So after reading the thread about why a CPU can't calculate Physics for nearly the same amount of objects that a PPU can, I was wondering; How efficient is a GPU when compared to a PPU? I can't imagine it being as powerful as a PPU, even with a 8800GTX calculating the physics.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Difference between Physics on a GPU and Physics on a PPU?
- Thread starter xFlankerx
- Start date
One would say that a PPU should be quicker with physics than a GPU, since it's obviously designed strictly for that. Of course though, we're comparing Ageia to much bigger companies like AMD and nvidia whom are able to stay at the bleeding edge of technology. Ageia is probably always going to be several steps behind in some respects, simply because they can not afford the same R&D costs.
Let me add another question to my initial question of how powerful a GPU is in calculating physics;
The GeForce 8800GTX is advertising something called "Quantum Physics." I read in the FAQ that a GPU can only calculate one at a time; graphics, or physics. How is Nvidia going about calculating Physics while also displaying graphics?
The GeForce 8800GTX is advertising something called "Quantum Physics." I read in the FAQ that a GPU can only calculate one at a time; graphics, or physics. How is Nvidia going about calculating Physics while also displaying graphics?
Good question. Unified cores. Nvidia and Ati are going to setup there cards so they dedicate "units" of the unified core to physics while the other "units" work on the graphics. Another method for simulating physics on GPUs is having 1 card do graphics while another does physics. Kinda like crossfire/sli but only 1 card does graphics while 1 does physics or having 2 cards do graphics while 1 does physics. Also GPUs are much much more powerful than PPUs. PPUs can do a lot of floating point calculations. Same with GPUs but GPUs have been refined a lot more times than PPUs so far.xFlankerx said:
I haven't been around the PC scene in a long time. So I cant remember the name for the "units" and I cant remember if floating point calculations is the right words/meaning. But you should get the idea.
Ah, I see about the Unified Cores. I actually read on that, but I guess I couldn't connect the dots, thanks for explaining that. As for the GPU - So a GPU that is only calculating Physics would be as powerful as, and maybe more powerful than, a PPU doing the same thing?
GPU use the Shaders for general computing. Because shaders are more general purpouse each next generation. Wich behave more like Coprocesors. From Sm3.0 also good for Physics. In raw Performance High-end GPU are more or less way powerfull then PPU. That what I expects.
PPU has Patented logic to do Physics. I can only asume it efficcent for Physics.
But they put it on 130nm with only a 125Mil of transisiors.
While G80 have almost 700Mil transistors. Where a large part are the shaders.
128 of them. Even if used a quarter of that it could be just enough to compete with PPU.
But nV have to compete also with ATI on this.
That G80 on 90nm. and a 65nm on the Way G81.
R600 on 80nm I think.
The problem is PPU is a light design on a old producktion procses. That hit it hard twice.
Where PPU doesn't compete with CPU. the first the CPU will catch up with is the PPU.
65nm 45 nm vs 130nm chip.
Ageia better skip 90nm and put a 500M PPU2 on 65nm
.
Because it going to compete with GPU in the near future not only CPU.
Need a PPU2 now then over 1 or 2 years.
Just like AMD at 90nm have a hard time to compete against iNtel 65 C2D.
PPU has Patented logic to do Physics. I can only asume it efficcent for Physics.
But they put it on 130nm with only a 125Mil of transisiors.
While G80 have almost 700Mil transistors. Where a large part are the shaders.
128 of them. Even if used a quarter of that it could be just enough to compete with PPU.
But nV have to compete also with ATI on this.
That G80 on 90nm. and a 65nm on the Way G81.
R600 on 80nm I think.
The problem is PPU is a light design on a old producktion procses. That hit it hard twice.
Where PPU doesn't compete with CPU. the first the CPU will catch up with is the PPU.
65nm 45 nm vs 130nm chip.
Ageia better skip 90nm and put a 500M PPU2 on 65nm
.Because it going to compete with GPU in the near future not only CPU.
Need a PPU2 now then over 1 or 2 years.
Just like AMD at 90nm have a hard time to compete against iNtel 65 C2D.
Thanks, lots of good info. But about the CPU catching up with the PPU -- Didn't the thread about why a CPU cannot currently compete with a PPU mention that a CPU could only handle like 2,000+ objects per core while the PPU could handle 30,000+. Either those numbers have been pulled out of someone's ass, or it seems like it would take the CPU a while to catch up. I know the CPU is a very general-purpose processor, so I don't personally think that it is capable of the FLOPS (am I using this wrong?) to compete with the GPU or PPU.
xFlankerx said:
Those numbers are estimates that should be pretty close. But keep in mind that Intel hopes to have 64-core CPUs by 2010. At that rate of expansion, the CPU could catch up in the next few years.
