Physx - Are you interested in it? Have your say! VOTE!

[cmdrdredd]

Unlike NVIDIA PhysX, you dont need specific hardware to run Havoc, every CPU can use it.

Same with every game engine physics, every Hardware being CPU or GPU can process it, you dont need specific hardware to run Frostbite 2/3 game Physics. Both NV and AMD GPUs have the same access to the Engines Physics.

The real question is if the potential quality for real time effects is the same. Havoc is always a joke when I have seen it in a game. The ragdoll physics for example.

 

[MrK6]

You're serious? You'll surely excuse me if I can only take this as a joke.
Good one. :thumbsup:

I still don't see you trying to refute any of my points. Every attempt at deflection or hiding from answering the rebuttals of your arguments just further proves my points.

1) Consoles still can display a couple of particles and sparks and basic (2D) fog
2) And still it is mostly PhysX games where these effects are shown in abundance
3) I never said the barrier was of a technical nature. It doesn't change anything - the effects aren't in the base game, that's a fact. Therefore nothing was removed, it was never there to begin with.
4) With that attitude you can fabricate theories for everything and claim they are the truth.

What you say is still not true, there is no decrease in quality. Or do you have access to a different timeline where PhysX doesn't exist? Do you know the developers personally and know their reasoning? And what other games do you speak of? Give some examples.

I'll give you cloth simulation since I've seen that feature in enough games already. But extensive particles effects are (while certainly possible) not the norm and occur only in a tiny amount of games if the occur at all. Crysis 3 has some cool spark effects, that Stalker game had dynamic fog, but apart from that, that's it. I cannot see a trend here.

I'm not sure what you're not understanding, PrincessFrosty completely disproved your previous points. Most of the effects that PhysX does have been rendered by a non-GPU accelerated method previously. Games with PhysX decrease quality because PhysX scales poorly and is poorly optimized for CPU's (obviously because nvidia doesn't make CPU's and wants to use it to sell GPU's). No developer that is being paid to use PhysX is going to waste more time and money to re-code the same effects with a different method. I'm guessing since current consoles are old, taking out the effects probably saves resources, however this approach might change with the expectations of new consoles. In either case, I'm not quite sure why that is so hard to follow.

 

[PrincessFrosty]

What you say is still not true, there is no decrease in quality. Or do you have access to a different timeline where PhysX doesn't exist?

No, I have access to this timeline where I have countless examples of similar games in a similar class that offer these basic features to everyone equally and aren't locked to GPU-PhysX only. The gaming industry forms a strong trend over time which the PhysX games lay outside of in this regard, that requires some explanation.

PhysX being the cause is the most reasonable explanation for the discrepancy, if anyone has a more plausible explanation then speak up. This is real life, we don't have 100% proof of everything, I'm not claiming absolute certainty but given the strong correlation and lack of a better explanation I think it's a completely reasonable position to take.

Let me just give one basic example, near the end of this video in Batman:AA - http://www.youtube.com/watch?v=irlmewwLyTo

PhysX OFF is missing basic spark effects from the ceiling, I want to know your answers to the following questions, honestly if you don't mind:

1) Do you agree that, at the very least most mainstream games have been capable of basic spark particles for at least the last 10 years?
2) Would you agree the physics of these basic particles can be calculated on the CPU easily?
3) Do you think it's reasonable these effects are limited GPU PhysX only?
4) What do you think is the most reasonable explanation they are limited to GPU PhysX only?
5) Do think it's fair that basic effects we've had for years are available only to 1 brand of video cards?

I'd like to point out that bringing these effects to both camps would be a simple matter of flipping a configuration variable that makes these effects be calculated on the CPU instead of the GPU, they're running under the same PhysX engine in both cases so no real work is required to correct this.

I eagerly await your answers to these questions, thank you.

 

[Jaydip]

If Physx is leveraging CUDA(not sure) then there is no way a CPU can compute it faster.CUDA is very well optimized and from my experience of using it professionally for many years it is a solid piece of work.

 

[Rvenger]

You must be a shill for AMD and bought a Titan and 780 just to throw some of the super sleuths on this board off the trail.

