PhysX games galore!

[f1sherman]

http://youtu.be/hJlkMAZTEPI

YAY

 

[zlatan]

The commercial license for PhysX isn't free if you're going to be using it for GPU accelerated effects.

I can write GPU accelerated effects in DirectCompute. I don't need PhysX for that. NVIDIA should focus on real use case, and not just graphics effects, because Havok is really expensive, so there is a market for PhysX.

NVidia is definitely positioning Flex for the game engine market. It's supposed to be introduced in PhysX 3.4. From their website:

I'm betting that Unity 5 and Unreal Engine 4 will have it integrated into the engines as an update.

Source

Good luck for them. But don't expect much. The problem with complex physics solvers is how to integrate it to the core gameplay simulation. You have to design the game fully for the physics solver, but there is a huge chance that won't work.
Doing a little demo for something and integrating it in a complex game is a different thing. Todays GPU PhysX effects are just graphics. Even if you integrate them in a bad way it won't affect the gameplay mechanics, so get the cash from NVIDIA, and who cares how it is works.

 

[Sable]

Have you seen the Witcher 3 trailer?

Powered by PhysX

Dat water! Dem clouds! Dat hair! :awe:

E3 2014 Trailer

I recall another E3 trailer of a recent game...

 

[KaRLiToS]

I don't understand the point of this thread?

Sounds like marketing done by a Nvidia representant.

 

[sontin]

I can write GPU accelerated effects in DirectCompute. I don't need PhysX for that. NVIDIA should focus on real use case, and not just graphics effects, because Havok is really expensive, so there is a market for PhysX.

Yeah, a visual computing company should not focus on visual computing. :|

 

[Carfax83]

I can write GPU accelerated effects in DirectCompute. I don't need PhysX for that. NVIDIA should focus on real use case, and not just graphics effects, because Havok is really expensive, so there is a market for PhysX.

If these effects can be done in DirectCompute, why aren't game companies doing so? I mean, why even bother with PhysX, which is vendor locked when you can just use DirectCompute and satisfy all of your customers?

And while I agree that advancements in gameplay physics would be nice, the graphical effects are also great as well. In fact, the graphical effects are often more strenuous than the gameplay physics, so it makes sense to run them on the GPU and not the CPU.

Just look at cloth physics for example. I remember when Ageia first came out with their PPU card years ago, and they released a really cool demo called Cell Factor. In the demo there was a single flag that was being simulated by the PPU card with good performance, and it was fully interactive and destructible.

In software mode however, the performance tanked because even the most powerful CPUs at that time (AMD FX dual cores) were incapable of handling it.. Now some of that could be due to inefficient coding, but I'm sure a lot of it was just because the CPUs just weren't powerful enough.

Even today, I have never seen a single game with software physics of any sort rival the simulation of GPU accelerated physics in performance or quality when it comes to cloth simulation.

Good luck for them. But don't expect much. The problem with complex physics solvers is how to integrate it to the core gameplay simulation. You have to design the game fully for the physics solver, but there is a huge chance that won't work.

I'm sure they'll be able to do it.

 

[Carfax83]

I recall another E3 trailer of a recent game...

You're referring to Watch Dogs I take it? If so, Ubisoft is well known for pulling that kind of crap. They did it with Far Cry 3 as well, but AC IV strangely enough was not downgraded at all.

But to be fair, Watch Dogs was a cross generation title so they had the problem of supporting the PS3, Xbox 360 and Wii U which undoubtedly impacted the quality of the assets in a negative manner.

CDPR on the other hand has a good track record with delivering the goods, and the Witcher 3 won't be supporting those legacy consoles so I have faith in CDPR that the Witcher 3 will be as amazing as what we've been led to believe.

 

[DarkKnightDude]

Pretty much still the same few devs over the years using PhysX, they really need to get more devs onboard if they want it to be more of a gimmick.

Of course most probably see it as a graphics fluff to force you to upgrade your card, even rarer then TressFX.