As for the GPU vs PPU, the G80 has changed some things. At first I thought that the PPU must still have better performance than the G80 since it is a dedicated physics chip. But, I realize that doesn't really make sense. A 300M 90nm chip with 128MB of memory on the PCI bus beating a 650M 65nm chip with 640MB of memory on the PCI-E X16 bus? Most likely, with all 128 cores dedicated to physics, the 8800GTX will beat the current-gen PPU. On the other hand, with only, say, 20 cores dedicated to physics (and the rest to graphics), the PPU would probably beat it. But it doesn't matter much anyway. Physics at the level they will be in the near-future won't max out either card.
SuperGee said:GPU use the Shaders for general computing. Because shaders are more general purpouse each next generation. Wich behave more like Coprocesors. From Sm3.0 also good for Physics. In raw Performance High-end GPU are more or less way powerfull then PPU. That what I expects.
PPU has Patented logic to do Physics. I can only asume it efficcent for Physics.
But they put it on 130nm with only a 125Mil of transisiors.
While G80 have almost 700Mil transistors. Where a large part are the shaders.
128 of them. Even if used a quarter of that it could be just enough to compete with PPU.
But nV have to compete also with ATI on this.
That G80 on 90nm. and a 65nm on the Way G81.
R600 on 80nm I think.
The problem is PPU is a light design on a old producktion procses. That hit it hard twice.
Where PPU doesn't compete with CPU. the first the CPU will catch up with is the PPU.
65nm 45 nm vs 130nm chip.
Ageia better skip 90nm and put a 500M PPU2 on 65nm
Because it going to compete with GPU in the near future not only CPU.
Need a PPU2 now then over 1 or 2 years.
Just like AMD at 90nm have a hard time to compete against iNtel 65 C2D.
The G80 doesn't have "shaders" exactly. It has 128 unified streaming processors, which are capable of doing vertex, pixel or geometry computations.
The pages here explains more about it.
http://enthusiast.hardocp.com/article.html?art=MTIxOCw1LCxoZW50aHVzaWFzdA==
http://enthusiast.hardocp.com/article.html?art=MTIxOCw4LCxoZW50aHVzaWFzdA==
Really, it is hard to tell how good the PPU is actually at physics calculations since there is no standard physics API that I know of. Therefore, comparing ATi, nVidia and Ageia's PhysX solutions at such a task is going to be difficult, if not impossible. If one game supports all three types of physics acceleration, then it would come down to implementation within the game or game engine. No doubt if more development time is spent on the Ageia solution, it might very well work better in a given game than ATi's solution would.
Plus, with there being a wide range of ATi and nVidia cards out there, it would depend specifically on the model of card being used for the physics calculations.
Sounds good.
So what are the chances of the G80 being used in the same fashion Physics-calculations-wise as the PhysX card? Because it sounds like people who have a 8800GTX don't really need to buy a PhysX card then. So will games like UT2K7 run the same Physics on a G80 as on PhysX?
So what are the chances of the G80 being used in the same fashion Physics-calculations-wise as the PhysX card? Because it sounds like people who have a 8800GTX don't really need to buy a PhysX card then. So will games like UT2K7 run the same Physics on a G80 as on PhysX?
xFlankerx said:Sounds good.
So what are the chances of the G80 being used in the same fashion Physics-calculations-wise as the PhysX card? Because it sounds like people who have a 8800GTX don't really need to buy a PhysX card then. So will games like UT2K7 run the same Physics on a G80 as on PhysX?
I have no idea. If I were to guess based on raw power, I'd say the 8800GTX would wipe the floor with a PhysX card. Really though, I have no idea if that's really the case. The Ageia solution might be uber effecient. I really can't say. Plus, I don't know how good the G80 is at physics calculations.
Right now only time will tell.
I think a PPU would wipe the floor with the G80 in UT2K7.
Because if not detecting PPU it put the Phsyics setting on low.
Unless it supports a second API besides PhysX. and that would be most likly Havok FX.
So far I know, it doesn't.
That's Possible Graw is a Havok(CPU) PhysX(PPU) game. But don't support GPU for physics.
