Ageia responds to GPU physics

jdLordHelmet said:
Funny how you sink to personal insults while simultaneosly proving your own lack of knowledge on the subject. Download that rocket demo. That is real physics, it is not scripted, and last I checked, I don't have a PhysX card. You aren't even using terminologies correctly.
Click to expand...

I don't remember insulting you.

EDIT: I'm looking for the rocket demo now. I like cool demo's.

You should as well d/l the demos on the ageia website listed in the earlier posts, run it and see the difference. You will see 4-5 FPS. If you had the PPU it would be 50-60 FPS, that is what was demonstrated.

What is your whole dislike for the PPU again? It obviously works and does what it's made to do, so it has a use.

What really grabbed me in the [H] workshop was watching the rocks/pebbles falling/rolling down the mountain side, that was an impressive use of physical interactions. And they showed the difference being PPU accelerated and CPU handled, the difference was from 4-5 fps to over 60.

The problem I have with people saying graphics cards can handle it is I have games right now that graphics cards can't handle with all the goodies cranked up, why would I want any more processing to be put on that card when a PPU can make a night and day difference (especially when we are talking about top of the line hardware like SLI 7800's)?

Just admit it, your biased against the PPU and no amount of reality will change your opinion.

It's ok man, really, I'm fine with it.
 
But, you want me to d/l your simple rocket demo
Click to expand...

It isn't my demo, it's a demo someone else linked that uses the PhysX SDK. Since you don't have a PhysX card, it runs in software mode. One demo on there does bring you to 3 fps, the rest run as high as 1000 FPS.

What is your whole dislike for the PPU again? It obviously works and does what it's made to do, so it has a use.
Click to expand...

I didn't say I dislike it, I just said I think it's uncalled for. Just because one demo an Ageia guy shows you runs 100 times better on their card doesn't mean it's needed. But I did say several times in this thread if someone shows me unbiased results I would be all for it, if the price is right. But I don't think such evidence exists yet. Also, the supposed demo's that a CPU can't handle are demo's that wouldn't work well for any system once advanced graphics are applied to the polygon, such as shadows and lighting and bump maps.
 
I see that now that I tried to find it...

I'll get it eventually.

Also, the supposed demo's that a CPU can't handle are demo's that wouldn't work well for any system once advanced graphics are applied to the polygon, such as shadows and lighting and bump maps.
Click to expand...

So when you run them in wire frame mode they work well?

I think light/shadows can be processed in the PPU engine, after all light does interact with real physics, in real life. Might as well calculate it as physics instead of graphics.

I think Graphics shadow-light interaction processing makes too much use of culling, or removing calculations for what you dont see in each frame...to allow a totally real or convincing effect. This culling does not allow completion of interactions of the light where processing the light particles in physics could. Would that not be another good reason for a dedicated PPU device?
 
http://www.autosimsport.com/issues/autosimsport_1_4.zip

There's a pretty good article in there about how the PPU can be utilized in racing sims, with comments from the Ageia guys and also some game developers. They also touch on whether the chipset will only use their novolak engine or be programmable, and whether the physics stuff can be done on a gpu. Good reading, it's under the 'Front Straight' section. Looks like the racing sim guys are saying a minimum of 18-24 months before we'll actually see games depend on any kind of PPU for gameplay, and until then it'll just be "world fluff" stuff (ie neat explosions, nice rain etc, just fluff).
 
BBA said:
What is your whole dislike for the PPU again? It obviously works and does what it's made to do, so it has a use.
Click to expand...

There's nothing impeding progress, the whole issue is practicality. Do we really need it? Will it make enough impact in games to make people want to buy it en masse? Physics is a large part of the gameplay, implementing it on a small user base would isolate them from the rest of the community as the other users are likely not to have it. See the catch 22 i noted before. 3D evolution didn't have that problem unlike physics.

The problem I have with people saying graphics cards can handle it is I have games right now that graphics cards can't handle with all the goodies cranked up, why would I want any more processing to be put on that card when a PPU can make a night and day difference (especially when we are talking about top of the line hardware like SLI 7800's)?
Click to expand...

