Physics Processing Unit?

[randycat99]

I'm not really understanding how "physics" has become so specialized that it "seems" so undesireable to do it on even the greatest general purpose processors (the usual suspects, that is). I think we can all agree how graphics have migrated to GPU's, but what's really so special about physics? How many people could really distinguish what is "realistic" dynamic behavior in a depiction? I frequently see cases in major movies where "credible physics" momentarily takes a backseat for no real good reason. Yet, this kind of stuff passes over people's heads just like that. [snaps fingers] So is there really a great impetus for "better physics" in games? I'm not debating if this would be a true technical improvement (because undoubtedly, it is). The question I pose is would people really notice? Could they really distinguish between "bad physics" from "good physics" or even from "average physics"? If not, then how can you really justify an extra "PPU" (presumably for "superior physics") when it is already possible to achieve "average physics" using simply a general purpose CPU that will already exist in a system?

You can add an extra spin on this idea and ask if the average person would rather have a (further) incremental increase in graphics "sparkle" or more realistic physics? I have a feeling their answer would not place as much interest in physics as we think. Why is this question significant? Naturally, if you have to budget a PPU into a system, you have to ask how much more value does it really add to the system (in the way the games will become and will people really be able to pick out the improvement) vs. why not just throw that money at more GPU-oriented refinements.

Given all that, I completely do not see what all this fuss is over having an extra PPU, when you have 1 of 3 ridiculously high-clocked PPC's to do just "better than average physics" (to pick XB2 as a console example). The only reason I could see to justify a PPU is if a general purpose CPU is simply incapable of doing the job. I think it is fairly well established that they most certainly can do the job. The real reason we don't see better physics is not because the processing resources are missing, but because the persons who "choreograph"/design/coordinate (for lack of a better term) the action that takes place in a game aren't particularly interested in realistic physics nor is the general audience particularly "informed" (in an engineering sort of manner) to appreciate if realistic physics is occuring before them or not. It may sound sad, but if it is true, then it is still moot whether it sounds sad or not.

Here's an example- when you have a prerendered car chase for a cutscene, how realistic does the dynamic behavior look? (imagine an NFS CG cutscene, as an example) For the most part, it is severely "canned" and would not pass for a movie sequence in the slightest even if the CG looked absolutely real. I suspect that most people do not even notice this. It's looks "normal" enough for them. To me, it is like watching a poorly done special effect in a movie where you actually see the "strings" (yeah, that corny).

There is no reason for this to have happened. It's prerendered- they had any amount of processing budget to invest to develop "real" trajectory paths and resolve inertial movements/rotations properly. Yet they didn't. Either it was not a great concern for whoever produced the CG cutscene and/or they thought the audience would not pick up on the "canned" motion. Given this, I have serious doubts that even if a PPU is made available as part of an architecture, it would get even paltry use to justify its presence. It's just not as great a concern as we think it is (talking about actual implementation and use, not just talking about it as a kewl thing to have), imo.

 

[Brimstone]

Personally I think people are going to totally notice the difference between previous generation games and next-gen games on X-Box 360 and PS3. In realistic military first person shooters espacially, the enviroments are so static, yet all the powerfull weapons and grenades are shooting and exploding everywhere.

The enviroment will react to what you do to it, this only happens in a very minor degree currently. Myself I like the violence aspect in in games like Rainbow 6. Physics will make the guns seem more violent. You shoot some glass and broken shards spray everywhere or shoot a phone on a desk in it goes flying off in some direction. Remember the scene in the Matrix movie where near the end where they were shooting out parts of the walls? While maybe a little over the top, to me it looked really cool and made the scene feel more violent.

 

[randycat99]

True those things should be good eye fodder. However, you can achieve those very same effects without hardcore physics simulation. Just as easily they could be pre-scripted motions (now, I'm not saying I'm in favor of pre-scripted actions- I'm just saying it's hardly something that only real physics simulation can achieve). Additionally, it's completely within the realm of general purpose CPU's to pull off simple physics actions such as that. ...but maybe what you had in mind is that this sort of reactivity of the environment should be more pervasive (so as to justify PPU)? I can buy that, I suppose. I think the real question then is will developers bother to put all sorts of extra reactive elements in a game scene? No doubt, some will. I don't think it will be pervasive amongst games, nor pervasive within a game, though. I think the fact remains is that this capability has been already possible with the best that Intel/AMD/IBM offers, yet "physics useage" continues to remain sparing.