Most Game will support one Physics API so they make a choice.
For real game bechin it would be nice to have a game supporting both.
Because if not detecting PPU it put the Phsyics setting on low.
Unless it supports a second API besides PhysX. and that would be most likly Havok FX.
So far I know, it doesn't.
That's Possible Graw is a Havok(CPU) PhysX(PPU) game. But don't support GPU for physics.
Most Game will support one Physics API so they make a choice.
For real game bechin it would be nice to have a game supporting both.
Like SuperGee said, until we get a standard physics API, the raw power of a card has no bearing on how it will perform in games. For now it's:
Game made for ATI physics: ATI cards blow away the other solutions, which don't do anything
Game made for NVIDIA physics: NVIDIA cards blow away the other solutions, which don't do anything
Game made for AGEIA physics: AGEIA cards blow away the other solutions, which don't do anything.
Granted, NVIDIA and ATI physics games don't actually exist yet, ATI's physics solution isn't even available, and AGIEA's solution only havs a few games that don't do much with it, so that's another reason why we can't compare the solutions right now.
All we know is that AGEIA's solution requires an inexpensive card, and NVIDIA's solution requires no card, but you lose graphics performance, or a very expensive card.
Game made for ATI physics: ATI cards blow away the other solutions, which don't do anything
Game made for NVIDIA physics: NVIDIA cards blow away the other solutions, which don't do anything
Game made for AGEIA physics: AGEIA cards blow away the other solutions, which don't do anything.
Granted, NVIDIA and ATI physics games don't actually exist yet, ATI's physics solution isn't even available, and AGIEA's solution only havs a few games that don't do much with it, so that's another reason why we can't compare the solutions right now.
All we know is that AGEIA's solution requires an inexpensive card, and NVIDIA's solution requires no card, but you lose graphics performance, or a very expensive card.
I'm no computer engineer or anything, so I leave this open to SuperGee (who is, or who at least sounds like one).
Cant raw processing power be rendered irrelevant by the architecture of the chip? Isn't there a big difference between the banks of floating point registers in the PPU and the psuedo-web structure a GPU uses?
I mean...sound cards aren't beefy processors by any means, but a GPU would suck at doing an X-fi's job right?
Cant raw processing power be rendered irrelevant by the architecture of the chip? Isn't there a big difference between the banks of floating point registers in the PPU and the psuedo-web structure a GPU uses?
I mean...sound cards aren't beefy processors by any means, but a GPU would suck at doing an X-fi's job right?
HavokFX is quite close to a standard api, in that anyone can support it by writing a glsl (or hlsl) implementation for it afaik.
This is a somewhat convoluted way of doing things in general, but the interface between HavokFX and the hardware in itself is an open standard. We'll certainly be able to compare different brand video cards with HavokFX.
This is a somewhat convoluted way of doing things in general, but the interface between HavokFX and the hardware in itself is an open standard. We'll certainly be able to compare different brand video cards with HavokFX.
While I am a large fan of the PPU idea, I think its important to think about this:xFlankerx said:
What exactly is a "physics" calculation? The calculating the graviational force between two or more objects? The electromagnetic force between two charged particles? The resulting movement when two objects collide nonelastically?
Well, I think it's pretty clear to anyone who has taken a high-school physics class that physics is a very, very large field of science. As such there is no "physics" calculation in general.
As such, the more interesting thing that Ageia is bringing to the table is an API for developers that ought to be able to simplify designing games with a lot of interactivity. One would hope that the engineers at Ageia designed their hardware to excel in the computations that are more frequently used by their API. As such, it'd be surprising to see a GPU solution be better at those than the Physx.
Let's look at a real-world problem on the other hand: Folding at home. It's a program that likely falls into the realm of physical biology, since it uses "physics" calculations (forces) to try and simulate how proteins fold. If you look at this thread you will notice that the group surrounding this thinks that the GPU is currently better suited for their needs than the PPU. However, it's important to note that this is not necessarily because a GPU is 'better', but rather because it is currently more versatile, which makes developing for it easier and is likely to produce better code.
Here are the cliff notes(tm):
1. There is no such thing as "a physics calculation" since physics engulfes a large field. It is likely that a lot of these calculations require floating point processing power.