Are you saying a PPU will offload the work from the GPU?

Just admit it, your biased against the PPU and no amount of reality will change your opinion.

It's ok man, really, I'm fine with it.
Click to expand...

No, we're just more aware with how the hardware works. Some of us program software for a living and we are well aware of how resources are juggled to make them work adequately. While a PPU may elevate the load off a CPU, it just isn't practical enough to be worth the $200.

BBA said:
I think Graphics shadow-light interaction processing makes too much use of culling, or removing calculations for what you dont see in each frame...to allow a totally real or convincing effect. This culling does not allow completion of interactions of the light where processing the light particles in physics could. Would that not be another good reason for a dedicated PPU device?
Click to expand...

Yes, the floating point calculations when computing for realistic lighting is enormous (Anybody tried rendering the library or the dragon in 3D Studio Max?). A physics card may be of use with that (if you can find a 3D rendering program that can make use of it), but there's already professional level cards that are specifically designed for that already with even better performance and better support.
 
bobzdar said:
I could really see this coming into play in simulations, ie car simulations. If any of you extensively played F1 Challenge, there was a way to enable high rate physics in the .ini files. This was only useable on faster processors (and was disabled by default) as it would slow the frame rate significantly, but if you had the processor to handle it, it would make the car handling kind of come alive. You could feel the car skipping over bumps (especially hitting curbs and the like), more subtle variations in traction etc. Off-loading to a PPU allowing 100 times faster (or more) physics calculations would really be a shot in the arm. A lot of the physics calculations deal with tire modelling, and allowing more complex tire modelling at a much higher rate would increase the realism incredibly, especially when a lot of games have to resort to look up tables vs. on the fly calculations at this point. I'm very excited about this for that reason as I know game devs. that deal with simulations are going to be using this.
Click to expand...

think about what fluid dynamics could do for flight sims :D

check out X-plane that uses its own physics engine to do this now but its limited to a small number of elements to processes the air over
 
Ageia needs a forum where they can claim rights to all posts.



IMO, a dedicated physics card doing poly collision (including soft bodies), interaction with effects (particles and volumes), and influencing game design through breakable objects has a lot of potential (one demo shows a plane built with probably 20-50 objects crashing into a hanger and breaking apart piece by piece, looks very cool).

I believe a goal of gaming is raytrace quality which means we're still years and years away. For so long as that standard isn't being met, anything that adds processing power to your system will help.

If this takes off, we could be seeing the half-life 2 buggy being build by bolt and nut, able to accurately deform to collision and damage instead of a box on top of 4 other boxes. Need for speed---dents, scraps, surface frictions, etc...Battlefield---craters and realisticly destructable buildings (50 buildings 400 pieces per building and still leaves 10,000 objects to play with), flight dynamics, etc... None of which impacts system performance because the card is doing the work.

If this takes off, game fundimentals change because it's now also about parts and internals.
 
jdLordHelmet said:
I didn't say I dislike it, I just said I think it's uncalled for. QUOTE]

3D cards were "uncalled for" once. SLI... In the future, games will take advantage of a PPU to present levels of realism we just can't expect today.
Click to expand...
 
jdLordHelmet said:
Yeah, that was one of my points a few pages back. Say for example you went to the next step and made every wall a stack of individual bricks that could come unstuck with the right force of impact. Right there you just multiplied the number of polygons in that scene by however many bricks are in that wall.
Click to expand...
That would be unnecessary and impractical. Furthermore, a brick wall doesn't just topple like a Jenga set when hit, it usually shatters according to the texture of the mortar. Break it into pieces on the fly according to material. Some bricks would be broken in half, some sections of wall would remain intact.
 
