What exactly is PhysX?

[anroin]

I really have been living under a rock,so enlighten me.I always thought it was a independent software that somehow "enhanced" the physics of a game if the game has support for it,but is it only existent by Nvidia cards? Can one not benefit from PhysX if using an ATI card?

 

[kingnoobe]

You can if you have a nvida card also, but ati doesn't use physx. Although last I heard even that has changed as I guess something along the line nvidia won't all the use of it if it detects an ati/different card. That of course could've changed or their could also be simple work arounds not to sure.

 

[reccy]

http://en.wikipedia.org/wiki/PhysX

 

[anroin]

Hmm,thanks for that.
Than,why would one want an ATI card if it's missing some technology? Or should I ask,can PhysX change things in a game that much?

 

[reccy]

I believe Ati cards work with Physx, youtube/google it, swear i seen an Ati working with the physx process.

Plus Physx is a nice graphical improvement, just not huge or necessary in gameplay. What it adds? youtube it mate - not sure myself, ive just turn the thing on and hoped for the best lol

 

[outlw6669]

PhysX is a nVidia propitiatory physics engine (similar to Havok, Bullet and a few others).

It is currently implemented one of two ways, to be processed on the CPU or to be processed on a GPU (historically it could also use an Ageia PPU).
If it is implemented to run on the CPU (as it is in the vast majority of PhysX titles, the Xbox 360, PS3 and Wii) it can be used with virtually all hardware configurations.

If it is implemented to run on the GPU (GPU PhysX) you are required to have a GPU that supports CUDA.
Unfortunately, nVidia does not allow its competition (mainly ATI) to run either CUDA or PhysX on their hardware (it has been hacked to run on ATI hardware before so it is technically possible).
Nvidia also artificially restricts your choices in GPUs by not allowing you to run a dedicated nVidia GPU for PhysX if any non-nVidia GPU is active in the system.

Due mainly to nVidia's desire to keep GPU PhysX 100% propitiatory to their hardware, it is rarely implemented.
When it is used, only minor aspects of the game are modified (fluttering drapes or flags, small amounts of debris, etc.).
If anything major where to be implemented, developers would be killing their customer base (lack of universal support) and profits.

In short, GPU PhysX should not be a major factor in deciding which GPU is best for you.

 

[gamerk316]

Here we go again...

If it is implemented to run on the GPU (GPU PhysX) you are required to have a GPU that supports CUDA.
Unfortunately, nVidia does not allow its competition (mainly ATI) to run either CUDA or PhysX on their hardware (it has been hacked to run on ATI hardware before so it is technically possible).

CUDA is NVIDIA's implementation of the API; there is no technical reason it couldn't be ported to ATI Steam, DirectCompute, or even OpenGL's programmable shader language. Now, whether NVIDIA will allow this to happen, or if ATI simply refuses to do so, is another debate all together.

Nvidia also artificially restricts your choices in GPUs by not allowing you to run a dedicated nVidia GPU for PhysX if any non-nVidia GPU is active in the system.

As explained before: Why should NVIDIA be responsable for a ATI/NVIDIA combination? Why should ATI benifit from NVIDIA's technology?

Due mainly to nVidia's desire to keep GPU PhysX 100% propitiatory to their hardware, it is rarely implemented.

Yep, its only used more then Havok is; that is to say, more then any other third party physics engine. The CPU portion of the API si doing quite well, and even the accelerated portions are seeing better use in new titles.

 

[Griffolion]

I must agree it is Nvidia's right to keep their own technology proprietary to themselves, they are in business to make money and are certainly not expected to be charitable to the rest of the industry by making Physx more open and thus taking away one of their edges in the market. The only problem with Physx is that its not so massive as you would include it in your pros/cons list when making a decision on graphics cards.

 

[iam2thecrowe]

if ATI made a GPU physx solution, and enabled it to run on any GPU, physx would be history.....Please do this ATI and knock nvidia off their high horse! Nvidia are no better than Apple by doing this proprietry crap!