2. Often times, it is not the raw power of a solution that makes it more feasible for a particular purpose, but rather the ease of developing for it and how much flexibility it allows the programmer.
freddiepm61
Gawd
- Joined
- Oct 19, 2004
- Messages
- 836
Nvidia and Ati physics though both implemented differently(though not that differently) will both work in the same situations as they both use the Havok physics API.
drizzt81 said:
Well, as for 1), all modern CPUs and GPUs have reasonably efficient floating point vector processing, with a reasonably generic programming model.
The bottleneck with physics is not really the calculations themselves, as they tend to be rather simple, but mostly the fact that the results have to be forwarded from one unit to the next quickly, since physics objects influence eachother.
The key to the PhysX chip is an intricate switching network that makes this possible.
I wonder how well GPUs will do, because they're generally designed to do a lot of operations in parallel, where you get the input from one place, and write the output to another, without any kind of influence in between.
Unless NV and ATi did something smart, I guess they'll just be doing a LOT of render-passes. Each pass will solve one iteration of the physics system.
In which case the dependencies of the physics system will dictate how efficient a GPU is at physics.
As for CPUs... they are very flexible, but they aren't very parallel, so while they can efficiently move data from one processing unit to the next, there are only a few processing units to work with.
As for 2), I don't think this will be much of a problem really. PhysX uses the reasonably common NovodeX library... The rest uses HavoK, afaik, which is the other big name.
And there are rumours of a physics component for DirectX. That would solve all programming issues right away.
So the actual physics solution will probably be transparent to the programmer. NovodeX at least works in software or PhysX-mode, without having to change code.
From what I understood, the GPU-physics will currently not support all features yet... But I'm not sure if that means that the features will be disabled, or just automatic fallback to software processing.
Dan_D said:
its a bit more then that with the way SM40 works and the unified design the thing can realy run ANY code you can write for the more parallel the code can be run the faster it will be run on the GPU then a CPU
physics lends it self to this well since you can have objects calculated individually at the same time or different forces on the same object at the same time
the CUDA language for the G80 is a big step to the GPU becoming nothing more then a VERY fast FPU
btw if you dont know CUDA is basically C for the G80
http://developer.nvidia.com/object/cuda.html
Yes while we can play and buy ageia's PhysX games and hardware now.
DirectX physics was in a early open stage. The news about it a while back, was that MS has a job open for a Phygisist. Maybe they have one now. And are lookin into it, what to do with the next version, interim or full DirecX update.
Then there comes a time that dev's start using that fresh Dirext PhysX and then some day DirecX physX games are popping out. By then 35nm Hardware will be out or 45nm. Like 2010. So far I way and so in the open. The have to deside what and how they are exactly gonna support with it. So Nice to know way later, but for the near future not so relevant. For the far future that is nice to know someday there will be a standard.
Nice for Developers espacialy the small once, not so relevant for games.
For them it's important what devs do with what ever solution.
First we must see HavokFX games.
Quantum and stream games sound more like ATI and nV specific Physics Paths and may not be so adoptive by dev's because it leaves out the competitors GPU hardware.
I think most dev go for the GPU standard Havok FX.
Also it's the same thing. Going to throw them on the GPU Physics API heap.
So i wonder is there a havok FX game coming out soon in the near future and wich one is it.
DirectX physics was in a early open stage. The news about it a while back, was that MS has a job open for a Phygisist. Maybe they have one now. And are lookin into it, what to do with the next version, interim or full DirecX update.
Then there comes a time that dev's start using that fresh Dirext PhysX and then some day DirecX physX games are popping out. By then 35nm Hardware will be out or 45nm. Like 2010. So far I way and so in the open. The have to deside what and how they are exactly gonna support with it. So Nice to know way later, but for the near future not so relevant. For the far future that is nice to know someday there will be a standard.
Nice for Developers espacialy the small once, not so relevant for games.
For them it's important what devs do with what ever solution.
First we must see HavokFX games.
Quantum and stream games sound more like ATI and nV specific Physics Paths and may not be so adoptive by dev's because it leaves out the competitors GPU hardware.
I think most dev go for the GPU standard Havok FX.
Also it's the same thing. Going to throw them on the GPU Physics API heap.
So i wonder is there a havok FX game coming out soon in the near future and wich one is it.
Well CF: revolution has a estimated releasdate in December. And that's close.
Well I'am very currious what this game could mean for ageia sucses for PPU from the gamers side.
nV do well at the moment with marketing and tech demo's. To influence gamers opinions.
A PhysX rich game would give Ageia some more possitive attention to the market.
My expectation about CFR are moderate. It's not a Class AAA title.
