I don't understand the point of this PPU stuff... Games like Half-Life 2 have great physics-driven enviroments without a PPU...

Here's a list of games using PhysX in some form.
http://www.nzone.com/object/nzone_physxgames_home.html

I think a lot of people saying it's a niche product or that no one will use it. No nothing about it. I think there are more than a dozen game studios on board now as well. Including the recent announcement from EA/2K.

Nvidia said that they only had to do small changes to cuda to make it OpenCL compatible I heard, and physX sits on top of the same foundation now I understand? And several more companies signed up for physx recently didn't they? So much for 'talking physx down' then eh.

As for stream computing, if you put the dll's with the driver then you should make a driver that people can install without random BSOD's every 4 minutes, else people are forced to us 6 months old drivers and find that when they download the stream SDK as an alternative it comes without pre-compiled dll's so that won't work either.. Also you should make your applications work on 64bit OS's since it's 2008 not 1998.
But let's just try to 'talk' competition down eh instead of doing a halfway decent job.

If you're running a game in a scenario where you're GPU limited... how is hoisting physics computations on to the GPU going to speed it up? Aren't you already stressing the GPU?

GPU-physics for gaming is already doomed. It is nothing more than a marketing pitch to justify expensive and hopelessly overpowered GPU solutions.

Mutli-core CPUs is where gaming physics should go. Quad-core are quite affordable these days and soon enough Sextet and Octal-core chips.

PhysX missed the boat and will die quicker than GLIDE.

It is bigger shame that none of the popular gaming genres are suitable for harder physics.

It's kind of interesting to me that AMD is essentially trying to kill the chicken in this "chicken and egg" scenario. Game developers won't use PhysX for PC games in a way that affects gameplay because only NVIDIA cards support PhysX. AMD is gambling that if they refuse to support PhysX, no developers will use it.* They may be right, seeing as DirectX Compute and OpenCL (maybe Havok?) will provide viable alternatives.

Still, I can see a possibility that things slide in NVIDIA's favor in the future. Some developers may write levels or maps that take advantage of PhysX, and some developers may write a PS3/Xbox 360 game that uses the PhysX library, and then figure that a PC port that only runs well on computers using an NVIDIA card is better than no port at all. PhysX does run on a CPU, they could assume that by the time their game comes out for PC that a high-end CPU would be enough. Engine writers could decide to write PhysX pathways, meaning that they would use PhysX on some systems and Havoc on others. id and Epic write engines that work on multiple platforms using multiple graphics libraries, so it's not too hard to imagine. Lastly NVIDIA could just give a company enough money that they decide to use PhysX.

I'm surprised that no one in this discussion has brought up this up. I see people saying 'PhysX sucks' or 'PhysX is great, and AMD has no future' or 'PhysX could be great but no developer is willing to use it' but no one is discussing that AMD is trying to kill a library. On that issue, I feel a little torn. As a consumer and programmer, I like choice. If PhysX is better, then games should use it. On the other hand, as a consumer and programmer I also like standards to be open, free, and widely supported. If PhysX won't be free, open, or widely deployed then maybe it should die. Nothing is stopping NVIDIA from writing a PhysX library for AMD cards using Brook+ (or Open-GL), and the fact that they aren't suggest to me that they're going to try some sort of lock-in. A PhysX implementation for AMD cards (even a slow one) provided by NVIDIA would probably make the engine much more palatable for developers. The fact that they're putting their money towards developers, bribing (edit: sorry, misread a different interview!) them to use PhysX, instead of towards making an open PhysX implementation is worrisome to me.

*Anyone can feel free to suggest that people with AMD graphics cards should buy Aegia's add-in card, but if they do they shouldn't hold their breath waiting for a response.

AMD has gone on record saying it sees no future for the game physics API.

LOL what a crock of sh**

Maybe no future for them.

Ppl get caught up in the "hardware acceleration" and API bits of Phsyx and forget that it also has a pretty large physics library.

DX11 Compute will enable developers to access hardware, but do you really want to reinvent a physics engine every time you make a game?

However, ati is right in that proprietary* implementations tend to die out - nvidia should know better than anyone, having gobbled up 3Dfx after it went bust.

DX11 is due out in Feb. Mirror's Edge may or may not be out by then. Like I said before, it's a race.

* Physx runs on CPUs, but so slowly compared to a GPU that you're better off turning it off.

Cyril,
I have been wondering, since you've done this in quite a few articles now...why do you, when referring to something NVIDIA said, say "boast" ? Isn't it the exact same thing AMD is doing too ? Why the difference in words, in so many articles ?