Sly said:
There's nothing impeding progress, the whole issue is practicality. Do we really need it? Will it make enough impact in games to make people want to buy it en masse? Physics is a large part of the gameplay, implementing it on a small user base would isolate them from the rest of the community as the other users are likely not to have it. See the catch 22 i noted before. 3D evolution didn't have that problem unlike physics.
Click to expand...
The first generations of games would have the physics processor as an optional component with the ability to render physics in software. As the technology matures and becomes more ubiquitous, it will become a required component much like the GPU, FPU, 32-bit CPU, hard drive, DVD-ROM drive, etc.

No, we're just more aware with how the hardware works. Some of us program software for a living and we are well aware of how resources are juggled to make them work adequately. While a PPU may elevate the load off a CPU, it just isn't practical enough to be worth the $200.
Click to expand...


pr0pensity said:
The PPU needn't be a $200 addon card. They can sell a gimped version as a $20 chip on your gaming card...
Click to expand...
 
jebo_4jc said:
I think you are right.
When I worked at Best Buy (probably 2 years ago...9800XT and 5900 Ultra were the top-end video cards at the time), I sold hundreds of GF4MX's, and 5200's, and etc. to pissed off people who took a copy of a game home for their kids and found out that their Intel Extreme graphics weren't so extreme. So now they are coming to me with an attitude saying things like "you want me to spend another $100 to play this stupid game?"
Now, if they take a game home, and find out that not only do they need to upgrade their graphics, but they need a new PPU too, they will be doubly pissed.
Click to expand...

Except that games that take advantage of a PPU will gracefully degrade to run on PCs without a PPU.
 
I can see no real use in arguing about it. I'm adopting a wait and see approach.

I will say common sense suggests a PPU is a good idea if it can be implemented without a hitch. The idea of idle cpu time doesn't stand up to reason because the biggest impact of the PPU will be when the system is being taxed the most. Simply tossing a grenade into a dynamic environment ie. the first time you met that heavy armored guy in F.E.A.R. would mean every shelf, every piece of machinery, every object in the game takes damage and reacts. Concussion would toss those objects around in a real world fashion. On top of that have 8 guys rushing you firing their weapons and/or tossing grenades with all the consequences of their actions.

I'd guess the difference should be slight over 30-10,000x in complexity (varying because of graphics complexity).

I'll wait for the real numbers. A good test would be have the user launch a missle at a 20,000 wet-brick building, using 64-bit textures, dynamic lighting, bump, volume (dust) and specular. Of course you'd have sound, particle effects and lighting from the impact.


Plus liquids are cool


Sry for the edits, engrish ain't too good this morning.
 
Da Frechman said:
It'd be cool if Havoc could run on PhysX processor, perhaps Ageia should think of broadening it's horizons a bit.
Click to expand...


thay have no one else whats to use it which is an issue but im sure there taking to MS about extetion to Direct X
 
Elios said:
thay have no one else whats to use it which is an issue but im sure there taking to MS about extetion to Direct X
Click to expand...


Yup, I agree, although Microsoft usually prefers to compete in existing markets.

On a tanget from the topic, I've moved far far away from the gaming aspect. It's all pretty gal-darn exciting (hair dyers, coffee pots, and key chains. grin...)
 
For me, I'm all for it actually....

If i'm not mistaken, PPU has already been implemented in next gen console like PS3, and some of the effect are shown in tech demo like mud splashing on the windscreen of a car.

In fact, i belive games developers would welcome this extra hardware as well. We're definetly gonna need it to achive something like what we've seen in the video of the next gen CryEngine, buildings that are destructable, tree leafs and plants that moves being push by the character moving through it, and i'm sure there are many other more application to utilize it. Stuff can be broken into much more pieces, such as glass shattering into thousands of pieces.

But of course like someone mentioned it, overall system have to be more powerful to utilize it.
 
id bet in a few years there will be a "Direct Physics" API added to Driect X
 
HOLY CRAP im good Brent you guys need another Hardware guy lol :D
 
id bet in a few years there will be a "Direct Physics" API added to Driect X
Click to expand...

