Does Physx really make a difference?

[GHarris]

The funny thing about your argument, is that you don't even mention Havok in there and it's as proprietary as PhysX is...So I'm sure you refused to play all the games that supported Havok until now, because they were using proprietary tech...

Also, PhysX is a physics API (how many times does this have to be said ????). You don't need to have a NVIDIA graphics card to run it, much less SLI, but if you do, you have the felixibility to choose where you want the physics calculations to be done. If you have graphics power to spare, do it on the GPU. If not, use the CPU or even a PPU.

"ATI cards don't" because AMD refused to use PhysX, when they were offered the chance to do it. They preferred to choose Havok instead, which is owned by Intel and, once again, is as proprietary as PhysX is.

You're never going to get a "free" Physics API. Havok isn't one and neither is PhysX, but you already have a standard to "play" with GPU resources directly. And it's called OpenCL, which both NVIDIA and AMD support. PhysX won't die out and neither will Havok (because they are physics APIs!!!!), but the flexibility of one of the over, will give it momentum and PhysX surely has that over Havok right now (deals signed with EA and 2K games for example). But the games that use it, don't come out over night. They take time to develop.

I'm sorry that my previous post was not clear enough. I am talking about hardware-accelerated physics, which is why Havok is not relevant to the points I made. The only physics API that can currently be hardware-accelerated is PhysX on NVIDIA graphics cards, yes?

As has already been said in this thread, GPUs can be far faster at physics calculations than CPUs. Which means that there are things that can be done in real time with GPU accelerated physics which just can't practically be done with a CPU. Which sets the two things apart.

Eventually, hardware-accelerated physics will either

a) become a "must-have" in games, or
b) die off.

I'm just trying to discuss the implications of both outcomes. I don't think either is particularly appealing, the way things stand at the moment.

If hardware-accelerated physics becomes a "must-have", and only one of the two big graphics card companies supports it, then pretty soon that company will have a monopoly (which will be bad news for us consumers).

If hardware-accelerated physics dies off, then that could be quite a waste of potential. Which would be a shame, but then just how much it would matter is what most of the discussion in this thread so far has been all about.

The only "ideal" outcome I can see, would be for GPU-accelerated physics calculations to run similarly well on both NVIDIA and ATI hardware. Then GPU-accelerated physics can achieve its full potential, whatever that may be. But there doesn't seem to be any realistic prospect of that happening.

I'd be interested to hear a full account of your claim that AMD refused to support NVIDIA's PhysX API. Was it a completely "free" decision? For example, was AMD forced to choose between
a) not supporting it at all or
b) starting to implement support while NVIDIA's own plans for supporting it were well advanced, so that they'd be lagging behind the speed/quality of NVIDIA's PhysX implementation while at the same time legitimizing NVIDIA's marketing claims that GPU-accelerated PhysX is important (in other words, basically endorsing their competitor's products).

Just wondering. I don't know the facts.

As for APIs that run only on the CPU, well, I don't really care who succeeds because they can run on both AMD and Intel CPUs more or less equally well.

 

[trinibwoy]

In the end, having physics calculated on the GPU means that graphics and physics are directly competing with one another..

How is that any different from the dozens of different things that directly compete with each other right now during the rendering of a frame? There isn't just one big blog of "graphics stuff" - developers make tradeoffs between different things even within the graphics space. In the end all these things contribute to what appears on screen, including physics.

 

[Elledan]

How is that any different from the dozens of different things that directly compete with each other right now during the rendering of a frame? There isn't just one big blog of "graphics stuff" - developers make tradeoffs between different things even within the graphics space. In the end all these things contribute to what appears on screen, including physics.

Indeed. With game engines written based on OpenGL, GLSL is used to program custom shader 'programs', small functions which modify the graphics data being processed by the GPU, adding effects and such. Depending on the GPU capabilities you have to scale the number and intensity of those shaders back.

CUDA/OpenCL/Brook+ are actually very much like GLSL in that they allow you write custom applications ('kernels') which run on the GPU. As an extension to those APIs there are physics APIs such as PhysX, which in essence are therefore also shader programs.

As graphics processing is identical to physics processing (both use linear algebra), there really isn't anything to differentiate a 'physics' shader from a 'graphics' shader. In the end it's all physics anyway

 

[Silus]

I'm sorry that my previous post was not clear enough. I am talking about hardware-accelerated physics, which is why Havok is not relevant to the points I made. The only physics API that can currently be hardware-accelerated is PhysX on NVIDIA graphics cards, yes?

Correct, but you said "PhysX" only. PhysX is a physics API. The GPU accelerated portion has absolutely nothing to do with PhysX itself. It's just something NVIDIA did in the drivers, to port PhysX instructions to CUDA, which is basically a bonus for all GeForce owners. They get additional use for their graphics cards, while not spending any extra cent.