But could give a nice surprise hopefully.
Well I'am very currious what this game could mean for ageia sucses for PPU from the gamers side.
nV do well at the moment with marketing and tech demo's. To influence gamers opinions.
A PhysX rich game would give Ageia some more possitive attention to the market.
My expectation about CFR are moderate. It's not a Class AAA title.
But could give a nice surprise hopefully.
NightRaven
[H]ard|Gawd
- Joined
- Feb 20, 2003
- Messages
- 1,359
Without a unified physics API, I don't really think the industry is really going to build up any significant market in physics. Having separate incompatible solutions isn't exactly encouraging for software developers. Having a single MS-lead standard will certainly help, and make it far more accessible, especially if it can also be used to do physics processing on spare CPU cores.
GPUs and PPUs are, at their heart, not all that different. They both deal extensively with floating point vector operations, one mostly with lighting and optical effects and the other with masses and forces. As is typical with newer technology, they both started out more specialized for their particular task. As GPUs become essentially advanced vector processors (at which point I hope Ray-Tracing takes off), physics APIs for them will also become much simpler and more efficient, and I wouldn't be surprised if the PPU as a distinct entity just vanishes
But who knows, as CPUs become increasingly parallelized and better integrated, there may eventually be no need for separate vector processors at all.
GPUs and PPUs are, at their heart, not all that different. They both deal extensively with floating point vector operations, one mostly with lighting and optical effects and the other with masses and forces. As is typical with newer technology, they both started out more specialized for their particular task. As GPUs become essentially advanced vector processors (at which point I hope Ray-Tracing takes off), physics APIs for them will also become much simpler and more efficient, and I wouldn't be surprised if the PPU as a distinct entity just vanishes
But who knows, as CPUs become increasingly parallelized and better integrated, there may eventually be no need for separate vector processors at all.
=drizzt81 said:
I would agree with you to a certain extent but I rationalize the calculations on a P.P.U. differently. Your defintion of molecular collision is wonderful but lets be real. When it comes to video games the primary factory in calculating object movement in general would be gravity, force, and mass. Or for anyone who doesnt know F=ma or force equals mass x acceleration.
With this said I dont believe that either gpu or a ppu could really outperform a cpu . The reason being is that each individual item would have to be assigned several values for which real time calculations could be made. Based on that wouldnt the most likely winner of all of the processing units be the one that had the highest real time number crunching ability? I think so, it just sounds a little more rational. Ideally, I would l love it if either Intel or AMD came up with their own ppu solution that allowed you to alter its logic chip. ( Like having a x6800 on ur mobo and a xeon on a secondary card that does only computations! THAT WOULD ROCK!)
CubicleGeek
Limp Gawd
- Joined
- Jan 13, 2006
- Messages
- 292
It would be difficult for general purpose processor like a CPU to outperform any specialized processor, even one running at an 8th of it's clock rate. A general purpose processor will provide all of the basic instructions required to perform any range of operations, but not optimize for special cases. Using the graphics card example, the reason why a GPU can outperform a CPU for rendering doesn't have anything to do with whether or not the CPU can do it, because it can. The GPU can do among other things the matrix manipulations required for vertex transformations in far fewer clock cycles than a CPU can, because it has logic circuits available specifically for that task. Tasks like vertex transformation, that occur all the time and take multiple clock cycles can be pipelined with a very reasonable assumption that the pipeline will always be (near) full.
I'm certainly no physicist and can hardly recall much of what I've learnt in my physics classes but if there are specific operations that can be optimized or specialized and certain patterns for physics operations used in gaming that can take advantage of tricks like pipelining you can design a chip that runs at low clock speeds that can output the values that you want at a greater frequency than any significantly higher speed general purpose processor.
It's not only about number crunching ability, but the ability to crunch the type of numbers that we want. All processors of all types do is number crunching.
CubicleGeek;1030442954]It would be difficult for general purpose processor like a CPU to outperform any specialized processor, even one running at an 8th of it's clock rate. A general purpose processor will provide all of the basic instructions required to perform any range of operations, but not optimize for special cases. Using the graphics card example, the reason why a GPU can outperform a CPU for rendering doesn't have anything to do with whether or not the CPU can do it, because it can. The GPU can do among other things the matrix manipulations required for vertex transformations in far fewer clock cycles than a CPU can, because it has logic circuits available specifically for that task. Tasks like vertex transformation, that occur all the time and take multiple clock cycles can be pipelined with a very reasonable assumption that the pipeline will always be (near) full.