 

[GwymWeepa]

Its one of those things that add polish to a game I suppose. The general gamer has a tough time telling a real good looking ps2 game and a real good looking Xbox game, despite all the effects that Xbox title is doing that the ps2 can't hold a candle to...but for those who appreciate graphics, hey its there. I would love a PPU in a system and can appreciate the effects, even if they seem subtle...like global collision detection, appropriate fluid dynamics, tens of thousands of particles flowing naturally in the wind (like the fragments from an explosion, or the resulting smoke).

 

[randycat99]

All of those things can be done on a "nearly believable" basis w/o PPU acceleration. That's where the notion of "appreciation" becomes a bit elusive. Anybody can appreciate "a thousand particles floating in a wind" even if it is just a pre-ordained effect. It is few who will bother to realize the particles are moving 27 deg astray from where they should according to the supposed conditions of the scene (to give an example). Is it even worthwhile to notice such things as "27 deg"? Just having a flurry of particles (meaning who's going to count how many particles as long as it looks like a "lot") floating in the wind will have achieved the desired affect w/o it being "physics-genuine". The way you've described these various uses of a PPU certainly sound impressive and exotic, but I suspect that many people are satisfied with quite a bit less in physics accuracy- as long as particles are "floating" in a winderly motion and shrapnel is radiating "outward". For that, you don't really "need" physics acceleration to accomplish (though it certainly can be employed and would be more "accurate").

 

[GwymWeepa]

I doubt most people can notice bump-mapping, super-duper anti-aliasing etc, but the cumulative effects add up. Imagine an RPG, your character is walking through a scene, its fall, there's an early snow. There is global hit detection, as you walk through a field all the grass bends out of your way realistically, the snow deforms properly under your feet, the snow flakes and the last of the fall leaves dance about you in a wind, the clouds above you deform with the wind as well. Very subtle effects, but if enough are portrayed properly, it gives a viewer the sense of a real world, even if they don't quite know what's going on.

 

[PiNkY]

I don't think this is necessarily about archieving 'true to live physics' in games but rather about noticably increasing the amount of dynamic geometry in scenes. Particles are fine for certain effects and certainly will be in heavy use for years to come, but once you want a significant increase in interactive geometry a dedicated PPU starts to make sense. Take a look at HL2 for example: the increased amount of dynamic geometry has quite an
(positive, imho) impact on gameplay. Point is dedicated graphics-HW has vastly increased the average scene complexity of games over the past years, but most of these scenes consist of quite non-interactive objects.

 

[gamepower]

Look at point 140 called" Additional Cpu Module":

http://xbox2.gamegossip.com/images/xbox2/xbox2_1.png

is this a indication of a physik coproz.???

 

[MfA]

Considering the SMP bit, not likely.

 

[randycat99]

I doubt most people can notice bump-mapping, super-duper anti-aliasing etc, but the cumulative effects add up.

There's the catch. It can look good one way, but if somebody else can make it look just as good using a less computationally intense way, no one is going to care if the first way needed special hardware to do it.

Imagine an RPG, your character is walking through a scene, its fall, there's an early snow. There is global hit detection, as you walk through a field all the grass bends out of your way realistically, the snow deforms properly under your feet,...

I dunno, it's probably just me, but using global hit detection to make footprints sounds like a huge, gratuitous waste of resources. It seems a far more efficient way to achieve functionally/visually the same effect is to simply put footprint marks and crushed grass patches where the characters feet have landed. No one is going to be zooming in on the feet as they land to watch grass be crushed underneath. That's just silly, imo. Without a doubt, the way I suggested is far less "fancy" and technically less impressive, but visually, no one is going to care how it came about.

...the snow flakes and the last of the fall leaves dance about you in a wind, the clouds above you deform with the wind as well. Very subtle effects, but if enough are portrayed properly, it gives a viewer the sense of a real world, even if they don't quite know what's going on.

