AGEIA PhysX Processor Preview

[bapenguin]

PC Perspective (http://www.pcper.com/article.php?aid=140&type=expert) has a writeup on the upcoming and first ever dedicated physics processor.What AGEIA and even game developers envision a PPU will enable for a gamer is a world with physics unlike anything we have seen in a real time game before. We are talking about thousands of rigid bodies, real flowing water, hair simulation, avalanches of rock, clothing simulations and more. Even more impressive is the idea of a universal collision detection system that allows you to interact with absolutely ANYTHING in a game world. All of it calculated in real time with nothing scripted in the game engine.

A bunch of problems with this.
1) The first cards coming out are 249-299 bucks!
2) Developers need to develop for the lowest system specs. In this case, it'll be really hard to add "extra physics" to a game for people with the PPU.
3) Is it really necessary? With Dual Core CPUs becoming standard this year, this could be overkill when the extra core could be dedicated to physics.

It'll be interesting to see if this can make itself into the market much like graphics cards did in the late 90's

[dr_qwandry]

are physics really going to play that much of an important role in the future?
mostly it's just been eye candy and only one game has actually used it in a practical way (HL2) and even that wasn't completly amazing.

[Zane]

Think of this in terms of graphics card. No desktop CPU out there can do software rendering anywhere near the level of quality of a current graphics card. This is probably similar. The processor on this add-on card most likely calculates physics many many times faster than a standard CPU, dual core or multi-processor systems included.

Not saying I'd be plopping down $250 but if it were in the realm of $50-100 and a few games supported it I'd probably want to pick one up.

[ben_bot]

the reason graphics cards made it so quickly was that you could see a marked performace boost if you had a graphics card.

For me, physics is part of the whole experience. So they need integrate their card with graphics cards so that experience can be realized easiest.

[kokyunage]

They just need on game that is a huge hit to take adventage of it. Like GLQuake for 3dfx. I concur that the cost seems prohibitive. I thinking it should be in the soundcard price range ($50-$100).

[Morrolan]

Physics need to be higher level before they will have a huge impact on gameplay.

Imagine shooting out pillars to bring cieling crashing down on an army of enemies, in a completely non-scripted way.

[Borys]

A bunch of problems with this.
1) The first cards coming out are 249-299 bucks!
2) Developers need to develop for the lowest system specs. In this case, it'll be really hard to add "extra physics" to a game for people with the PPU.
3) Is it really necessary? With Dual Core CPUs becoming standard this year, this could be overkill when the extra core could be dedicated to physics.

It'll be interesting to see if this can make itself into the market much like graphics cards did in the late 90's


1) Agree, way too expensive. $99 - $150 max would be a sweet price.
2) 3DFx did break into mainstream, maybe this will too?
3) Yes it is. In theory (?) this could calculate physics up to 100 times faster than dual or even triple core CPUs.

What we need is a clear jawdropping example of that chip in work. Show me a beatiful castle made of individual bricks and destroy it LOTR style. Or a battlefield Rome: TW style with havok 2.0 on every soldier. Of course make it so complicated that it runs @ 30 FPS with the physics accelerator and @ 1 FPS without (no cheating).

[Borys]

One more thing:

this could be a BIG advantage of PC gaming. After all you won't get this stuff in the next-gen consoles. Imagine HL3 with full physics accelerator support that simply can't be done on PS3/ Xbox 360.

[Stupid Fat Hobbit]

the reason graphics cards made it so quickly was that you could see a marked performace boost if you had a graphics card.

For me, physics is part of the whole experience. So they need integrate their card with graphics cards so that experience can be realized easiest.

This is exactly what I was thinking. Given that games are the only (common) application that would require a physics card, and most games already require a graphics card, integrated physics/graphics makes perfect sense.

Also, I agree that all this needs is one GLQuake-esque application to push it into the mainstream.

How much were the first 3d cards, out of interest?

[Tyrant]

Yeah, with no games that really make use of "realistic" physics in a gameplay standpoint, save for Half-Life 2, a physics processor doesn't come across as a "must buy" type of addon. Especially at a price above $100.

[Hellstorm]

One more thing:

this could be a BIG advantage of PC gaming. After all you won't get this stuff in the next-gen consoles. Imagine HL3 with full physics accelerator support that simply can't be done on PS3/ Xbox 360.

Hmmm... could it be... in the Rev?

[Borys]

Also, I agree that all this needs is one GLQuake-esque application to push it into the mainstream.

Duke Nukem Forever.

Dare I dream?

[Stupid Fat Hobbit]

Yeah, with no games that really make use of "realistic" physics in a gameplay standpoint, save for Half-Life 2, a physics processor doesn't come across as a "must buy" type of addon.

News just in: archeologists have just unearthed a fragment of a stone tablet, believed to date from 1996. In it, an unknown author writes: "...with no games that really make use of accelerated 3D graphics in a gameplay standpoint, save for GL Quake, a 3D accelerator doesn't come across as a "must buy" type of addon."

[president_fred]

One thing I would appreciate is if the next batch of Nvidia SLI graphics cards would come with this technology, with dual cards it would be great and I need to update my 6800 for Oblivion anyway. It would simply be one less thing to worry about.