I'd back that bet with a lot of money :)

Hey, look at this:

You will be a member of the core engine team who will be primarily responsible for working closely with our Direct3D team, helping to define, develop and map optimized simulation and collision algorithms onto data structures that are optimized for the GPU.
Click to expand...
You should have a good understanding of 3D Math as well as current CPU and GPU architecture.
Click to expand...

Hahaha :)

At any rate, a Microsoft physics API is VERY exciting. I love programming in DirectX, and I can't wait to get my hands on this.
 
jdLordHelmet said:
I'm an amateur graphics programmer. Some of the stuff you guys are talking about goes beyond what we think of when we say "physics" and gets into things that would require a complete overhaul of how we create the graphics that show on your screen. Graphics is about drawing polygons and mapping textures to them, and physics is about moving those polygons around in a realistic way. From that perspective, stuff like deformation, entropy, "blowing holes in walls", etc, is gonna take a lot more than just physics. If fact, blowing realistic holes in walls has nothing to do with physics, in the graphics programming sense of the word. At that point, you aren't even dealing with polygons, you would be dealing with some type of pixel-by-pixel mapping or some other kind of wacky, not-yet-created method. Which makes sense, if you look at graphics progression through a sort of calculus filter, as graphics get better, the number of polygons gets larger/the size of polygons get smaller - eventually we may end up with polygons 1 pixel large - in other words, just plotting billions of 3D pixels instead of polygons (this is already done, but not for entire scenes). The "physics" for such an engine would be insane, sure, but all of it would be insane. I still say a physics card will never be needed. Ok, let me rephrase that: a physics card should never be needed. Physics will always be a laughably inexpensive venture compared to graphics.
Click to expand...

Well put and well written.. thanks for that.
 
AndoOKC1 said:
I have to disagree with the second paragraph...alot of what we see with games arent supported by more than a 1/3 of the gaming population...such things include sounds cards that support things like X-FI, EAX, Dolby Live but game companies still implement them for those that do. Most people dont even have the hardware to run alot of their games with the eye candy all on. If you look at Half Life 2 system survey ...most people still run Nvidia 5200's, ATI 9600 and 9800. These are barely capable of any AA or AF and other options at high resolutions. I dont believe for one second that game companies produce for the general population b/c that would only create and an average game for the average player. Instead they code for the highest end systems that can support the features that put their game on the map and those that cant run it just have to run the game sub-optimal and those that can will enjoy the game as the programmers intended (100% eye and ear candy).

Dont get me wrong, they keep the average player in mind but that is done simply with on/off options for AA AF, and different resolutions. THis I believe will be the same mechanism for implementing the physics processor. Just an option...

Since you brought up voodoo...i remmeber when i got my voodoo2 and some games would let me use it by selecting it optionally. This is the same picture im seeing for the PPU.
Click to expand...

In essence, this also ellongates the longevity of the game. Games will be purchased at higher volumes way well into the future when people who weren't able to play with all features on can in the future when technology prices drop, as opposed to the games that were created for the current times' "standard" machines, which will be old tech in a matter of months. (Think selling quarters).
 
jdLordHelmet said:
I'm an amateur graphics programmer. Some of the stuff you guys are talking about goes beyond what we think of when we say "physics" and gets into things that would require a complete overhaul of how we create the graphics that show on your screen. Graphics is about drawing polygons and mapping textures to them, and physics is about moving those polygons around in a realistic way. From that perspective, stuff like deformation, entropy, "blowing holes in walls", etc, is gonna take a lot more than just physics. If fact, blowing realistic holes in walls has nothing to do with physics, in the graphics programming sense of the word. At that point, you aren't even dealing with polygons, you would be dealing with some type of pixel-by-pixel mapping or some other kind of wacky, not-yet-created method. Which makes sense, if you look at graphics progression through a sort of calculus filter, as graphics get better, the number of polygons gets larger/the size of polygons get smaller - eventually we may end up with polygons 1 pixel large - in other words, just plotting billions of 3D pixels instead of polygons (this is already done, but not for entire scenes). The "physics" for such an engine would be insane, sure, but all of it would be insane. I still say a physics card will never be needed. Ok, let me rephrase that: a physics card should never be needed. Physics will always be a laughably inexpensive venture compared to graphics.
Click to expand...


