PhysX analyzed on CPU

[Chumbucket843]

http://www.geeks3d.com/20100521/gpu-...e-cpu-support/
gpu physx isnt that much faster than cpu physx, only by a factor of ~2 (i7920 and gtx260). if anyone wants to bench a gtx480 or 980x and post results go for it.

[Eastcoasthandle]

Anyone remember the Fluidmark 1.2 article?

[Final8ty]

There is no specific level of physics, but game physics is similar to 3D graphics in how it perceived by gamers. The closer it is to reality, then the more impressive it becomes.

Just as developers strive to make their graphics more realistic, so should they strive to make their physics more realistic as well.

Unfortunately, adherence to reality requires LOADS more computation, and presently, the CPU is very inferior to the GPU in that regard.

You say Havok is sufficient for most physics, but the most commonly touted Havok game (BFBC2) doesn't even use Havok completely for it's most impressive physics effects.

Instead, it uses a combination of scripted animation (which has been around for years) combined with Havok to render the impressive explosions that you see in the game.

Thats all fine and good if you want to be stuck in the past, but technology like Time, doesn't stop for anyone.

With hardware PhysX, not only do we get a greater level of realism, but because developers no longer need to spend time animating effects that can now be run in engine and in real time, development time and cost are reduced.

With that said, the case for hardware accelerated physics is quite strong imo.

So far PhysX has not demonstrated anything real at all, Ghost busters in my view has done better physics wise than any GPU phyxs game & your understanding of what the average joe gamer is impressed by is way off & that's why we see more & more complaints about the quality of today's games from the PC gamers because the average gamer is a console gamer & there standards are far from high & in fact it seems to be dropping.

[Hell Hound]

Anyone remember the Fluidmark 1.2 article?

Therefore, bechmarking seems to be a little tricky in new FluidMark. We are curious if someone will come with solid method after app release.

Does not seem to be a accurate benchmark prog.

[cegras]

There is no specific level of physics, but game physics is similar to 3D graphics in how it perceived by gamers. The closer it is to reality, then the more impressive it becomes.

Just as developers strive to make their graphics more realistic, so should they strive to make their physics more realistic as well.

Unfortunately, adherence to reality requires LOADS more computation, and presently, the CPU is very inferior to the GPU in that regard.

You say Havok is sufficient for most physics, but the most commonly touted Havok game (BFBC2) doesn't even use Havok completely for it's most impressive physics effects.

Instead, it uses a combination of scripted animation (which has been around for years) combined with Havok to render the impressive explosions that you see in the game.

Thats all fine and good if you want to be stuck in the past, but technology like Time, doesn't stop for anyone.

With hardware PhysX, not only do we get a greater level of realism, but because developers no longer need to spend time animating effects that can now be run in engine and in real time, development time and cost are reduced.

With that said, the case for hardware accelerated physics is quite strong imo.

The case you're making for PhysX versus 'shortcut' methods involving a mashup of Havok and prerendered scenes is very similar to (I think) the case for rasterization versus ray tracing. It simply ends up giving equivalent results for less effort and overhead.

For example, Valve has an excellent presentation on how it randomly generates gore on the zombies in L4D2. The 'physx' way (the 'ideal' way) is to generate unique wounds for every zombie everytime you shoot it. However, clearly this is not possible, so a lot of compromises were made. I know nothing about computer graphics but I found the presentation somewhat understandable:

http://www.valvesoftware.com/publica...l4d2wounds.pdf

[zir_blazer]

That doesn't make any sense. x87 supports max 80bit but because physx runs on g80 and agia ppu, which are both single precision 32bit, means its already running with inferior precision.

Worse, because SSE can be used to manipulate up to 4 32 Bits numbers per Instruction.


The truth is that I don't care about physics adoption in games or whatever, all this eyecandy trash doesn't make a game better or worse, or more fun to play. Actually, all the time and resources that developers waste on focusing on graphics alone made games much less original and fun to play.
Two decades ago, developers had to hand optimize code in Assembler to make it run as fast as possible. These days, they rush to release buggy and unstable releases, then simply patch it, and throw more Hardware at the games to compensate for their coding deficciences. You are feeding them.

[Stukov]

I don't know about you guys, but I don't want physics in games for accurate cloth movement, I want physics in games to produce accurate representation of bullets/projectiles used firearms (grain weight, velocity, flight, wind affect, barrel length, etc) and proper penetration of material by bullets/projectiles. Not just walls of wood, stone, or sheet rock, but also body hits.

That way you don't just have damage to body parts based on area (arms, chest, legs, head), but actual accurate body parts like central nervous system, circulatory system, skeletal structure and damage to them actually affecting the person accurately. Bullet hits arm, breaks bone, arm no longer able to be used. Bullet enters body in mid section, bounces off spine, travels upward and nicks aorta leading to quickly bleeding death.

Then you also have projectiles going through material and accurately deflecting, slowing down, or just completely breaking apart and if the projectile can go through, hitting the object behind that material at the properly reduced speed.