How nice of you to call me a shill. Well you are really off base because I bought a 7970 because I felt that I had too much invested in my current rig so I downgraded. I have a GTX 650 in my other rig so yeah I am an AMD shill alright. I don't even own an AMD CPU anymore.

Right, so you got rid of your Titan and your GTX 780 and bought a 7970 instead........ :sneaky:

Like I said. $650 vs $250 and 30% slower... I don't need physx because I don't like the performance hit it gives me. Hey but if I have an issue with Nvidia, I am a shill right? If I want to enjoy PhysX I can certainly pull a previous gen Nvidia card out of my closet and stick it in my rig to run PhysX.

I can say that I did like my 780 and my Titan, they were too expensive for me though. If they were cheaper and if the recent drivers would have been better I would have kept them!!

 

[BoFox]

I voted "unimpressed" because I've seen many other non-proprietary engines do physics at least 90% as visually impressively as Phsyx (Havok, Crytek, Infernal Velocity, Red Faction, etc..). It becomes that MUCH more impressive considering that decent physics are done on the CPU with a well-written algorithm.

Usually, Physx is just an add-on/ tack-on gimmick onto a game where the actual gameplay has nothing to do whatsoever with PhysX effects. The original game engine also usually had nothing to do with physics being built around PhysX. In most PhysX games, it's just a few instances in the game "here and there" like the waving flag that suddenly demands billions of CUDA calculations per second compared to the rest of other ordinary objects in the game. In Mirror's Edge and Batman games, the seemingly billion pieces of broken glass shards on the floor just did not feel balanced with the rest of the game at all - only to disappear altogether in a second or two.

It's like Crysis 2's overkill use of tessellation when turned to the max, but worse in terms of "overkill" - a vain show-off of a poorly optimized algorithm that only applies to a select few objects in the game (or at the cost of rendering that game "bland" for those without capable PhysX cards, like complete removal of fog altogether from the console version).


What I would be really interested in is FRAME TIME measurements with PhysX games (with the settings turned to high/max) on capable Nvidia cards.

My experience with Physx games in the past (at least with games older than PhysX 3.0 for sure) was that there was a certain kind of hitching/stuttering (sometimes intolerable) even though my card was capable enough of maintaining a minimum of no lower than 25-30 fps even during the most demanding scenarios. That was, I was not even using a dedicated card - it was only a single GPU.

It would still be interesting to see frame time measurements on a system with a dedicated PhysX card as well, in addition to a single GPU handling PhysX on its own.

 

[FiendishMind]

Let me just give one basic example, near the end of this video in Batman:AA - http://www.youtube.com/watch?v=irlmewwLyTo

In this case shouldn't the blame be placed on that specific developer? I've seen many games that support GPU accelerated PhysX but clearly the developers still made the effort to include the "industry standard effects" when GPU PhysX wasn't being used.

 

[taserbro]

No, I have access to this timeline where I have countless examples of similar games in a similar class that offer these basic features to everyone equally and aren't locked to GPU-PhysX only. The gaming industry forms a strong trend over time which the PhysX games lay outside of in this regard, that requires some explanation.

PhysX being the cause is the most reasonable explanation for the discrepancy, if anyone has a more plausible explanation then speak up. This is real life, we don't have 100% proof of everything, I'm not claiming absolute certainty but given the strong correlation and lack of a better explanation I think it's a completely reasonable position to take.

Let me just give one basic example, near the end of this video in Batman:AA - http://www.youtube.com/watch?v=irlmewwLyTo

PhysX OFF is missing basic spark effects from the ceiling, I want to know your answers to the following questions, honestly if you don't mind:

1) Do you agree that, at the very least most mainstream games have been capable of basic spark particles for at least the last 10 years?
2) Would you agree the physics of these basic particles can be calculated on the CPU easily?
3) Do you think it's reasonable these effects are limited GPU PhysX only?
4) What do you think is the most reasonable explanation they are limited to GPU PhysX only?
5) Do think it's fair that basic effects we've had for years are available only to 1 brand of video cards?