A solution would be very similar to a transparency (or alpha channel) subtracting from the texture and mapped as desired according to the object. The physics would come into play with collision detection (physics engine) according to the size of subtraction. (angles and types of damage would also have to be considered). The geometry of the brush remains untouched, it's simply a new map (texture) with a given properties. LOL, unroll a black hole, throw it against the wall or floor and jump through it or paint a door on a wall and then go through it.

A concrete pillar supporting an upper floor would receive little damage from a 9mm, where as the same pillar can be taken out with a rocket launcher (causing the very surprised contents of the upper floor to be hastily reloacted to the preceeding). Whether that pillar is consistent throughout or has rebar is up to the developers (the conditions are set by the developers for the catastrophic event). That pillar can be taken out the way the user wants under the conditions set by the developers (a little or a lot, how and where should be up to the player). What those *very surprised* contents do until they find their very desireable quiet is entirely up to the physics engine.
 
D3v01D said:
Have you ever fired a rocket at a wall and wondered why it doesn't make a hole? Or why you can't make waves in a pool of water?
Click to expand...
In unreal you can, and everquest II I think (make wavy water). I don't understand why other games can't do it too.
 
Da Frechman said:
A solution would be very similar to a transparency (or alpha channel) subtracting from the texture and mapped as desired according to the object. The physics would come into play with collision detection (physics engine) according to the size of subtraction. (angles and types of damage would also have to be considered). The geometry of the brush remains untouched, it's simply a new map (texture) with a given properties. LOL, unroll a black hole, throw it against the wall or floor and jump through it or paint a door on a wall and then go through it.

A concrete pillar supporting an upper floor would receive little damage from a 9mm, where as the same pillar can be taken out with a rocket launcher (causing the very surprised contents of the upper floor to be hastily reloacted to the preceeding). Whether that pillar is consistent throughout or has rebar is up to the developers (the conditions are set by the developers for the catastrophic event). That pillar can be taken out the way the user wants under the conditions set by the developers (a little or a lot, how and where should be up to the player). What those *very surprised* contents do until they find their very desireable quiet is entirely up to the physics engine.
Click to expand...

I doubt simply making a dynamically shapped 2d hole would look that great, not to mention the problems it would cause in terms of looking up close at non orthogonal angles.
 
scrumdiddly said:
jdLordHelmet said:
3D cards were "uncalled for" once. SLI... In the future, games will take advantage of a PPU to present levels of realism we just can't expect today.
Click to expand...

Again, the reference to the 3D evolution. The transition phase for 3D was virtually painless, there were no drawbacks, no issues, no gameplay changes. A 3D card user and a software one could play hand in hand with no penalty, neither has an advantage over the other aside from the higher res. This was especially demonstrated when HL became popular. 3Dcards were expensive, but they were completely optional and changed <NOTHING> ingame.

Has anyone else here actually been around when the Voodoo1 came out? Or did you guys just arrive around the time of the GeForce2.

Elios said:
id bet in a few years there will be a "Direct Physics" API added to Driect X
Click to expand...

Hopefully sooner. A lot of the issues will likely be solved by the introduction of a primary standard. Any chance we'll get one by the time longhorn comes out? And when it does, will the Ageia PPU be compliant? The early video cards had issues with DirectX too when the 3D API came out.
Click to expand...
 
jdLordHelmet said:
People keep talking about the RISC aspect of a PPU, but exactly what reduced instructions would you use to model complex calculus-based physics?

.....


But a physics card, in order to do all the things you guys are talking about, would have to handle a wide variety of instructions, and wouldn't be much of a RISC processor at all.

.....

Well, after all my ranting and theorizing
Click to expand...


Really!

go read up on microprocessor architecture, what you said shows the only thing you know about RISC is what it stands for, and doesn't even make sense.



jdLordHelmet said:
Funny how you sink to personal insults while simultaneosly proving your own lack of knowledge on the subject.
Click to expand...
 