 

[Noctifer616]

Even today, I have never seen a single game with software physics of any sort rival the simulation of GPU accelerated physics in performance or quality when it comes to cloth simulation.

You can't use GPU compute without sacrificing graphics performance on PC. It's the same reason why Physx runs better with a dedicated card rather than a single card, even though games aren't making use of the compute units on that card.

Till asynchronous compute becomes a standard on PC, you aren't going to see much of GPU compute in games.

 

[zlatan]

If these effects can be done in DirectCompute, why aren't game companies doing so? I mean, why even bother with PhysX, which is vendor locked when you can just use DirectCompute and satisfy all of your customers?

They are doing it. They just not create a video about effects on/off.

And while I agree that advancements in gameplay physics would be nice, the graphical effects are also great as well. In fact, the graphical effects are often more strenuous than the gameplay physics, so it makes sense to run them on the GPU and not the CPU.

Just look at cloth physics for example. I remember when Ageia first came out with their PPU card years ago, and they released a really cool demo called Cell Factor. In the demo there was a single flag that was being simulated by the PPU card with good performance, and it was fully interactive and destructible.

Cloth physics is not really good for GPUs. Havok has a cloth version which is CPU-based and it can run much faster than the GPU-based option. There are some games that use cloth simulation in DirectCompute. Dirt and Grid series for example. Sometimes it is useful, but the standard APIs use synchronised compute, so it is not ideal for todays hardwares.

A destructible environment needs a very special rigid body solver. The PPU was a DSP, so it was ideal to accelerate it, but in NVIDIA PhysX all released games use CPU-based rigid body solver.

In software mode however, the performance tanked because even the most powerful CPUs at that time (AMD FX dual cores) were incapable of handling it.. Now some of that could be due to inefficient coding, but I'm sure a lot of it was just because the CPUs just weren't powerful enough.

Well it depends on coding. Havok's destruction module is extremely fast. You can see it in Battlefied 4.

Even today, I have never seen a single game with software physics of any sort rival the simulation of GPU accelerated physics in performance or quality when it comes to cloth simulation.

Play with the Star Wars: The Force Unleashed games. You will be impressed.

 

[24601]

Sure Havok can be fast.

Havok also has always looked fake as hell from the beginning to today.

The point of PhysX is to make something have actual physics, not "physics-ish"

If developers find implementing actual physics in gameplay is difficult, hire developers that actually know math.

 

[tviceman]

Yawn. Wake us up when PhysX is something more than 100% graphical fluff.

You should just stay asleep until HBAO, SMAA, TressFx, and HDR are also more than just graphical fluff because having more realistic and detailed effects apparently is not your cup of tea.

 

[DarkKnightDude]

You should just stay asleep until HBAO, SMAA, TressFx, and HDR are also more than just graphical fluff because having more realistic and detailed effects apparently is not your cup of tea.

SMAA makes more realistic and detailed effects? What are you smoking? :biggrin:

I guess if you count blurrier and smoother edges sure.

 

[Paul98]

UE4 does use PhysX, but only CPU based.

I know at least one person is working on adding bullet also.

UE4 is freaking great

 

[SoulWager]

Yawn. Wake us up when PhysX is something more than 100% graphical fluff.

I think Kerbal Space Program uses PhysX for non-graphical purposes. It isn't GPU accelerated though.

 

[Rakehellion]

Why should I get excited about PhysX? Aren't there about a million other physics engines out there? What makes the proprietary one so special?

 

[thilanliyan]

The PPU was a DSP, so it was ideal to accelerate it, but in NVIDIA PhysX all released games use CPU-based rigid body solver.

I really wish nVidia kept the PPU cards separate. I would have given them my money for one if they had. I needed to keep my primary cards ATI (mining) so there was no chance I was I going to buy a Geforce card. If they had kept the PPU separate they would have got SOME money out of me instead of none, and I'm sure there are others in the same boat as me.

 

[BFG10K]

Even today, I have never seen a single game with software physics of any sort rival the simulation of GPU accelerated physics in performance or quality when it comes to cloth simulation.

