View Full Version : Is physics the future?
When we look at the only thing GPU's are meant to do, making first person shooters more gruesome / lifelike / accesible (depending on your perspective), is physics going to offer more than better things to launch with your awesome gravity gun?
:shock:
:D
If we expand the target audience, is better physics processing offering FPSes better destruction? More satisfying ways to make small pieces of virtual people and buildings? Do we want lots of intestines, blood, real-life severed limbs and buildings destroyable into gravel?
:lol:
Anyway, what do we need to make more life-like virtual environments? And what is going to provide it for us?
Bottom line: are consoles able to do it? If not, forget about it. No consoles -> PC users need to buy additional hardware -> it isn't going to happen.
Better ways to simulate real life would be awesome. But then again, MMOGs tend to stylize their idea of reality to a common demeanor that has nothing to do whatsoever with real-life, but everything with reducing advancement and misuse by other players.
The only thing we might all agree upon, is that we want objects to behave "naturally" when handled. Better objects. Who make it easier to make those objects in the first place as well. And automatic, life-like animations!
So, do we want first person shooter games to be realistic, gruesome and short (as you would die REAL FAST and there are no reloads in real life), or do we want them to be gruesome, with lots of blood and horribly realistic severed limbs, but in some kind of fantasy world where you, the player, are a God and kill them all?
And for the REAL enthousiasts, what do we do with the other games? :P
Btw.
How long do you think a swordfight or gun battle takes in real life? Seconds at most? Exactly. No long drawn out game-style or hollywood-style battles. You come, see, and win or die in a heartbeat. Where is the fun in that?
Unless... but no... that would be sacrilegious, you want a game that ISN'T focussed on killing everyone else! The horror!
:wink:
I think with AGEIA does have potential with physics (CPU or PPU form) but I dont see it happening anytime soon. (At least mainstream). With dual core chips coming down the pipe i think this will be able to hold over until either the ATI/NV bring the technology and include it in the GPU or it does go the way of a seperate chip ala a distinct chip (AGEIA) or similar PPU etc.
I think that if developers really strive to make A.I. a focul point (and GPU's and SLI have yet to be used to their potential in current engines and will keep on progressing) then we might see this move. Personally (as I have said in other threads) that physics although I would love to see them brought up a few notches, I would prefer more emphesis on A.I. progression then physics. Yes as you pointed out there would be more of a natural feel with improved physics but with a serious dedication of A.I. and a steady progression of physic methods I think the impacts of a games nature (be it combat tactics to NPC interraction) would raise the game immersion bar immensely. The graphics are definately progressing. With A.I. progressing at an accelerated rate then physics, (in my opinon) would have no choice but to progress along with GPU and A.I developments to keep the balance. (Least I hope the developers focus on balancing each technology).
I know this is just rambling (its been a long day and medications after surgery dont help the matter much) :) But i hope i tried to be clear on my points.
Anyone else have opinions?
***There is no Luck...Only the Will and Desire to Succeed !!!***
IgnorancePersonified
08-Apr-2005, 12:10
my thinking is that a GPU or VPU has enough complexity as it is.
GPU's are definately not enough. There's are lot more beyond 3D. Pun intended. The most obvious being:
- Physics
- Character animation
- AI
All these elements are important in creating an immersive experience. And the amount of processing to be done to make all these 3 elements work together is SO OUT of today's CPU's league, it's not even funny!
Imagine one example: Say you have a charater, in the form of a beautiful woman in a long dress, and you want to make her sit on a chair at in the middle of the room, just by giving her a script command "sit on chair A". You'd have to:
- Give her dynamic realistic movements, with hair and clothing deformation (having the dress sway and conform to her form),
- Sense of awareness to the environment, both in facial expressions/head-eye movement, and transposition of obstacles. All in a realistic manner.
- Realistic movements while interacting with the chair
- etc...
All these steps would use a whole freakin' amount of processing power to be done in real-time, and encompass all three elements I mentioned earlier (physics, character animation and AI). There's no way we're going to be able to pull this off in the years to come, unless we dedicate the same amount of power to it, like we did with 3D renderings. Dual-Core or Quad-Core general processors like AMD/Intel are just NOT the way to go. We'll need massive parallization and vetor processing power.
3dilettante
08-Apr-2005, 13:05
One of the problems I think will slow the adoption of physics hardware is that it is such a niche product.