Eventually, hardware-accelerated physics will either

a) become a "must-have" in games, or
b) die off.

a) It will never become "must have", because you would always be limiting your options, even if all hardware manufacturers, jumped on that bandwagon. It's not in the developers best interest to limit the "audience" of their games. But it will be something that will appeal to all users, that want to have all the eye candy they can get an all those that appreciate realistic simulations, that no doubt improve the immersion in a game.

b) It won't die off either, because GPUs are much better at these sort of calculations than CPUs. In order for CPUs to reach the efficiency and speed in these calculations, we are looking at a consumer based product, with no less than 16 cores (and they'll still be lower than powerful GPUs) and that's still quite a bit far away. When a <$100 GPU does these much faster than a $999 CPU, the advantages are quite clear...

I'd be interested to hear a full account of your claim that AMD refused to support NVIDIA's PhysX API. Was it a completely "free" decision? For example, was AMD forced to choose between
a) not supporting it at all or
b) starting to implement support while NVIDIA's own plans for supporting it were well advanced, so that they'd be lagging behind the speed/quality of NVIDIA's PhysX implementation while at the same time legitimizing NVIDIA's marketing claims that GPU-accelerated PhysX is important (in other words, basically endorsing their competitor's products).

Just wondering. I don't know the facts.

Your choice of options are "interesting". It's either AMD just decided to not support it or NVIDIA is evil and just wants to make AMD look bad...

As for links, there are plenty...Google search is your friend, but here's one:

http://www.custompc.co.uk/news/602205/nvidia-offers-physx-support-to-amd--ati.html

The title says "offers to AMD/ATI", but it's really "offers to everyone".

And this is a reply from AMD, after the efforts of a guy that said he had a way to run CUDA (and thus physX) on Radeons:

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

I think bit-tech had an article on this guy's reply too.

Despite the fact that PhysX is available for anyone and that AMD could have support for both APIs (Havok and PhysX), they refused to use PhysX or even to support this guy, in his efforts to "enable" PhysX on Radeons.

 

[Elledan]

I think what many people don't seem to realize is that we are seeing the same thing with physics as the 90s saw with graphics with the shift to hardware acceleration. No one in his right mind would want to use a software renderer at this point, and in the same vein in 5-10 years time no one would want to use a software physics library.

Then again in the early 90s quite a few users on this forum were probably still in diapers and missed most of the graphics API wars

 

[Tudz]

To get all technical and stuff, graphics is a subset of physics. Both use linear algebra to calculate vectors and matrices. In the Real World (tm) visualization ('seeing') forms only a minor fraction of the entirety of our physical experience of and interaction with our environment. The part of the EM spectrum we perceive ('light') ranges only across a few hundred nm. Physics stretches far beyond just that. In essence you could say that our eyes just allow us to see what the other 99.99% of physics affects and changes in our environment. In that respect it's somewhat crazy to consider that in computer games 99.99% is focused on visualization and the actual environmental interactions crammed in as an afterthought. Understandable considering the limitations of the medium (2D, only visual and auditory feedback, limited processing power), but still interesting.

The most realistic game will therefore be basically a big physics simulation

Soooo, basically what you're saying is graphics is physics based on the fact they are electromagnetic waves and the study of electromagnetic waves is physics? So basically all of art is physics because it comes down to sight (EM waves), sound (pressure waves) and I guess touch (lets not go there). You just end up coming to the conclusion that everything is physics and reach the point that so many people reach turning physics into philosophy

 

[Silus]

I think what many people don't seem to realize is that we are seeing the same thing with physics as the 90s saw with graphics with the shift to hardware acceleration. No one in his right mind would want to use a software renderer at this point, and in the same vein in 5-10 years time no one would want to use a software physics library.

Then again in the early 90s quite a few users on this forum were probably still in diapers and missed most of the graphics API wars

That too, but it's evident that most of the nay-sayers are nay-sayers, because they are attaching a "color" to the tech and if that "color" doesn't suit them, then the tech sucks and should die.

In all the previous discussions concerning PhysX, there are always people that:

1) Mix GPU accelerated physics with PhysX itself and so it should die, because it only runs on GeForces.
2) Claim that PhysX is proprietary and should die for that, conveniently forgetting that it's as proprietary as Havok
3) Claim that almost no one is using PhysX right now, so it means it's not worth it, again conveniently forgetting that games take years to develop.
4) Say It's only eye candy and it doesn't influence gameplay, also forgetting that this is not supposed to influence gameplay right now. Except one or two games (that had the advantage of being the first to use physicalized environments), there are no games that use physics effects except for eye candy purposes or simply the same types of gameplay impacts that the first games that ever used physics API, had.
Physics calculated on the GPU are not going to change that right away. What they change is the ability of developers to create much more realistic physics simulations, that could never have been done on a CPU.