I'd like to point out that bringing these effects to both camps would be a simple matter of flipping a configuration variable that makes these effects be calculated on the CPU instead of the GPU, they're running under the same PhysX engine in both cases so no real work is required to correct this.

I eagerly await your answers to these questions, thank you.

I don't really have a horse in this race but I'd like to point out one possible source of misunderstanding here.
The sparks you saw on those videos were one of the many mundane and non-gameplay-affecting implementation of a physics particle effect, which I agree, could have easily been done without gpu-acceleration. But that doesn't mean that all previous implementation of these spark effect (or any of the other many visual effects that people often incorrectly attribute to cpu physics like landscape shifting, interactive building destruction, most if not all water effects...etc) in the last 10 years were really physics implementations. Most of them are not.

In fact, many of the older and more impressive things that people credit to cpu physics were just (well-done) scripted animations cleverly triggered by code that hadn't anything to do with physics functions, gpu-accelerated or not. The benefit of physics implementation is that it would allow real-time interaction on a greater level (what you didn't see was that the spark would bounce realistically off of any object that happened to be in its way based on the material properties assigned to those objects whereas a non-physics spark would fall in the pre-animated path and just clip through whatever were in its way, which no one would bat an eyelash at) and much more easily implemented if the assets involved were compatible but even those moderately complex effects can be hard-coded cleverly to look like physics but without really calling any physics functions at all and thus without requiring the computational power needed. In fact, years of having to work with antiquated console hardware have made industry game devs mind-blowingly competent at making these cool-looking effects without requiring much cpu horsepower at all. If you want more complex things that would take too long for humans to recreate, then yes it would make sense to have calculations done on GPU but when you have to make it run on an xbox, it makes sense to animate things (sometimes using physics enabled toolsets) on the workstations and then exporting the pre-computed bezier curves.

To date, some of the more impressive application of real physics in games, putting tech demos aside, only involved some basic primitives and ragdoll skeletons of rudimentary complexity (I'll cite crysis and cryengine in general if anyone cares to correct me on a more impressive physics implementation available commercially). Obviously, nvidia's proprietary solution can't be used by game devs for more ambitious, gameplay-affecting mechanics which is hardly because of their dastardly scheming, but rather because the prospect of a game cutting out a large (and vocal) portion of their potential userbase would be disastrous financially and devs, like most people, like money.
Nonetheless, you do raise a good point in that many of the graphics effects we players expect in the scope of this generation of games really don't need "real" physics to create a perfectly valid illusion of having it.

 

[ICDP]

If Physx is leveraging CUDA(not sure) then there is no way a CPU can compute it faster.CUDA is very well optimized and from my experience of using it professionally for many years it is a solid piece of work.

He is asking about basic particle effects that have been removed entirely from some GPU PhysX enabled games if you use PhysX off. To be honest most GPU enabled Physx games do give a PhysX low setting which would be CPU only. Though clearly there are occasions when even rudimentary effects that could easily be rendered on the CPU are missing entirely.

Sparks
Smoke
Fog

I am currently running a GTX780 because I got it for a very good price after selling my 2x 7950s. Not once did the idea that I could use PhysX come into my decision to purchase it. It's nice to have and is definitely a welcome option, it just isn't a deciding factor for me and obviously many others. If Keys was correct in his mistaken assumption that everyone wants PhysX then AMD would never, ever sell a single GPU.

 

[wand3r3r]

He is asking about basic particle effects that have been removed entirely from some GPU PhysX enabled games. To be honest most GPU enabled Physx games do give a PhysX low setting which would be CPU only. Though clearly there are occasions when even rudimentary effects that could easily be rendered on the CPU are missing entirely.

Sparks
Smoke
Fog

I am currently running a GTX780 because I got it for a very good price after selling my 2x 7950s. Not once did the idea that I could use PhysX come into my decision to purchase it. It's nice to have and is definitely a welcome option, it just isn't a deciding factor for me and obviously many others. If Keys was correct in his mistaken assumption that everyone wants PhysX then AMD would never, ever sell a single GPU.

His statement was the most absurd statement in the thread to this point.