No word processor or spreadsheet program will ever need physics, nor will most games. With the exception of games that involve heavy immersion, there isn't much else that really needs good physics.
Why should I care if my simulated city's traffic doesn't show proper inelastic collision behavior?
How often will the user notice the difference between a rough software approximation and a more accurate physics model?
"Look! That see how that ball bounced to 1.2 cm? It should have bounced 1.1, that $300 physics card sure payed off."
One of the problems I think will slow the adoption of physics hardware is that it is such a niche product.
No word processor or spreadsheet program will ever need physics, nor will most games. With the exception of games that involve heavy immersion, there isn't much else that really needs good physics.
Don't underestimate the gamming market. The only reason people buy expensive home PC's, is so they can play HL2, EQ2 or whatever... Any crappy 2 year old PC can run Windows, Wordprocessors and Spreadsheets. Even with Microsoft's effort of making them slower and slower.
"Look! That see how that ball bounced to 1.2 cm? It should have bounced 1.1, that $300 physics card sure payed off."
I'm sure you know that there's more to physics then bouncing balls and flying objects.
Simon F
08-Apr-2005, 13:45
Imagine one example: Say you have a charater, in the form of a beautiful woman in a long dress, and you want to make her sit on a chair at in the middle of the room, just by giving her a script command "sit on chair A". You'd have to:
- Give her dynamic realistic movements, with hair and clothing deformation (having the dress sway and conform to her form),
That can actually be done with the graphics system (well, the cloth modelling to some extent).
3dilettante
08-Apr-2005, 13:49
Don't underestimate the gamming market. The only reason people buy expensive home PC's, is so they can play HL2, EQ2 or whatever... Any crappy 2 year old PC can run Windows, Wordprocessors and Spreadsheets. Even with Microsoft's effort of making them slower and slower.
Not all games need heavy-duty physics. Even those that are good fit for a physics processor are going to run into diminishing returns. Many physical effects are very subtle, and I think a lot of people are going to balk at adding another expensive component to their system that essentially exists only for gaming, and only for some games.
I'm sure you know that there's more to physics then bouncing balls and flying objects.
Yes, but not all of it is readily apparent to the user. Games that leverage it will find it hard to justify the end result given the additional effort put into programming for another processor.
Unless the physics processor is really cheap and can be placed as a checkbox feature on a sound or video card, I don't see it doing too well for a long time.
I think physics has the potential to bring a completely new level of interaction to games. It literally can add a game to a game because you can't really predict what realistic interactions can do for you until they are there.
Physics don't have to be realistic either. Why not create your own world with its own physical laws? Never had a game like that have we.
The possibilities are very exciting to me.
Inane_Dork
08-Apr-2005, 19:35
Physics is becoming important because correct-looking motion is becoming more important. Surface detail is shooting through the roof, but so much animation is still old-school keyframe blending with hands going through swords, swords going through walls, gliding walking, etc.
Where I disagree with many people is over the need of a PPU (or something of that nature). CPUs and GPUs are already converging. Why introduce a third branch in between the two now? In 4 years we'll just be folding it into the CPU (assuming it takes hold in 4 years, which is another debate).
Physics chips will need a "The Sims of physics" game that will get people out buying because they NEED that hardware because its such a great game.
I'm thinking something involving beseiging beautiful castles made of large numbers of physics bricks and lovely manually aimed seige weaponry...
Mmmmmmm castle seiges.
Preferably with a nice custom castle editor :D
I would imagine that the PPU is a short fad. Dual and quad core chips will quickly replace the need for them. Intel itself has said to expect 8 threads per chip by 2010 I believe. Whether that's hyperthreaded or not is moot, 4 cores would still be extraordinary.
I think physics is just one aspect to certain types of games. Each area or aspect will need to get better for more immersion or should I say more realistic experience if that is the goal of a game. Physics isn't needed in all games but most games do need graphics but then again not all. I do believe better physics could enhance certain games, now do we need dedicated PPUs? In some cases yes, in others no. Now having a dual core CPU combined with a potential dual graphic core(card) setup, do we really need to add another processor into the mix? Another API. It does seem like physic calculations should be done more on a vertex level or geometry based processor. How would ten of thousands of interactions between all the objects in a given scene be coordinated if behavior is physic based between GPU and PPU over a bus? For example using particles where your action affects all the interactions of the scene.?