I'm certainly no physicist and can hardly recall much of what I've learnt in my physics classes but if there are specific operations that can be optimized or specialized and certain patterns for physics operations used in gaming that can take advantage of tricks like pipelining you can design a chip that runs at low clock speeds that can output the values that you want at a greater frequency than any significantly higher speed general purpose processor.
I'm certainly no physicist and can hardly recall much of what I've learnt in my physics classes but if there are specific operations that can be optimized or specialized and certain patterns for physics operations used in gaming that can take advantage of tricks like pipelining you can design a chip that runs at low clock speeds that can output the values that you want at a greater frequency than any significantly higher speed general purpose processor.
It's not only about number crunching ability said:
Efficient compared to?
Power (watts)?
Price (money)?
Cycles (Hz)
Naturally the PPU is built specifically for calculating certain things and so can do it quite quickly, a CPU might be less efficient in that it might require more clock cycles to perfom just 1 set of calculations but generally speaking they're faster to begin with.
I dont think power is an issue to anyone, large wattage PSU's are cheap, additionally no one really cares about efficiency per clock providing theres some balance that gets the work done, it's all about price. I think that if you take the price of the PPU and add that onto what you'd usualy spend on a CPU you'd get a better overall deal
My old single core 64 4000 @ 2.7Ghz could chew through CellFactor with no PPU installed @ 1280 960 running on average 30fps, only the liquid physics and cloth physics were missing. With my new E6600 @ 3.0 Ghz I think any dual core optimised version of the demos would run pretty sweet.
Power (watts)?
Price (money)?
Cycles (Hz)
Naturally the PPU is built specifically for calculating certain things and so can do it quite quickly, a CPU might be less efficient in that it might require more clock cycles to perfom just 1 set of calculations but generally speaking they're faster to begin with.
I dont think power is an issue to anyone, large wattage PSU's are cheap, additionally no one really cares about efficiency per clock providing theres some balance that gets the work done, it's all about price. I think that if you take the price of the PPU and add that onto what you'd usualy spend on a CPU you'd get a better overall deal
My old single core 64 4000 @ 2.7Ghz could chew through CellFactor with no PPU installed @ 1280 960 running on average 30fps, only the liquid physics and cloth physics were missing. With my new E6600 @ 3.0 Ghz I think any dual core optimised version of the demos would run pretty sweet.
Well I play it with Cloth and Fluids on. smooth fps. With a PPU.
Because Revolution will have more PhysX in it. There will probaly be a larger difference in PPU non PPU Play & eyecandy compared to Combat training.
Caffeinated
[H]ard|Gawd
- Joined
- Oct 16, 2002
- Messages
- 1,451
So far, they should be called Placebo Processing Units.
I don't think anyone is doubting what the cards are capable of doing. It's not like they are useless, they just aren't being used that much right now. Certainly none of the titles out right now are worth having it for. That's just my opinion and the opinion of many people that have posted in this thread.
So your card can do alot of this stuff really well in theory. That's nice. Does it help in UT2004, Quake 4, CS:S, HL2, HL2 Episode One, FEAR, FEAR Extraction Point, BF2, BF 2142, or any of the popular games right now? The most popular game I can think of that the card gets used in is GRAW, and I'm sorry, but what the PhysX card does in GRAW isn't that spectacular. At least not $200.00+ worth of spectacular.
Caffeinated
[H]ard|Gawd
- Joined
- Oct 16, 2002
- Messages
- 1,451
This is the key point right here.
The biggest difference between PPU vs. GPU physics (other than people not owning a PPU) comes down to using actual game physics vs. "fluff" physics. You can utilize the PPU to actually do a lot of your real game physics if you want to such as collision, etc, and things that actually effect the gameplay. Using physics on the GPU readbacks are insanely expensive and you would never think of putting actual game critical physics functionality on the GPU.
Although I don't think having critical physics on the PPU will ever be viable even if everybody started running out tomorrow to buy them, or even if it is integrated into "gamer" motherboards. If certain large scale physics calculations are critical to your game play (an example would be say boats attacking each other where the actual physics of the water effects the gameplay) than people without a PPU will be unable to play it. Especially now that dual and quad core systems will become mainstream over the next couple of years and they can handle a good amount of physics calculations while offloading the fluff to the GPU.