Even today I've never seen a single PhysX game that allows the player to dig arbitrary tunnels like Red Faction 1 was doing back in 2001 on a Pentium 2 @ 400 MHz.

 

[24601]

Even today I've never seen a single PhysX game that allows the player to dig arbitrary tunnels like Red Faction 1 was doing back in 2001 on a Pentium 2 @ 400 MHz.

Blame developers.

Consoles ruin everything.

Games stopped advancing in 2001-2002ish.

I think Kerbal Space Program uses PhysX for non-graphical purposes. It isn't GPU accelerated though.

Trine

 

[Carfax83]

Cloth physics is not really good for GPUs. Havok has a cloth version which is CPU-based and it can run much faster than the GPU-based option.

In many ways, this is the exact opposite of what I've seen. GPU based cloth physics are not only much faster, but much more complex. One of the most impressive games with cloth physics was Batman Arkham City. Lots of characters in the game had cloth physics on their person (and there was environmental cloth physics as well), and all of it was being simulated by the GPU with PhysX. Same thing with Mafia 2.

Now compare that to Watch Dogs which uses Havok. In Watch Dogs, the cloth simulation is there, but it's very basic in comparison. Aiden's jacket does respond to his movements, but the complexity of the simulation has been severely curtailed to the point where it doesn't look natural. Same thing with the NPCs in the city, which have even more basic cloth simulation.

The point is, that while Havok can do cloth simulation without impacting speed that much, the complexity and accuracy is nowhere near what PhysX can do.

A destructible environment needs a very special rigid body solver. The PPU was a DSP, so it was ideal to accelerate it, but in NVIDIA PhysX all released games use CPU-based rigid body solver.

The GPU can do it as well, but developers are reluctant to do so given that it is vendor locked. Look at the Hawken, which offers full scale destruction.

Well it depends on coding. Havok's destruction module is extremely fast. You can see it in Battlefied 4.

I feel as though Havok places too much of a premium on speed, and not enough on accuracy or authenticity.

And the large scale destruction events are all scripted..

Play with the Star Wars: The Force Unleashed games. You will be impressed.

I actually have this game, but I had no idea it even had cloth physics. I uninstalled it after a few hours of playing it, as it wasn't a very good game and it took up a ton of space..

 

[Carfax83]

Even today I've never seen a single PhysX game that allows the player to dig arbitrary tunnels like Red Faction 1 was doing back in 2001 on a Pentium 2 @ 400 MHz.

A game has to be designed from the get go with that sort of thing in mind..

The closest thing I can think of is Everquest Next which offers a lot of environmental destruction and uses PhysX.

 

[f1sherman]

physics model can be executed faster on cpu only if

• a) it's a very simple model
• b) problem can not be parallelized well enough
• c) GPU solver is outright terrible

In all other cases GPU>CPU.
That's why today we have GPU renaissance in physics and supercomputers.

As for Havok, I thought it can use GPU nowadays.

 

[dn7309]

I like the Physx effects, but the performance hit is too high. It like running software T&L back in the ati rage days. I thought when you have dedicated hardware, there should be little to no hit in performance.

 

[Carfax83]

I like the Physx effects, but the performance hit is too high. It like running software T&L back in the ati rage days. I thought when you have dedicated hardware, there should be little to no hit in performance.

You can't get something for nothing. For some reason, many people believe that you should be able to have sophisticated physics effects which use tens of thousands of individual particles for practically no cost in performance..

Yet, they accept the fact that running higher levels of AA or ambient occlusion etcetera will lead to a performance hit

 

[24601]

You can't get something for nothing. For some reason, many people believe that you should be able to have sophisticated physics effects which use tens of thousands of individual particles for practically no cost in performance..

Yet, they accept the fact that running higher levels of AA or ambient occlusion etcetera will lead to a performance hit

It's not like old PhysX capable Nvidia cards are that expensive.

I'm sure my 460 GTX 768mB would still make a great dedicated PhysX card.