Many of the arguements against the idea of a PPU have already been used unsuccessfully for many years against GPUs.
Sure, GPUs are more typically vital to most people but like GPUs, a dedicated chip is always going to be easier to scale up in power than a CPU.
Sandwich
09-Apr-2005, 12:11
I think physics is just one aspect to certain types of games. Each area or aspect will need to get better for more immersion or should I say more realistic experience if that is the goal of a game. Physics isn't needed in all games but most games do need graphics but then again not all. I do believe better physics could enhance certain games, now do we need dedicated PPUs? In some cases yes, in others no.
I agree. I think the best hope for PPUs is Xbox2 and PS3 games using alot of vector processing power to create lifelike animations like hair blowing in the wind and clothes folding and small bouncing fragments in explosions.
Things that won't interfere with game mechanics, but make the game look better.
That way developers can still port games to our molasses PCs, sacrficing visuals, while including PPU support for the extra animations.
Maybe the public will notice the dfiiference and want those PPUs. Else the PPU is probably dead.
Now having a dual core CPU combined with a potential dual graphic core(card) setup, do we really need to add another processor into the mix? Another API. It does seem like physic calculations should be done more on a vertex level or geometry based processor. How would ten of thousands of interactions between all the objects in a given scene be coordinated if behavior is physic based between GPU and PPU over a bus? For example using particles where your action affects all the interactions of the scene.?
If the PPU doesn't succeed we have 3 options:
1. Intel or AMD will soon come up with a solution that is a fast as Cell. Unlikely.
2. Nvidia or ATI will manage to fit a PPU and a GPU on one card or chip.
3. no more games for the PC. Everybody get a console.
nutball
09-Apr-2005, 13:40
1. Intel or AMD will soon come up with a solution that is a fast as Cell. Unlikely.
Why is that unlikely? Why is it implausible that Intel (specifically, or AMD if you like) are capable of coming up with something to match Cell?
I find the mythical status that Cell has attained over the past months quite amusing (particularly as there haven't really been that many public demonstrations of its *actual* performance, people are judging by numbers).
Sandwich
09-Apr-2005, 14:39
Why is that unlikely? Why is it implausible that Intel (specifically, or AMD if you like) are capable of coming up with something to match Cell?
Pentiums and athlons must work with existing software. The cores are held back by x86 legacy and bloated with OOOE logic.
An athlonFX-55 runs at 2.4ghz and can execute 3 instructions in parallel.
Cell is predicted to reach 4.0ghz. 8 SPEs, each capable of 2 instructions in parallel.
Even for a 4 multicore athlon to come close, it would have to be 2X more efficient than cell.
Sofar we haven't seen any dual AMDs yet and intels pentium D is at best 70% faster than a 3ghz P4.
nutball
09-Apr-2005, 14:48
Cell is predicted to reach 4.0ghz. 8 SPEs, each capable of 2 instructions in parallel.
Right, but the SPEs are special-purpose units, are they not? Cell also have a PPC-derivate general-purpose core that acts as a front-end (if I undertstand the architecture correctly). If you are comparing the 8 Cell SPEs with the x86 cores, you're not comparing like with like. Think of the x86 core as the PPE, and the SPEs as the putative new unit(s) to be added. My question is why it's impossible for Intel to design and build something like the SPEs and stack 8 of them around an x86 core.
(OOOE isn't specific to x86 either, and in-order doesn't have such a good reputation on some problems, so the SPEs in Cell may yet prove to be a PITA to code for).
Sandwich
09-Apr-2005, 14:53
My question is why it's impossible for Intel to design and build something like the SPEs and stack 8 of them around an x86 core.
If intel have something like this planned, sofar they haven't made their plans public. The roadmaps show dualcores and multicore cpus.
chavvdarrr
09-Apr-2005, 14:55
An athlonFX-55 runs at 2.4ghz and can execute 3 instructions in parallel.
Cell is predicted to reach 4.0ghz. 8 SPEs, each capable of 2 instructions in parallel.
Even for a 4 multicore athlon to come close, it would have to be 2X more efficient than cell.
Sofar we haven't seen any dual AMDs yet and intels pentium D is at best 70% faster than a 3ghz P4.
you forgot that x86 cpus can execute SIMD instructions too (as Cell)
Well, yes, while I agree with most of the above (and especially that the AI is much more interesting at the moment), but what is it we actually want to do with those physics?