In regards to AMD's comments, they're pretty funny. So because PhysX is proprietary, it shouldn't be used, but Havok, which is also proprietary, should...

This is obvious PR talk. They have nothing to compete with NVIDIA in this front and so they just talk it down. It's the typical move from any company in this business.

Yes and no about "eye-candy". I mean, when it boils down, isn't it really just about eye candy anyway? I mean, I can play UT3 quite comfortably on a Radeon HD 2400 pro quite easily with everything turned all the way down. (I know for a fact). The game is still fun, but doesn't look as nice as my desktop with a 8800GTX and all the settings maxed. I'm not sure I get any less "fun" out of the 2400 except that I can't play the PhysX only maps.

I wholeheartedly agree with AMD. If nVidia is incapable of lifting PhysX-exclusive things from their "eye candy" status up to "extra added interactivity, challenge or gameplay content" status, then PhysX will plain simply die. That's not to say it's inherently useless, but they're not using it to make games, they're making tech demo elements on top of games. That's where this is all going wrong.

You're not going to get physics effects that impact gameplay until the vast majority of hardware is capable of producing those effects with reasonable performance. The fact is that there are certain effects that require more horsepower than what's available from CPUs today. So you can either appreciate them for what they are or bitch and moan for a few years till CPUs catch up.

Whether or not one IHV is pushing it as a marketing ploy it's hard to argue that more eye candy isn't a welcome. We obsess over DX10 particle systems for more realistic fire, smoke and water but now those same things are irrelevant and unnecessary when it comes to hardware accelerated physics? Once the politics around PhysX support goes away hopefully developers get on board and show us something impressive. But whether PhysX goes opencl or some other standard emerges it will happen eventually.

I think a lot people are confused about what PhysX is. PhysX is a library of physics calculation related algorithms and functionality. It is not inherently tied to any particular hardware and can be implemented in various forms.

That should be obvious since it runs on Ageia's PPU, Sony's Cell, Intel's x86 and Nvidia GPU's via CUDA. There is no reason why the library can't be ported to OpenCL. It should be a very easy port considering the similiarity between CUDA and OpenCL.

People are confusing PhysX the library with the hardware that it runs on. What makes it worse is that people are talking about Havok like it's more open than PhysX. It's not.

I'm not quite sure if I'm making any point with this (sry, it was a tough day...) but game physics in recent titles are sort of what was happening back than when dx9 rolled out...particularly the version c, i believe. I can clearly remember I read somewhere that most of the tricks could have been done with previous versions as well (even with 8.x), but either the programmers were lazy or they didn't have the resources to develop those things that we first saw in games with dx9.0c support. What I mean here: it doesn't matter what the GPU or the API is capable of, if the big shot studios just don't want to utilize it in full. And nV and AMD both help this tendency by fighting over sale rates - they won't give full support to something what's not their own. The enduser is the least important. Unlike the share holders.

Personally, I'm not interested in cloth pieces flying around in the air, neither in disappearing debris. I want to see real body weighting, momentum, truely interacting pieces of environment. Etc.
But neither nVidia, nor AMD cares about my opinion.

...:
http://www.tgdaily.com/content/view/38392/118/

It was reported about 6 months ago nVidia offered to license PhysX to ALL 3rd-parties and also offered to license the technology for the drivers for non-nVidia GPUs as well. Well, the first part of the last sentence has now come to fruition with the licenses to EA and Take Two. At the same time, I read that there was direct communication from nVidia to AMD offerring to license PhysX, and that AMD/ATi (hopefully) politely said no, since they had just got into bed with Havok/Intel. So, it seems that the non-PhysX "high-road" that ATi is taking in not integrating PhysX functionality into their GPU drivers is entirely of their own making, with business-politics no doubt playing its usual "not-invented-here" role. They could license PhysX tomorrow if they wanted to, as there is nothing in the ATi GPU architectures that prevents effective driver implementation.

Hopefully, the refusal to license PhysX does not damage ATi's medium-term profitability; it is going to be a long while until a more universal GPU hardware-accelerated physics alternate is available, presumably within the Dx 11++ family. It is pretty obvious from the fact that EA and Take Two have both licensed PhysX already, that their expectation of something more universal for game-related hardware-accelerated physics in the medium term is pretty low.

As for running the PhysX-enhancement of PC future games on desktop CPU cores when lacking the GPU driver support, there will be massive computational hits to these games' "frame-rates" when PhysX is enabled. Certain types of computations are appropriate to current multi-core desktop processors and massively-parallel particle-physics algorithms are distinctly not in that group.

Hmm, isn't DirectX a closed and proprietary standard?