Although I don't think having critical physics on the PPU will ever be viable even if everybody started running out tomorrow to buy them, or even if it is integrated into "gamer" motherboards. If certain large scale physics calculations are critical to your game play (an example would be say boats attacking each other where the actual physics of the water effects the gameplay) than people without a PPU will be unable to play it. Especially now that dual and quad core systems will become mainstream over the next couple of years and they can handle a good amount of physics calculations while offloading the fluff to the GPU.
Games getting more complex. And Physics add to this complexity.
Because fideling with game play have a hard impac on Dependancy in that complexity. Game mechanics. It something Dev's don like to burn on. So it's more a Requierment Feature risk assesment. Go for it or not.
Effec Physics is the save way. light impact on game design, something that can be taged on if there is some spare time or wishes from publisher.
While Gameplay Physics have a hard impact on gameplay and net load. It can trow the balance completly over. So some hurdles Dev's must go headon with. If they feel brave and want to make the next best thing and take the risk to make a better and Physics rich game wich is possible with the tech now.
Well The Cellfactor dev team focus on this Gameplay Phsyics
As for GPU there are some flavors. and all with it's restriction and strong points.
you got nV SM3.o wich sucks at Physics in Performance way. Maybe even in write back.
ATI SM3.0 differs in being better at Physics then nV counter part. Write back would be possible the GPU firms claim and havok to in that old quad interview.
nV SM.4.0 nV is for a long time aware of ageia plans. maybe the Sm4.0 shaders are so general purpouse they can write back easier in a basic way. It can handle Physics well.
ATI looks like the have a headstart with Phsyics on GPU with there SM3.0 solution. So I expect that R600 shader tech would be more like a second generation of GPGPU while nV sm3.0 is a inbetween step. And G80 there realy full first attempt.
ATI plans to come with a 2GB GPGPU card. Looks like they where planning it for some time.
Imangine a R600 on 80nm with 2GB for local Physics memory.
VS Ageia PPU on 130nm and a limit of 128MB
I think Ageia needs a PPU2 at least on 90nm or smaller with 1GB or so to compete in the near future. a 400Mil or more transistor beast
Because fideling with game play have a hard impac on Dependancy in that complexity. Game mechanics. It something Dev's don like to burn on. So it's more a Requierment Feature risk assesment. Go for it or not.
Effec Physics is the save way. light impact on game design, something that can be taged on if there is some spare time or wishes from publisher.
While Gameplay Physics have a hard impact on gameplay and net load. It can trow the balance completly over. So some hurdles Dev's must go headon with. If they feel brave and want to make the next best thing and take the risk to make a better and Physics rich game wich is possible with the tech now.
Well The Cellfactor dev team focus on this Gameplay Phsyics
As for GPU there are some flavors. and all with it's restriction and strong points.
you got nV SM3.o wich sucks at Physics in Performance way. Maybe even in write back.
ATI SM3.0 differs in being better at Physics then nV counter part. Write back would be possible the GPU firms claim and havok to in that old quad interview.
nV SM.4.0 nV is for a long time aware of ageia plans. maybe the Sm4.0 shaders are so general purpouse they can write back easier in a basic way. It can handle Physics well.
ATI looks like the have a headstart with Phsyics on GPU with there SM3.0 solution. So I expect that R600 shader tech would be more like a second generation of GPGPU while nV sm3.0 is a inbetween step. And G80 there realy full first attempt.
ATI plans to come with a 2GB GPGPU card. Looks like they where planning it for some time.
Imangine a R600 on 80nm with 2GB for local Physics memory.
VS Ageia PPU on 130nm and a limit of 128MB
I think Ageia needs a PPU2 at least on 90nm or smaller with 1GB or so to compete in the near future. a 400Mil or more transistor beast
SM4.0 shaders are not 'general purpose' they are just able to act as either vertex or pixel shaders instead of having a constant set amount. Writing back in SM4.0 is almost the same as in 3.0 other than fetch can be a bit faster due to the DX10 architecture without having to swap user/kernel mode. Also note that even if it was possible to do full gameplay physics on a GPU (which it's not atm) nobody in their right mind would do this because you would alienate 99.9% of customers that do not have the highest end GPUs or SLI.
That is the reason that we will either continue to see physics other than fluff done on the CPU, unless they start building the physics solutions on all boards as part of the chipset, which I highly doubt.
they need to make something like hyper threading for GPUs so when a game is not using some potential power then it can be used for physics. as long as it can be done without another 100 million transistors and power connector