If we look at games as they are now, we have this fairly simple state machine, in which objects are moved and transformed through simple rules, like the bone animations, and force / weight calculations, after which we calculate the exterior and send that to the GPU to be drawn.
But we want it to be more realistic. We want clothes to flow, we want objects to move naturally. (Edit: and we want the feet of the actors to actually touch the ground on slopes ;) ) At first, that would mostly be updating the rules that move and transform the objects. But that isn't enough. You want those objects build out of many smaller objects, that all work independently.
Like this:
If you want realistic swordfights, you cannot use predefined animations. Impossible, unless you want to spend more than 90% of your total budget on those. You need the objects to be able to move independently.
To do that, you need to define building blocks, like bones that form a skeleton, and muscles that move it. And if you want the looks to reflect your actions (bulging muscles, anyone?), you need to give those muscles a volume according to their power. And you would want the computer to make them fill the volume they would take up. Wrap that in skin, and you have a decent, really life-like representation of a human.
To make it move, you want to make lists of muscles involved in performing actions, extend force, calculate the new skeleton orientation, calculate the new volume of the muscles and wrap the skin around it.
And to make fights possible, you use the forces those muscles extend, the mass of the limbs and sword, collision detection and hit actions, like disabling the muscles / bones that are hit, or chopping off the limb. Which would change the functional parts of your object.
At that time, the AI would have to be clever enough to compensate for the damage done. No small task!
All of the above would require hardware acceleration. That means, we want to off-load the scene management to a specialized device. So, no scripting of actions, no pre-cooked animations, no triggers that are sure to fire. And if we can do all that, you want to be able to pick up an object, throw it to your opponent, see your opponent react (move aside? get hit?), and take advantage of all that.
Which brings me to the last point: how in hell are we going to control all that action as a player? With a full body suit that follows our movements? And how about feedback when hitting something? Otherwise, you would reduce the interaction to triggering predefined moves...
DiGuru, yes that is definitely the way we need to go in the future but it's such an incredible challenge...yikes!
I really think the middle-ware software development in gaming is just beginning to explode. Middle-ware companies can focus on one specific are and provide their engine plugin for the game developer to utilize. The level of realism you describe is on the level that I just can't imagine a single company being able to manage it alone.
Perhaps we'll have a middleware solution that is a universal human emulator lol. It could help a game company almost instantly have lifelike human NPCs with realistic properties.
Things like that. Parts of the overall program developed separately from the game itself. Like Havok, etc.
universal human emulator
Yes, that would make sense. And I think, that it would be the single most prominent thing that makes games feel "alive". Architecture that is too smooth can be forgiven when textured right, but human looks and animation is the defining thing.
Ok.
So, to answer the questions about what to do with physics and where the physics engine is to reside from the other threads about it, I would suggest: in the device that offers scene management and a universal human emulator.
:D
And in the mean time, expect physics to be used mainly for more items you can launch with your gravity gun and better / more horrific destructions.
Edit: I almost forgot: and breasts that move, of course! :wink:
As the complexity increases or physic computations manages object and vertex manipulation, how will three separate processors or multiple CPU's and GPU's keep in sync with the available bandwidth between all the processors? The GPU will not know what to render until the physic calculations are completed for each object affected and then again AI will also need data points or information to process behavior etc.. Seems like a small boost by off loading the cpu is possible but once real complex scenes are sought, coordination and bandwidth may become an issue.
To me it just seems more logical to have the physic calculations incorporated onto the GPU or have fast access to the GPU since that is where physic calculations have the most affect and a place where the highest bandwidth is available. Vertex animation is already done on the GPU, adding some physic rules or processing ability seems to me to be what is needed in the future. Animating hair and detecting collision and movment with other objects and atmosphere should be a vertex program or should I say a smart vertex program, all self contained for the hair object. No need to tie up the bandwidth of the cpu, memory and a separate PPU for realistic air animation of characters. Let each object stand on their own until a new variable is inserted into the calculation (like water pouring on the hair changing its characteristics) which will once again become it's own entity in behavior without a lot of processing overhead outside of the GPU.
While there is no doubt that physics plays its part in a modern game, I do wonder as to what depth of detail we'd all really notice. In something like a racing game, to my mind, it would be wasted processing time modelling stone chips flying about or turbulent airflow around the head of motorbike rider; it's wasted because such micro-detail would not be noticeable in 17" screen and a few million pixels. DiGuru touched on this by questioning how one would adapt a player's controls or movement from the actions of these effects.