All these misconceptions make no sense, because there's plenty of information about it on the web, to read and actually learn about. But some people prefer to ask in forums, where these "color attached" fans exist and will stop at nothing to spin and twist easily accessible information about the benefits / problems that a certain technology has.

 

[trinibwoy]

Soooo, basically what you're saying is graphics is physics based on the fact they are electromagnetic waves and the study of electromagnetic waves is physics? So basically all of art is physics because it comes down to sight (EM waves), sound (pressure waves) and I guess touch (lets not go there). You just end up coming to the conclusion that everything is physics and reach the point that so many people reach turning physics into philosophy

"Graphics rendering" is the simulation of what your eyes see when looking at a scene. Just because rasterization doesn't directly simulate light transport doesn't mean there aren't any other methods that do. Raytracing and photon mapping literally simulate shooting rays of light around. How is that not physics?

Optics is regarded within theoretical physics as a subfield of electromagnetism

 

[Tudz]

"Graphics rendering" is the simulation of what your eyes see when looking at a scene. Just because rasterization doesn't directly simulate light transport doesn't mean there aren't any other methods that do. Raytracing and photon mapping literally simulate shooting rays of light around. How is that not physics?

I never said it wasn't I said if you want to get technical about it, everything is physics. I move my mouse, it has friction... physics... movements picked up by an LED and I'm guessing a photodiode... physics... signal passed through electrons in wires... physics... wire made from a metal (metallic lattice) coated in a polymer... physics... I could trace this all the way to the cursor moving on the screen and the lens in your eye projecting the image onto your retina and your brain picking up the signals, but you get the idea You look at anything close enough and it becomes physics.

Saying graphics is physics is about as significant as saying toothpaste is physics. But I dont want to get into an argument about it, my original post was simply reflecting the fact that I thought the post I quoted was an odd statement.

 

[trinibwoy]

Saying graphics is physics is about as significant as saying toothpaste is physics.

We're talking about what happens behind the scenes to generate the image that shows up on your screen when playing a game. That's a pretty focused topic. Not sure how a comparison to toothpaste is relevant.

 

[Tudz]

We're talking about what happens behind the scenes to generate the image that shows up on your screen when playing a game. That's a pretty focused topic. Not sure how a comparison to toothpaste is relevant.

I never said it was as "relevant" to say toothpaste is physics, I said its as "significant" to say toothpaste is physics.

 

[Valset]

That too, but it's evident that most of the nay-sayers are nay-sayers, because they are attaching a "color" to the tech and if that "color" doesn't suit them, then the tech sucks and should die.

In all the previous discussions concerning PhysX, there are always people that:

1) Mix GPU accelerated physics with PhysX itself and so it should die, because it only runs on GeForces.
2) Claim that PhysX is proprietary and should die for that, conveniently forgetting that it's as proprietary as Havok
3) Claim that almost no one is using PhysX right now, so it means it's not worth it, again conveniently forgetting that games take years to develop.
4) Say It's only eye candy and it doesn't influence gameplay, also forgetting that this is not supposed to influence gameplay right now. Except one or two games (that had the advantage of being the first to use physicalized environments), there are no games that use physics effects except for eye candy purposes or simply the same types of gameplay impacts that the first games that ever used physics API, had.
Physics calculated on the GPU are not going to change that right away. What they change is the ability of developers to create much more realistic physics simulations, that could never have been done on a CPU.

All these misconceptions make no sense, because there's plenty of information about it on the web, to read and actually learn about. But some people prefer to ask in forums, where these "color attached" fans exist and will stop at nothing to spin and twist easily accessible information about the benefits / problems that a certain technology has.

Those all valid points (and well said) but I would add this: I honestly don't think its worth doing on the GPU because there is so much more the GPU can do that does affect game play. (I am aware that this is subjective, its JMO) I will use the new batman demo for example (wow, talk about a juvenile plot, its hard to play it even with the eye candy). I can't run it maxed out on my system, given the choice between good frame rates with AA and physx, well it was no contest. I just don't see this changing any time soon. It not that the physics is not worth pursuing it just that there are better things to allocate your resources.

the other point I would add is the chicken and the egg problem is killing physx faster then anything, people aren't buying in to physx because nothing meaningful has come of it yet, and nothing meaningful is coming because people aren't buying it yet. I have been a green fan for a long time (last 4 cards in fact) but every time I hear about physx I tend to roll my eyes. If they are that desperate to get this tech going then they are going to have to get something produced that warrants it otherwise it will continue like it is, with little or nothing differentiating red from green beyond the bottom dollar.