Nice to see that "adults" are still capable of poo-slanging matches akin to those of children arguing over Yugi-Oh cards ("you're gay!", "no, you are!", "your dad is!" etc). Those of you throwing said poo - be quiet, stick to the topic at hand and keep the abuse to a minimum, please.

The only question we must ask of the PhysX card is "how long until it becomes affordable?". Pity anyone who asks why we need it, and shun anyone who says we don't need it.

The greatest question ever asked is "Why?". The greatest answer ever given is "Why not?".

If you have difficulty understanding the moral of my post, please return to school and finish Year 8.
 
acidfr33k said:

Really!

go read up on microprocessor architecture, what you said shows the only thing you know about RISC is what it stands for, and doesn't even make sense.
Click to expand...


Seriosly, do you know? Did you realize the Power IBM chips that Apple uses are less RISC than P4's?

Pretty much all CPUs are "post RISC" these days. And the PPU is more of a SIMD streaming CPU than a RISC CPU.

http://arstechnica.com/cpu/4q99/risc-cisc/rvc-1.html

==>Lazn
 
what i would like to know is what is on this add on card except a chip.... i mean would it not be more efficent to just simply dump the PCB board it sits on and add it to the graphics card? This way it doesnt have to travel over the bus to get down the the GPU? if it only going a few math calc's then you could probly go out and get an old zilog chip plug it on a graphics cpu and bammo it will work.. and an increase manufacturing cost of like a dollar..
 
Pharacon said:
what i would like to know is what is on this add on card except a chip.... i mean would it not be more efficent to just simply dump the PCB board it sits on and add it to the graphics card? This way it doesnt have to travel over the bus to get down the the GPU? if it only going a few math calc's then you could probly go out and get an old zilog chip plug it on a graphics cpu and bammo it will work.. and an increase manufacturing cost of like a dollar..
Click to expand...

Yeah, because I SO want to pay for a new PPU EVERY time I buy a new GPU.

Chances are PPU's will have a much longer lifespan than GPU's, even for enthusiasts, so I would rather buy it once and then not have to worry about it for a few years.

As for whats on this card.... It's a powerful chip, over half the size of a GPU, and I think it requires active cooling. I believe it also has 128MB of RAM... So your videocard would be considerably bigger too if you added this too it.
 
acidfr33k said:
blah blah blah
Click to expand...

I was trying to keep the argument simple because most people here wouldn't know what I was talking about. From a basic standpoint, everything I said was true.
 
i get 80-90 fps in css stress test with a stock 3000+ venice and OCed GTX. Obviously it's a CPU problem, since my friend with a 3500+ winnie and GTX gets 140. Same this happens in the fear demo, lag that isn't explained by the graphics card. (Let's not get into OCing my proc or anything, it was just an example.)

So obviously in those scenarios physics was the limiting factor, not graphics. That's why I think a PPU will help. My friend was talking about quake 4 and said they were limited to so many AIs and so many lights at once (something like 6 and 12 respectively. or thereabouts) b/c otherwise they'd bring anything current to its knees. A PPU will let you process so many more lighting effects and particles it won't even be funny.

My .02
 
Pharacon said:
what i would like to know is what is on this add on card except a chip.... i mean would it not be more efficent to just simply dump the PCB board it sits on and add it to the graphics card? This way it doesnt have to travel over the bus to get down the the GPU? if it only going a few math calc's then you could probly go out and get an old zilog chip plug it on a graphics cpu and bammo it will work.. and an increase manufacturing cost of like a dollar..
Click to expand...

128mb of GDDR3. And a high MHZ SIMD CPU on a FPGA? not going to happen. This thing has more horsepower than a modern CPU when it is doing it's kind of math.

What we know so far:
*125 million transistors
*182 mm^2 die size
*130 nm process technology from TSMC
*20 watts power consumption
*128 MB of GDDR3 memory

==>Lazn
 
You must log in or register to reply here.
Back
Top