Sandwich
10-Apr-2005, 16:13
While there is no doubt that physics plays its part in a modern game, I do wonder as to what depth of detail we'd all really notice. In something like a racing game, to my mind, it would be wasted processing time modelling stone chips flying about or turbulent airflow around the head of motorbike rider; it's wasted because such micro-detail would not be noticeable in 17" screen and a few million pixels. DiGuru touched on this by questioning how one would adapt a player's controls or movement from the actions of these effects.
How about..
flags and tree branches in the wind
bits bouncing on the track after accidents
debris scattering when bikes run through
offroad dirt trails
animation of riders, specially when they fall
What I don't think would be a good idea, is letting the PPU do the actual bikes.
Most of the objects would consist of shaped volumes, that would react on external forces, like gravity, wind, force extended and touching other objects. Volume, mass, anchor points, collision detection and force applied. And they would have to calculate their new volume, orientation and mesh according to that. All those volumes together, with the bounding volumes around them, would make a scene.
As an artist, you would build volumes, attach them, specifiy the forces, mass, flexibility and if they are self-powered, and you would let the global object composed of those small objects take care of itself.
For a human (or a chair, for that matter), you want some basic rules, like: "this is how you stand upright!". And you would control the action by issuing commands, like: "walk that way!".
What would be the simplest way to make something like that possible?
I'm thinking about just using an ordinary bone structure and mesh at first, and just adding the composite objects step by step. So, first the muscles, and a script interface to be able to control them. Then the curvature, and a way to bulge the meshes. And then a sort of second skeleton, that consists of points that the external forces act upon, extended into the other objects around it, like the clothes and hair. Etc.
Would that work?
For a human (or a chair, for that matter), you want some basic rules, like: "this is how you stand upright!". And you would control the action by issuing commands, like: "walk that way!".
What would be the simplest way to make something like that possible?
I'm thinking about just using an ordinary bone structure and mesh at first, and just adding the composite objects step by step. So, first the muscles, and a script interface to be able to control them. Then the curvature, and a way to bulge the meshes. And then a sort of second skeleton, that consists of points that the external forces act upon, extended into the other objects around it, like the clothes and hair. Etc.
Would that work?
Both dynamic human motion synthesis and fully dynamic muscle deformation are subjects of study by inumerous universities out there. The fact is, that we're not there yet, even for non-realtime processing, let alone for games.
I think the PPU will allow game creators to have more flexibility and create stunning dynamic environments that we haven't had in the past. Look how static the gaming enviroments are now. You can't blow up buildings or make craters in the ground, etc.
Both dynamic human motion synthesis and fully dynamic muscle deformation are subjects of study by inumerous universities out there. The fact is, that we're not there yet, even for non-realtime processing, let alone for games.
Yes, that's why we would need those hacks. Like with most other things in rendering graphics: we can't really do it exactly as it should be done, even if we knew how, but we make an approximation that works good enough.
Chalnoth
12-Apr-2005, 21:25
Well, I"ll just respond to the original post:
Keep in mind that adding different sorts of simulations to the game will require different amounts of processing power. What you see today is basically the easy stuff: rigid body dynamics.
What you'll see in the immediate future, then, is the expansion of rigid body dynamics. You'll see more movable objects, and breakable objects made of many pieces. After this, you'll start to see some soft-body dynamics. These are a bit harder to do, but still fundamentally related to rigid body dynamics, and thus don't require a dramatic change. This just means that you'll start to see the possibility of bouncing boobies (from physics, not from pre-baked animations), blunt weapons that actually crush flesh, or anything else that would constitute a continuously-deformable object.
Further into the future it gets more interesting: fluid dynamics. Fluid dynamics really are a challenging problem, but would also be essential to the immersion of games. The most obvious example of how fluid dynamics would work in a game is smoke. Wouldn't it be fantastic to see smoke in a game that billows and churns and moves around objects realistically?
Water is also a problem that has never come close to being solved realistically in games. It'd be nice to see water surface deformation that's actually realistic. But this will likely also be a little ways off.
When 3d acceleration first came about, we had games that offered both software and hardware rendering modes to bridge the gap. In what way could a game offer both hardware(complex) and software(simple) physics modes where the complex mode isn't useless fluff, and at the same time the simple mode isn't missing out on too much of the potential game.