 

[Griffolion]

if ATI made a GPU physx solution, and enabled it to run on any GPU, physx would be history.....Please do this ATI and knock nvidia off their high horse! Nvidia are no better than Apple by doing this proprietry crap!

Like i said, Nvidia have the right to keep THEIR technology proprietary, the only difference is what type of stance you take to it. But i agree, if ATI made a physics API that was open and matched/surpassd Physx in what it did then Physx would soon go out of the window. Open > Proprietary, proven time and time again. But still considering this, it is always Nvidia's right to keep their tech prorietary if they want to.

 

[jedimasterben]

Like i said, Nvidia have the right to keep THEIR technology proprietary, the only difference is what type of stance you take to it. But i agree, if ATI made a physics API that was open and matched/surpassd Physx in what it did then Physx would soon go out of the window. Open > Proprietary, proven time and time again. But still considering this, it is always Nvidia's right to keep their tech prorietary if they want to.

Last I heard, AMD is working on accelerating the Havok API with its GPUs using OpenCL, so it would be able to run on any GPU that had OpenCL enabled.

In nVidia were to optimize/rewrite the code for the API (IE, making run x86 code instead of x87), then I don't think that it would need GPU acceleration to reach the same effect. I don't know that much about coding, so I don't quite know how big of an undertaking it would be to do that, so it must be the reason that they haven't done anything like that yet.

 

[outlw6669]

CUDA is NVIDIA's implementation of the API; there is no technical reason it couldn't be ported to ATI Steam, DirectCompute, or even OpenGL's programmable shader language. Now, whether NVIDIA will allow this to happen, or if ATI simply refuses to do so, is another debate all together.

Yeah, I have heard it both ways.
Nvidia insisting that they would allow others to support it, ATI claiming that nVidia will not allow them to implement it on their hardware.
Without being in the industry, I could not say with 100% certainty which argument is correct, probably it is in the middle somewhere.
Based on nVidia's business practices and several older editorials about this however, I am heavily leaning towards nVidia working to block ATI.

The only thing I do know for sure is that GPU PhysX can not (officially) be ran on anything other than nVidia hardware.
Blame aside, limiting the install base of GPU PhysX capable cards is a major minus for PhysX.

As explained before: Why should NVIDIA be responsable for a ATI/NVIDIA combination? Why should ATI benifit from NVIDIA's technology?

Why?
For a multitude of reasons.

Because it artificially not supported (remember the 'Bug' that allowed a nVidia PhysX GPU and ATI primary to work together?).
Because systems running older or hacked drivers do it just fine.
Because Windows 7 allows multiple display drivers to work together just fine.
Because they could put a disclaimer with the driver 'may not work properly with an ATI GPU installed' and still grow their install base.
Because every PhysX card sold is money in nVidia's coffer and a larger install base of GPU PhysX capable systems.
Because every GPU PhysX capable system out there increases the odds that a developer will make GPU PhysX a major part of their game, driving nVidia GPU sales.
Etc, Etc.....

Also, I really fail to see how ATI is benefiting from nVidia's tech?
ATI would not sell any extra cards with PhysX/CUDA support, nVidia would.
No matter how I look at it, more sales for nVidia is good for nVidia.

The only potential down side I can find is if someone is looking to upgrade an older nVidia card (soon to be a dedicated PhysX card) and purchases ATI.
In a case like that, it would be a clear indicator that nVidia needs to work to make their cards more competitive and are probably loosing sales across the board.

Yep, its only used more then Havok is; that is to say, more then any other third party physics engine. The CPU portion of the API si doing quite well, and even the accelerated portions are seeing better use in new titles.

I was specifically refering to the GPU PhysX implementation of PhysX which is a quite Short List.
The CPU implementation has quite a few More Titles available (comparable, or a few more than Havok), but also does not need a nVidia GPU to work, as it the point of this thread.