Have you tried Physx yet? I hope you got a good deal with the 9970 right around the corner. :thumbsup:

 

[Carfax83]

I don't really have a horse in this race but I'd like to point out one possible source of misunderstanding here.
The sparks you saw on those videos were one of the many mundane and non-gameplay-affecting implementation of a physics particle effect, which I agree, could have easily been done without gpu-acceleration. But that doesn't mean that all previous implementation of these spark effect (or any of the other many visual effects that people often incorrectly attribute to cpu physics like landscape shifting, interactive building destruction, most if not all water effects...etc) in the last 10 years were really physics implementations. Most of them are not.

In fact, many of the older and more impressive things that people credit to cpu physics were just (well-done) scripted animations cleverly triggered by code that hadn't anything to do with physics functions, gpu-accelerated or not. The benefit of physics implementation is that it would allow real-time interaction on a greater level (what you didn't see was that the spark would bounce realistically off of any object that happened to be in its way based on the material properties assigned to those objects whereas a non-physics spark would fall in the pre-animated path and just clip through whatever were in its way, which no one would bat an eyelash at) and much more easily implemented if the assets involved were compatible but even those moderately complex effects can be hard-coded cleverly to look like physics but without really calling any physics functions at all and thus without requiring the computational power needed. In fact, years of having to work with antiquated console hardware have made industry game devs mind-blowingly competent at making these cool-looking effects without requiring much cpu horsepower at all. If you want more complex things that would take too long for humans to recreate, then yes it would make sense to have calculations done on GPU but when you have to make it run on an xbox, it makes sense to animate things (sometimes using physics enabled toolsets) on the workstations and then exporting the pre-computed bezier curves.

To date, some of the more impressive application of real physics in games, putting tech demos aside, only involved some basic primitives and ragdoll skeletons of rudimentary complexity (I'll cite crysis and cryengine in general if anyone cares to correct me on a more impressive physics implementation available commercially). Obviously, nvidia's proprietary solution can't be used by game devs for more ambitious, gameplay-affecting mechanics which is hardly because of their dastardly scheming, but rather because the prospect of a game cutting out a large (and vocal) portion of their potential userbase would be disastrous financially and devs, like most people, like money.
Nonetheless, you do raise a good point in that many of the graphics effects we players expect in the scope of this generation of games really don't need "real" physics to create a perfectly valid illusion of having it.

Good post taserbro. That's one of the biggest reasons why having discussions about PhysX is so annoying, because so many people cannot differentiate between actual physics, and scripted animations.

The two are completely different, although they "look" similar at times.

 

[ICDP]

His statement was the most absurd statement in the thread to this point.

Have you tried Physx yet? I hope you got a good deal with the 9970 right around the corner. :thumbsup:

Ironically in many GPU PhysX enabled games the PhysX effects detract from the immersion. The effect looks comical when an extreme amount of wall debris fills the screen, only to "melt" a few seconds later and the wall is perfectly intact apart from some bullet hole decals.

I tried using PhysX in Metro Last Light at 2560x1600. It causes a ~20%+ performance hit according to FPS figures, which doesn't sound like much but it also introduces some very bad lag when the effects "turn on". With the lag and the minimum FPS dropping to unplayable levels PhysX had to go. The thing is that many PhysX advocates seem to forget that it does come with a hefty performance hit.

I sold 1x 7950s for £180 a few weeks ago and my remaining one last week for £150, so a total of £330. The GTX780 was available for £425 as a refurb, so total cost for my "upgrade" was less than £100. It is an MSI GTX780 Gaming edition so the warranty is still valid for 3 years. In most games I am taking a drop in performance but I play a lot of flight/race sims that simply do not work with SLI/CF. So the GTX780 is giving upwards of 30%-50% performance increase in those games that do not work with CF.

I figured that by October when 9970 is available my 7950s would be worth well less than £300 as they are already available for £160. I honestly think 9970 will be around Titan speed and that at most a 10% delta will exist between GTX780/Titan/9970.