Even portrayed "improperly", few, if any people will notice the difference. The effect need only be visually "convincing", not genuinely physics driven. That's the thing I have been trying to convey here is that all of these examples can be done convincingly even without a PPU or even an intensive physics engine. It just seems like people have thought up all these ways to utilize a PPU, just for the sake of putting one to use, not because these sorts of effects could not be possible w/o one. They most certainly can be done. They just wouldn't be physics driven. The only crucial thing is to "can" the motion convincingly (maybe less exacting a science, but that's what makes an artist a "good" artist), and no one is going to care how the motion/effect was derived (as far as getting the "vibe" of a scene"- just that it is there.

This is really coming out to sound more like a "buzzword" feature, rather something that can really add to the experience in a way that cannot be simulated/duplicated via alternate means. I hate to sound so negative about it, but I have yet to really hear anyone come up with a compelling use for a PPU that just could not be done with a general purpose CPU. Conceivably, the pervasively reactive environment scenario (where anything and everything can be busted up or moved) seems like the most likely scenario. However, that gets into some seriously daunting issues of content creation. Literally, you would have to build everything right down to their smallest components- a wooden floor plank by plank, every book on a shelf, every rock in a cave, etc. (not to mention assign specific physics properties to each of these "elements"- if anybody here has ever tried to model an entire environment "FEA style", you know what a huuuuuge pain in the A that can spiral into...) This could raise the tedium of designing a game level to impractical proportions. I'm just not sure every game developer is really up to a challenge like that, no matter how kewl we would find this on our end.

 

[Brimstone]

Physics are barely even used toady. The computational demand to have numerous objects become interactive just isn't possible.

A combat game based in the jungle will be totally different visual experience because of massive physics processing power. Long grass and low branches will bend as character models walk through them. They will become torn apart from explosions and bullets impacting them.

 

[gurgi]

Physics are barely even used toady. The computational demand to have numerous objects become interactive just isn't possible.

A combat game based in the jungle will be totally different visual experience because of massive physics processing power. Long grass and low branches will bend as character models walk through them. They will become torn apart from explosions and bullets impacting them.

Yeah, it reminds me of ray tracing conversations. Why do it, when you can do all these little tricks that amount to the same thing. Why?...so you don't have to do all those little tricks. Hardware to do ray tracing is a ways off, but for something like physics, if we have the capability now, why not use it instead of the hacks.

It sure would be nice to have most games have a high amount of physics interaction, instead of a few games partial support. Oh, in this game, grass bends for you, but in that one, you can destroy walls. None of it is together, and only a few games bother with these details.

With a good game API, this could really bring realism to the table and maybe mitigate the diminishing returns of art budgets. Less programmers, more artists/designers!! =P

 

[GwymWeepa]

I stick by the argument that it enhances the experience. I mean, why ever bother with more sophisticated physics if most people don't mind an approximation? As graphics get more realistic it becomes increasingly odd not to have accurate physics to match the accurate graphics. For instance, my friend was playing HL2, he tossed a can of paint at a wall, it left a splash...he threw it again...it left the same texture. That sort of ruined the moment, because at first we were saying "Oh cool", the next moment we thought it was lame.

 

[Frank]

Physics can help the art department as well: instead of making and animating multiple versions of each object (clothing, water, plants and destructible objects come to mind), you just have the meshes calculated. They could do that now, but it would require a much faster CPU.

A single Cell processor could do it as well, and when you have the PPU / SPU's calculate the meshes, you can combine the vertex processing while you get better physics.

 

[digitalwanderer]

I still think the big thing the PPU will end up bringing to the table will be much more/realistic destructible environments.

I'd pay $200 to be able to blow the bricks out of a wall in a game I think.

 

[Frank]

I still think the big thing the PPU will end up bringing to the table will be much more/realistic destructible environments.

I'd pay $200 to be able to blow the bricks out of a wall in a game I think.

Try Silent Storm. Fallout (Tactics), Jagged Alliance 2? Yep, that's the only real succesor. And you can blow just about anything to small pieces in 3D. GREAT game!