I'm just mostly underwhelmed by what physics in games do today, the only one that seems to have a semi use of what I'm talking about is Stalker series, but I don't know if it is even accurate physics or pre-programed.

[Manicdan]

ill make it easy for you then, how about getting shot in the chest with any gun will just kill you? o and health packs dont work, they just slow down the inevitable. lol

[El Mano]

With hardware PhysX, not only do we get a greater level of realism, but because developers no longer need to spend time animating effects that can now be run in engine and in real time, development time and cost are reduced.

LOL you don't really believe that, do you?

Right now Havok is as realistic as Physx. I don't feel that BC2's physics are anyway worst than say Metro 2033.

FarCry2 is as realistic as any other Physx game. And so on.

Besides, if the difference between hardware Physx and CPU physx is going to be just some cloths... then I don't really care at all.

Just tell us a Havok game that's so unrealistic compared to other Physx tittle just because it's physics are ran in the CPU rather than in the GPU.

[Toysoldier]

Actually I would like to see a game like Split Second being done using PhysX. Not because I like PhysX (because I don't) but to see how it would run compared to Havok.
It's simply amazing what they do with Havok in that game.

[Hell Hound]

I would like to see Serious Sam 3 engine with hardware physx support.Nvidia needs to release a driver for the ageia ppu also.


Uploaded with ImageShack.us

[cegras]

I don't know about you guys, but I don't want physics in games for accurate cloth movement, I want physics in games to produce accurate representation of bullets/projectiles used firearms (grain weight, velocity, flight, wind affect, barrel length, etc) and proper penetration of material by bullets/projectiles. Not just walls of wood, stone, or sheet rock, but also body hits.

That way you don't just have damage to body parts based on area (arms, chest, legs, head), but actual accurate body parts like central nervous system, circulatory system, skeletal structure and damage to them actually affecting the person accurately. Bullet hits arm, breaks bone, arm no longer able to be used. Bullet enters body in mid section, bounces off spine, travels upward and nicks aorta leading to quickly bleeding death.

Then you also have projectiles going through material and accurately deflecting, slowing down, or just completely breaking apart and if the projectile can go through, hitting the object behind that material at the properly reduced speed.

I'm just mostly underwhelmed by what physics in games do today, the only one that seems to have a semi use of what I'm talking about is Stalker series, but I don't know if it is even accurate physics or pre-programed.

Oh god no. A Tom Clancy game taken to the extreme would be the most boring thing ... ever.

[hurleybird]

The fact is that the GPU will always be superior to the CPU in this regard, regardless of the amount of optimization on behalf of the CPU.

GPUs are naturally suited to that kind of computation, and when you look at it from a cost/performance perspective, the GPU easily takes the number one spot.

Spot on. Kinda... ok, not really.

Of course the GPU will beat the CPU in terms of a massive amount of simple, inheritley multithreaded single precision floating point math. No one is going to argue otherwise. If you were to benchmark pure physics performance on both, it makes sense that the GPU is going to come out ahead.

Are GPUs a better choice for physics in the context of a game though? Hell no. The GPU is already doing a massive amount of work rendering the game. To make matters worse, no matter how powerful GPUs become game developers and manufacturers will always find a way to use that power, whether it be super complex shaders, massive tessellation, giant textures, huge resolutions, enormous amounts of anti-aliasing, or more action on screen, it is very easy to find a use for increased graphics power. CPUs on the other hand, are generally not used to their full potential in most games. It is hard to adjust a single value like resolution or anti-aliasing to make use of added CPU power. Worse yet, the gap is doing nothing but increasing and the more cores a CPU has the harder it is to utilize all of it's resources (except through physics). In this context, it makes a lot more sense to let the CPU do the Physics work since it will generally have more performance that isn't being utilized by other aspects of the game.

[Xoulz]

A 32 core gpu only cost $40 usd and we have not nearly used physx/physics like we could.

gpu core is not a cpu core. They are saying that a GPU isn't capable of doing what a 2 cores of a modern CPU can do. It is roughly 2x more.

read the article again.. it what every sane person has been saying for the last year. It is what Bad Company 2 proves and why physx is just a marketing tool..!

Need I bring up the Ghostbuster demo again...

[Johnny87au]

I would not mind running some extra threads on CPU for something like PhysX, as no game can utilize 6 cores so far.

Yer true, 980x would come in handy, still overkill though

[saaya]

very nice article!
this finally proves what we all suspected, physix is very poorly executed on cpus, most likely on purpose...

[STEvil]

I would like to see Serious Sam 3 engine with hardware physx support.Nvidia needs to release a driver for the ageia ppu also.


Uploaded with ImageShack.us

I'd rather not, the one they are building is more powerful.

[W1zzard]

Actually I would like to see a game like Split Second being done using PhysX. Not because I like PhysX (because I don't) but to see how it would run compared to Havok.
It's simply amazing what they do with Havok in that game.

yup, split second is the most impressive demonstration of cpu physics so far. awesome gameplay physics

[saaya]

I would like to see Serious Sam 3 engine with hardware physx support.Nvidia needs to release a driver for the ageia ppu also.