 

[Carfax83]

He is asking about basic particle effects that have been removed entirely from some GPU PhysX enabled games if you use PhysX off. To be honest most GPU enabled Physx games do give a PhysX low setting which would be CPU only. Though clearly there are occasions when even rudimentary effects that could easily be rendered on the CPU are missing entirely.

Sparks
Smoke
Fog

Well you can blame the developer for that, as they are the ones that determine these things. And dynamic fog and smoke isn't exactly rudimentary.

Ironically in many GPU PhysX enabled games the PhysX effects detract from the immersion. The effect looks comical when an extreme amount of wall debris fills the screen, only to "melt" a few seconds later and the wall is perfectly intact apart from some bullet hole decals.

That's a memory saving design decision implemented by the dev, and has nothing to do with what PhysX is, or isn't capable of.

I am currently running a GTX780 because I got it for a very good price after selling my 2x 7950s. Not once did the idea that I could use PhysX come into my decision to purchase it. Using PhysX in Metro Last Light at 2560x1600 causes a ~20%+ performance hit and introduces some very bad lag when the effects "turn on". The minimum FPS was dropping to unplayable levels so PhysX had to go. It's nice to have and is definitely a welcome option, it just isn't a deciding factor for me and obviously many others according to the poll. If Keys was correct in his mistaken assumption that everyone wants PhysX then AMD would never, ever sell a single GPU.

The latest PhysX driver fixes a problem with Metro Last Light so you may want to give it a go and see if it fixes that problem.

Latest PhysX driver

ncludes the latest PhysX runtime builds to support all released PhysX content.



Changes & fixed issues in this release

• Fixes a bug that caused the Metro Last Light to not be GPU accelerated on some systems

 

[Carfax83]

I voted "unimpressed" because I've seen many other non-proprietary engines do physics at least 90% as visually impressively as Phsyx (Havok, Crytek, Infernal Velocity, Red Faction, etc..). It becomes that MUCH more impressive considering that decent physics are done on the CPU with a well-written algorithm.

Sorry to burst your bubble, but Havok, Crytek, Infernal Velocity etcetera are ALL proprietary.

Might want to look up what the word means.

Also, the effects used in those engines are basically standard fare, and don't begin to approach the level of complexity and realism that PhysX is capable of.

 

[wand3r3r]

Everyone is concerned about vendor lockouts or vendor specific stuff and use the term proprietary for that. I assume you've figured that out by now. It's been beaten to death in this thread already.

 

[Carfax83]

Everyone is concerned about vendor lockouts or vendor specific stuff and use the term proprietary for that. I assume you've figured that out by now. It's been beaten to death in this thread already.

That's not an excuse to change the meaning of a word.. Communication in any language depends on commonly accepted meanings and definitions.

And I don't know about anyone else, but if I kept using a word with the wrong meaning repeatedly, then I'd want to be corrected.

 

[ICDP]

Well you can blame the developer for that, as they are the ones that determine these things. And dynamic fog and smoke isn't exactly rudimentary.

I didn't lay blame on anyone, I merely pointed out that it is fairly common in some GPU PhysX games. It doesn't matter what the reason is, the fact it happens is the salient point. I also didn't mention anything about the missing effects needing to be dynamic, just present.

That's a memory saving design decision implemented by the dev, and has nothing to do with what PhysX is, or isn't capable of.

Without proof you are offering nothing but an assumption and pure speculation. The fact is that no matter what PhysX is capable off it has never went beyond some extra debris, smoke or cloth effects that do not fundamentally affect the game play. It's the same as TressFX, it adds a bit of extra eye candy at a substantial performance impact. It's nice to have but it isn't game changing, at least with TressFX everyone has the option of trying it.

Do you find it strange that not one single game has adopted PhysX for calculating flight/car physics models or damage models? Should I take it from this fact that it purely because the devs wanted to save memory? Or should I go ahead and believe it's because PhysX simply can't do those kind of complex calculations? Or should I accept the commonly accepted reason, that it's because PhysX is proprietary and no sane developer would deliberately eliminate ~40% of their potential sales? PhysX may have the capability to produce the most amazing physics possible, the problem is that this potential has never been realised because it is proprietary. It is for this simple reason that PhysX has always been and always will be add nothing but simple eye candy.