Club3D says it'll be European manufacturer for Aegia PPUs this winter
http://forums.hexus.net/showthread.php?t=43060
Also Crytek appears to have signed up.
Jawed
scificube
14-Apr-2005, 20:45
Club3D says it'll be European manufacturer for Aegia PPUs this winter
http://forums.hexus.net/showthread.php?t=43060
Also Crytek appears to have signed up.
Jawed
I...
I am happy.
Unknown Soldier
15-Apr-2005, 07:19
AGEIA Technologies Interview
In the last few years, we have seen collision detection and rigid body physics, with Half-Life 2 representing the pinnacle of software accelerated physics; featuring between 10 and 100 objects interacting on screen. How different will games with hardware accelerated physics be? Do you think that the PhysX PPU will revolutionize gaming?
Manju Hegde: We believe PhysX will revive PC gaming innovation and that AGEIA’s technology will re-ignite the passions and creativity of game developers who can now integrate dramatic game play for a totally immersive environment.
PhysX accelerates physics-related calculations in games. Its unprecedented power to perform advanced real-time physics simulation enables deeper player-environment interaction and immersive, living detail.
With the PPU, car fenders will at last crumple with satisfying detail, as the gamer slides his car head-on into a wall, thick with clouds of dust, curls of smoke and shards of glass one expects from being so close to the action.
In the short term, perhaps the most significant enhancement will relate to game effects, where environmental and other factors currently unavailable to game developers can easily be incorporated. As the wind blows through a lonely game world, trees will bend and sway, leaves will blow and scatter, with shutters banging loudly and curtains blowing in the breeze. When the monk walks through the catacombs, smoking torch in hand, his flowing robes will drape realistically over his legs; and the heroine’s pony tail will blow in the wind as she rides her mount across the landscape. As the castle explodes, the spray of rubble will rupture the walls on the surrounding streets. These subtle yet essential effects will add a completely new layer of visual realism to games, immediately boosting immersion. These physical simulation based visualizations will be altered by inputs and actions and so will substantially greater variety and richness than canned animations.
Soon thereafter, physics will work its way into gameplay itself. Perhaps the blowing breeze exposes hidden soldiers behind the branches, while the monk’s shifting robes reveals a knife or a hidden letter. The number of active characters in a scene can jump from a handful to dozens or even hundreds, each with its own unique physical and logical characteristics.
I have no doubt physics are more important to a game than graphics, since they are far beyond eye candy; they have a true impact on gameplay. But how will gamers “see” this? You know, both ATI and nVIDIA develop special demos to showcase their products. When can we expect the first demos and movies that showcase the difference between non-hardware accelerated physics and PhysX-powered games?
Manju Hegde: Over the next couple of months you will be able to see both special demos as well as game demos from PhysX powered games that will be available during the 2005 Christmas season.
Regarding your business plan, would you make your own PhysX cards and also sell the chips to card manufacturers, ala ATI, or would you only sell the PPUs for others to build the cards like nVIDIA does?
Manju Hegde: As mentioned above, we see the add-in market as just one of the channels to bring this technology to market. We are now working with the high end OEMs, Volume OEMs, motherboards manufactures, white box distributors, laptop manufactures, Express Card and many others to bring the technology to the gamers. We see this as critical in making our game developers and publisher partners successful by driving the chip out to the wide market. Our plan is to sell the PhysX chip into any device that can benefit from physical simulation applications.
We have been meeting with a number of the top tier companies in the OEM, ODM and board manufacturing space and plan on making announcements shortly.
Can you reveal (of course without telling us who the company is) if the PhysX PPU will be in any form inside at least one of the next generation consoles, or was it too late in the process to contribute to these upcoming products?
Manju Hegde: On this one, you will have to wait.
Read More: TeamXBox (http://interviews.teamxbox.com/xbox/1117/AGEIA-Technologies-Interview/p1)
scificube
15-Apr-2005, 17:33
You do know no one likes a tease right...oh yeah...well...I still don't like being teased.
I think this would mean a yes or it's still possible at this point because if this weren't the case why would he just say no?
It would rock so hard for the consoles to have a PPU IMO and it would seem that at a minimum this may become the reality of things.