 

[Mousemonkey]

I can't believe this is still going, basically ATi do not want to pay Nvidia for a licence to use PhysX and Nvidia don't want to give it to them for free because they keep slagging it off whilst promising that they are going to come up with something that is much better, a promise they have yet to deliver on after some four years or so.

 

[reccy]

I can't believe this is still going, basically ATi do not want to pay Nvidia for a licence to use PhysX and Nvidia don't want to give it to them for free because they keep slagging it off whilst promising that they are going to come up with something that is much better, a promise they have yet to deliver on after some four years or so.

Hense why i hate Ati.. Bunch of Bum Holes..

 

[rofl_my_waffle]

Physx isn't popular. Nvidia just makes it sound like their physics engine is the ONLY engine for physics in the entire world which is absolutely not true.

Theres like 13 titles according to Nvidia that actually support Physx.

CPU based physics Havok claims a mind numbing 250+ titles. Thats right, almost every game you play already has some type of physics engine, probably Havok but nobody advertises it like Nvidia does.
Havok is just as powerful perhaps even more so in certain aspects. Not to mention your CPU can run physics calculations several times better than a GPU and most computers these days have extra cores that are just idling.

Physx is a bunch of marketing garbage and not anything useful. A quick way to sell people dedicated physx cards, and you would need dedicated physx cards because their cards can't even run their own engine efficiently. They purposely used x87 to do math instead of SSE which would have been faster in every single case, they also limited CPUs to a single processing thread. Consoles on the other hand don't have physx enabled GPUs but they can do it with only their CPUs because it is optimised for it, and an Xbox barely has the power of a P4.

This is why Havok is a dozen times more popular. Not to mention Physx will never work for multiplayer because you just can't have only Nvidia users to play the game, thats why games like Bad Company 2 uses Havok as well. Physx isn't catching on and never will. But Nvidia can try very hard to make you think it matters.

 

[gamerk316]

Theres like 13 titles according to Nvidia that actually support Physx.

Get it right; only a handful of titles use GPU accelerated PhysX for the more advanced PhysX effects; but in 2009/2010, more games use CPU PhysX then CPU Havok; comparing CPU Havok to GPU PhysX is silly.

to mention your CPU can run physics calculations several times better than a GPU

Wrong; Multi-object dynamic physical effects can not be run on current CPU's at any decent speed; thats why no game has event attempted such yet. [Red Faction 3 used a VERY powerful scripting engine. The only game that can even claim to use dynamic phyics is Backbreaker, which coincidentally, is Eurphoria wrapped around a PhysX layer].

Todays implementations of one-object interactions can run fine, as can most linear (simple) physics solutions, but full scale dynamic physics would bog down any CPU in existence.

 

[gamerk316]

Because systems running older or hacked drivers do it just fine.

So therefore, there is no possibility of there EVER being an incompatability between NVIDIA and ATI's drivers. Where I work, Software engineers get fired for those types of assumptions. In such a situation, it becomes NVIDIA's problem to ensure compatability with ATI's drivers, while ATI has to do nothing.

Also, I really fail to see how ATI is benefiting from nVidia's tech?
ATI would not sell any extra cards with PhysX/CUDA support, nVidia would.

How many people wanted to get an ATI card as their primary, with a NVIDIA card for PhysX? A lot; lets face it, for people who already have a 8800GTX, wouldn't it make more sense for them to get a 5xxx series and keep the 8800 for PhysX? Hence, ATI benifits.

 

[Glacier]

Hmm,thanks for that.
Than,why would one want an ATI card if it's missing some technology? Or should I ask,can PhysX change things in a game that much?

Like the other poster said, only a few games actually use PhysX. PhyX makes those few games look better, but it's not enough to ignore ATI. Hardcore gamers might want PhysX, but we mainstream gamers couldn't care less about it. We want good cards that play games well. ATI makes very good cards at a reasonable price. The 5000 series owned the market for a long time, and ATI will soon re-take the market with the 6000 series.