The latest PhysX driver fixes a problem with Metro Last Light so you may want to give it a go and see if it fixes that problem.

Latest PhysX driver

Thanks for the suggestion but I am already using those drivers.

 

[cmdrdredd]

I tried using PhysX in Metro Last Light at 2560x1600. It causes a ~20%+ performance hit according to FPS figures, which doesn't sound like much but it also introduces some very bad lag when the effects "turn on". With the lag and the minimum FPS dropping to unplayable levels PhysX had to go. The thing is that many PhysX advocates seem to forget that it does come with a hefty performance hit.

You do know that there was a bug where it can drop to below 10fps and uninstalling physx and installing the stand alone version fixes it right? No more unplayable. That means uninstall the one that comes with whatever driver you're using and install the physx system software alone.

 

[Final8ty]

Cr

Good post taserbro. That's one of the biggest reasons why having discussions about PhysX is so annoying, because so many people cannot differentiate between actual physics, and scripted animations.

The two are completely different, although they "look" similar at times.

Then Physx is not doing a very good job if people in general are finding it hard to differentiate.

 

[cmdrdredd]

Cr
Then Physx is not doing a very good job if people in general are finding it hard to differentiate.

That's not true at all. It's that people don't understand what it means when a scene does the same thing every time no matter what. Like Battlefield 3 when you blow up the building in the first single player mission. That scene is always the same all the time. There's no actual physics calculations there going on but I've seen people try to describe it as such.

 

[ICDP]

You do know that there was a bug where it can drop to below 10fps and uninstalling physx and installing the stand alone version fixes it right? No more unplayable. That means uninstall the one that comes with whatever driver you're using and install the physx system software alone.

Thanks for the info but I am using the latest separate Physx installer. I just have to accept that even a GTX780 can't do PhysX at 2560x1600 without an unacceptable (IMHO) performance hit.

 

[borderdeal]

Quite untrue. If you had the ability to run them both, why would you not? I mean of course this solely depends on the situation and hardware used, but if you could, you would. Explain the alternative. Why "wouldn't" AA be used if it was perfectly fine to do so without fps problems? Same for PhysX or ANY OTHER image quality or feature available for the situation?

Correct and ONLY answer:
People will generally use as many IQ enhancements as they can so long as game is still playable. This of course is completely reversed for pro gamers in FPS. They want sky high fps IQ be damned.

All these people can surely say that they aren't interested, and I'd maybe believe a very small percentage of them are being truthful. Any gaming enthusiast wants it ALL. Yes, I said ANY with the exception of pro gamers dependent of orbital fps.

Not true I am not a pro gamer but I have said it before I cannot see a difference with AA on or off when "I" am playing. So I do not care if some one else sees the difference I do not so I could care less for it. I just finished playing Metro 2033 and Metro Last Light and I played with the lowest AA settings because I could not disable it completely. Again just because you can see a difference and you want it does not mean we all do.

 

[Keysplayr]

I still don't see you trying to refute any of my points. Every attempt at deflection or hiding from answering the rebuttals of your arguments just further proves my points.

Show me your "points". Please.

What you and I call points might be two very different things.

 

[Deders]

Thanks for the info but I am using the latest separate Physx installer. I just have to accept that even a GTX780 can't do PhysX at 2560x1600 without an unacceptable (IMHO) performance hit.

It runs perfectly fine on my 670 with everything set to highest and 2xSSAA enabled. Do you have SSAA enabled? if so it may be causing the slowdown as it has to render everything twice the size of the current resolution. On mine that would be 3360*2100. On yours it would be 5120*3200.

 

[Final8ty]

That's not true at all. It's that people don't understand what it means when a scene does the same thing every time no matter what. Like Battlefield 3 when you blow up the building in the first single player mission. That scene is always the same all the time. There's no actual physics calculations there going on but I've seen people try to describe it as such.

We know what it means but the point is in general most people are not noticing or not caring, at least the scripted stuff looks similar and a bonus of physical game change properties and that part people are noticing more.