I think these guys are doing well and playing their cards right thus far. They simply need to get some demos up and into view and I think they will really get smoking.
trinibwoy
15-Apr-2005, 19:27
I'm not sure if anyone has already touched on this but is the synchronization of CPU and PPU going to be a difficult task? I'm assuming the CPU will need to have the output of the PPU before processing data for the next frame.
interesting I would think,
the ai is done by the CPU then the cpu tells the PPU what animations are done, then the PPU does its thing then sends that data to the GPU for the final results,
For collision cpu, if collision is detected then sends info to PPU and that calculates the collisions, then the end results are set back to the cpu.
nutball
15-Apr-2005, 21:18
I'm assuming the CPU will need to have the output of the PPU before processing data for the next frame.
Maybe not if the PPU can send the geometry direct to the GPU, and the GPU is PPU-aware and can digest it directly. Synchronisation is certainly an issue, but the looser the better performance-wise I think. The whole situation is starting to smack of MPI over PCIe. Which could be interesting.
trinibwoy
15-Apr-2005, 22:38
I'm assuming the CPU will need to have the output of the PPU before processing data for the next frame.
Maybe not if the PPU can send the geometry direct to the GPU, and the GPU is PPU-aware and can digest it directly. Synchronisation is certainly an issue, but the looser the better performance-wise I think. The whole situation is starting to smack of MPI over PCIe. Which could be interesting.
I was more thinking along the lines of AI and other routines (sound is one example) reacting to the result of physical interactions. The CPU would need to have a good picture of the state of the environment if a true sense of immersion is going to be achieved.
For example, if there is a ricochet - the CPU does not know any details of the deflection until the PPU does it's thing so there needs to be some communication back to the CPU to update state.
Sandwich
16-Apr-2005, 10:56
I'm assuming the CPU will need to have the output of the PPU before processing data for the next frame.
Maybe not if the PPU can send the geometry direct to the GPU, and the GPU is PPU-aware and can digest it directly. Synchronisation is certainly an issue, but the looser the better performance-wise I think. The whole situation is starting to smack of MPI over PCIe. Which could be interesting.
The suggestion isn't entirely unreasonable. Maybe you could have PPUs and GPUs on one card in the future.
Until PPUs are common stuff, the only thing a PPU should be used for is eye-candy. You simply cannot have the AI or the game mechanics rely on results from the PPU, because games will also need to run on PCs without PPUs.
Results must also be consistent with those from the CPU. I suppose that's where the novodex api comes in, but PPUs will typically sit in the high-end rigs of the enthousiast anyway. There's no point to take a load of the cpu, unless the game always runs like crap without a PPU.
There's plenty PPU potential that doesn't have anything to do with gameplay, like long cloaks that don't have clipping issues with doorways or stairs.
trinibwoy
16-Apr-2005, 14:41
The suggestion isn't entirely unreasonable. Maybe you could have PPUs and GPUs on one card in the future.
Until PPUs are common stuff, the only thing a PPU should be used for is eye-candy. You simply cannot have the AI or the game mechanics rely on results from the PPU, because games will also need to run on PCs without PPUs.
Results must also be consistent with those from the CPU. I suppose that's where the novodex api comes in, but PPUs will typically sit in the high-end rigs of the enthousiast anyway. There's no point to take a load of the cpu, unless the game always runs like crap without a PPU.
There's plenty PPU potential that doesn't have anything to do with gameplay, like long cloaks that don't have clipping issues with doorways or stairs.
I don't think it's possible to limit PPU output to eye-candy since a lot of physics (at least in current games) involves collisions and collisions generate sound. So until the PPU and/or GPU has a built in sound-card the CPU has to know what's going on after the PPU does its thing.
Depending on the design it may be possible to guarantee very similar results between a software and hardware implementation of the Novodex API so machines without PPU's (and powerful CPU's) may not be so disadvantaged.
I think most (rigid body) physics API's are state machines triggered on the collision detection. If you want to off-load physics, you need to off-load that as well. And if you want to be able to calculate things that are close to one another, you get problems if you only calculate and trigger collisions, as you can have multiple objects colliding multiple times with one another before the new state is satisfactory. Otherwise, they might clip. So you would need to off-load the whole scene management.
For volumes, there is so much interaction while calculating the meshes, that you can't really have the CPU manage them, I think.
vBulletin® v3.8.6, Copyright ©2000-2013, Jelsoft Enterprises Ltd.