with the extra PPU action. Agiea must finally have something worth showing off because those cards have been done for months now.


Ill buy one if only to see it for myself but my expectation have been lowered.

 

[Velk]

Originally posted by: seanp789
I for one have been drooling at the mouth for nearly a year at the mention of a dedicated PPU. So what do we get with hardware physics simulation?

http://www.ageia.com/products/release_notes.html
This is the current version of Agiea's physX drivers. If you take the time to read it you will notice some misleading contradictions to http://physx.ageia.com/

+fluids (The maximum number of capsules, boxes, and are limited to 1024 in a scene.
+rigid body (There is a limit of 2048 D6 joints per scene, joint limits must be symetrical)
-soft body (nope)
-cloth (software only, no hardware support)


Algeia and Nvidia both bragged about their "boulder scenes" 50,000 and 15,000 respectively.

With no offense meant, you really can't make any predictions on things until you know what they are talking about. For example, the limit of 2048 D6 joints is just that - a limit on joints, it has nothing whatsoever to do with boulders and how many of them it can calculate paths for.

That's not to say that the result won't be underwhelming, but the 'evidence' you provide suggests nothing either way (Except, of course, that they are still adding major features to the SDK now so the chances of anything using them in a half decent fashion for a may release date is pretty much zero ).

 

[LOUISSSSS]

^^ well said. i will be waiting until this card actually comes into mainstream market to do further research, everything u read now is preliminary and many things will change from now till the day its availible

 

[seanp789]

Yes its performance is indeed all speculation. If you look at the limitations again:

boulders 50,000
joints 2,048
fluids 1,024

I think without a PHD in physics and engineering you can agree that the document implies it is more processing intensive to process fuilds as opposed to boulders. My point was that a number like 50,000 boulders should not be relied upon as a performance measure because their are far more factors in a real game that will affect performance.

For example, lets you were using 500 fluids, maybe that would reduce your boulder to only 25,000 boulders. Add 1,000 joints and now you may be down to only 12,500 boulders. These exact number are not accurate but the point stands that advanced effects will severely drop the overall performance of the PPU.

I used a analogy to the video card market that boulders are about as useful as a videocard stating it can do 10M triangles per second. Hence the marketing up to this point has been misleading on what we can actually expect in a game and a side note that cloth is infact not hardware accelerated.


My thoughts were largely based on what consumers could expect at launch and which for me may be a little bit of a let down. Of course I would expect them to improve over time but you may end up having to wait a year to see something just added to an SDK or even wait for a second generation of PPU.

I will of course be proven right or wrong after these cards go retail in 2 months. Its still a safe bet that there has been some serious exageration on what to expect at launch and in the near future.

Originally posted by: Velk

With no offense meant, you really can't make any predictions on things until you know what they are talking about. For example, the limit of 2048 D6 joints is just that - a limit on joints, it has nothing whatsoever to do with boulders and how many of them it can calculate paths for.

That's not to say that the result won't be underwhelming, but the 'evidence' you provide suggests nothing either way (Except, of course, that they are still adding major features to the SDK now so the chances of anything using them in a half decent fashion for a may release date is pretty much zero ).

 

[Childs]

Well, since this is hardware this may not be an either/or situation. Its possible that different parts of the ppu are responsible for different effects, so number of boulders and number of fluids have nothing do to with each other. A general purpose cpu would be effected by it, but the functions for fluids vs boulders wouldnt overlap...at least I wouldnt think so. Its been 10 years since I opened a physics text book, and I have no idea how the ppu is designed.