Uploaded with ImageShack.us

fyi, those shots look way fishy... those are fake if you ask me...

[Hornet331]

fyi, those shots look way fishy... those are fake if you ask me...

lol what?

Those screens don't look particularly good. Crysis can do better. (if you don't belive me, play SS:FE HD it uses the Serious engine 3)


Anyway, ever since Aeagia descided to rename novodex (which was a good physic sdk ack then) intp physx to sell there useless hardware i hoped it would be doomed... to bad nv picked it up.
But well if you look closer, aegias and nv philosophy are a perfect match in this regard...

[Carfax]

So far PhysX has not demonstrated anything real at all

Well PhysX has been limited to eye candy physics due to the fact that ATI isn't on board.. It has far more potential, but it won't be realized until hardware driven physics becomes mainstream.

Also, PhysX was the first to feature fully interactive cloth and smoke effects in a game, so it's not as though it hasn't been revolutionary in certain ways..

Ghost busters in my view has done better physics wise than any GPU phyxs game & your understanding of what the average joe gamer is impressed by is way off & that's why we see more & more complaints about the quality of today's games from the PC gamers because the average gamer is a console gamer & there standards are far from high & in fact it seems to be dropping.

LOL, you think Ghostbusters is impressive? OK, I'll bite. Explain whats so impressive about Ghostbusters

[Carfax]

The case you're making for PhysX versus 'shortcut' methods involving a mashup of Havok and prerendered scenes is very similar to (I think) the case for rasterization versus ray tracing. It simply ends up giving equivalent results for less effort and overhead.

No, the results are definitely not the same.

Look at the latest PhysX trailer for Mafia 2.

You see fully interactive cloth effects, and destruction effects that utilize tens of thousands of particles, and even the explosive pressure wave, that are all being driven by a physics engine.

Show me a game that uses software physics and scripted animations that has explosions anywhere near as realistic as what you see in Mafia 2.

Check out 1:09 in the trailer. Look at the car bumper. Thats a lot of attention to detail

[Carfax]

LOL you don't really believe that, do you?

Right now Havok is as realistic as Physx. I don't feel that BC2's physics are anyway worst than say Metro 2033.

FarCry2 is as realistic as any other Physx game. And so on.

Besides, if the difference between hardware Physx and CPU physx is going to be just some cloths... then I don't really care at all.

Have you seen the Mafia 2 PhysX trailer?

Show me a game that uses CPU driven physics that comes even close as to what you see in that footage.

Just tell us a Havok game that's so unrealistic compared to other Physx tittle just because it's physics are ran in the CPU rather than in the GPU

Thats easy. BC2 (and tons of other games) uses Havok for physics, but for the big explosions, they also rely on scripted animations as well.

In other words, if you blow up a building in BC2, it will blow up the same way every single time; with the exception of a few hundred objects or so that are calculated with Havok.

This is done because the CPU doesn't have the horsepower necessary to calculate all of the debris and objects that are generated from a massive explosion.

In Mafia 2 however, we see that because the GPU is much more powerful, the destruction effects can be calculated completely and we end up with dynamic explosions that not only generate a lot more debris and objects, but look different every single time.

[saaya]

lol what?

Those screens don't look particularly good. Crysis can do better. (if you don't belive me, play SS:FE HD it uses the Serious engine 3)

maybe its a weird dof implementation but the background looks like its one single blurry texture manipulated in photoshop, and the only thing special about the tank and wall behind it are high res textures, the lighting and shadows are nothing particularly special...

[Carfax]

Are GPUs a better choice for physics in the context of a game though? Hell no. The GPU is already doing a massive amount of work rendering the game. To make matters worse, no matter how powerful GPUs become game developers and manufacturers will always find a way to use that power, whether it be super complex shaders, massive tessellation, giant textures, huge resolutions, enormous amounts of anti-aliasing, or more action on screen, it is very easy to find a use for increased graphics power. CPUs on the other hand, are generally not used to their full potential in most games. It is hard to adjust a single value like resolution or anti-aliasing to make use of added CPU power. Worse yet, the gap is doing nothing but increasing and the more cores a CPU has the harder it is to utilize all of it's resources (except through physics). In this context, it makes a lot more sense to let the CPU do the Physics work since it will generally have more performance that isn't being utilized by other aspects of the game.

It all comes back to price/performance.

All of what you say is true, but you neglected to mention that using a dedicated physX card mitigates all of those problems. In fact, you always get the best results when you use a dedicated physX card.

My GTS 250 which I use for PhysX only, will b!tch slap my Core i7 @ 4.2ghz when it comes to anything involving physics, while costing a lot less as well.

Also, dual cores still comprise the majority of gaming PCs.

The latest Steam hardware survey.

Quadcore, let alone hexcores won't become fully mainstream until a year or two from now.