[dr_wily]

i sorta like half assed physics.. thats what made max payne 2 fun.

[Tyrant]

Can someone tell me how accelerated 3d graphics are used in actual gameplay?

[snubber]

This won't hit mainstream in my prediciont. GFX did it but not sure about Physics. The reason is that you can design a game that has varying levels of graphics that don't affect gameplay. For example, you can play at a lower resolution, or use 16-bit textures instead of 32. Or maybe the engine renders fewer particle effects, or uses blocky shadows instead of nice soft shadows. All of this looks pretty, and so there's a need for a fancy, high-end 3d card, but it's not necessary.

But with physics? To me it seems it would affect gameplay a LOT more, and you can't just "drop down" like you can with graphics. You're not going to build an engine/level where you can say knock down an entire wall, brick-by-brick that is only supported by less than 1% of the market. It's not a fancy particle effect or a cool water shader, but providing actual "physical" (well, virtually physical :) ) changes to the game world -- bricks flying around, etc.

I dunno. If I were a game engine developer, I'd be ignoring this and just looking to dual-core, despite less power.

[KarmaGhost]

I skimmed the article and couldn't find anything on the price. I could have missed it, but I didn't see anything saying it would be that expensive. But, it would definately need to be much less than that for it to become truly mainstream.

[Paranoia]

No way this stuff is going to succeed. Up until now I cannot find a reason to buy a physic accelerated card to add more fun to my gaming. Half-Life 2 physic is more of a novelty and it doesn't lend much to replay values.

[KarmaGhost]

No way this stuff is going to succeed. Up until now I cannot find a reason to buy a physic accelerated card to add more fun to my gaming. Half-Life 2 physic is more of a novelty and it doesn't lend much to replay values.Maybe that's because there are no games that really use physics to any great extent. Once there are, you'll see the need. As someone posted before, no one really thought there was a need for gfx cards back in the day, let alone a dedicated GPU.

[Talanvor]

Got to agree. Right now there's nothing that even supports the cards. Now, if they can get the Unreal Engine 3.0 games to use this.... well. Plus, come on, $250? You need to drop that down at least $100 before I'd be interested. If I've already put down $500 for a graphics card, $250 is just breaking my balls.

[SlamDunk]

Here's my little page, dedicated for PhysX info. I will keep that page up-to-date as good as I can.

http://personal.inet.fi/atk/kjh2348fs/ageia_physx.html

I'm very excited about PPU and I think physics will eventually become one of the biggest factors for a absolutely immersive gameplay experience. Physics in todays games are on a level of knocking down a few chairs here and there. That's nothing.

[DiBiddilyBop]

Got to agree. Right now there's nothing that even supports the cards. Now, if they can get the Unreal Engine 3.0 games to use this.... well. Plus, come on, $250? You need to drop that down at least $100 before I'd be interested. If I've already put down $500 for a graphics card, $250 is just breaking my balls.

Once upon a time I heard Unreal 3 had full support for this technology. Seeing as how Unreal 3 is going to be the best thing since sliced bread when it comes to engine design and every game under the sun is licensing it... maybe this will become a big deal in the near future.

[Lint of Death]

I think this card could be a great thing to have in the future. How can I believe that? Age of Empires 3. Just read about the sorts of things they're using physics for. They won't be the last to use physics-based collapsing towers that kill troops. One day, I'm sure that these physics cards will be necessary to get a great experience out of modern (our future) games.

[lpmiller]

Dual core procs are not standard this year, nor will they be. They are out, yes, but way too expensive to be considered standard. Nor is there a lot of point to them for most gamers. Yet. However, they are the wave of the future as far as goes. But standard? Not for a year or two, yet.

[bapenguin]

Dual core procs are not standard this year, nor will they be. They are out, yes, but way too expensive to be considered standard. Nor is there a lot of point to them for most gamers. Yet. However, they are the wave of the future as far as goes. But standard? Not for a year or two, yet.

By standard I mean, after this year, Intel and AMD will no longer produce mainly single core processors. On the road maps they are mostly all dual core....hence the new standard.

[RMan]

Well, I'm definitely looking forward to these things, as a developer I'm currently much more limited by what I can't do with physics than what I can't do with graphics. I’m using Novodex, which is excellent, but in software there’s just so much you can do, so I really want the hardware base to be there. The current graphics cards are powerful enough that graphics technology is no longer a significant limiting factor, and advancing it will yield diminishing returns. Physics, on the other hand, isn’t near that, and when proper demonstrations are available I think the average gamer will see that it’s going to be almost as significant as graphics hardware.

[XenonCJ]

Prediction : AGEIA will be out of business within 1 year.

Why? Your average Joe gamer ain't buying this as a stand-alone product EVER.

Also, soon, as dual-core tech becomes more common, the extra CPU will probably provide enough extra processing to make a dedicated physics chip's benifit negligible.

I think the only hope for them is to license this tech to be on chip on a graphics card, or on a CPU as extra instructions or something.

Flame away.

[A Lusty Alien]

Gee...

It seems that for the added capacity and gaming experience that a physics card would offer, $250 - $299 doesn't seem like all that much to spend. Not when you have high end graphics cards selling for between $400 - $800 dollars.

[Paranoia]

Prediction : AGEIA will be out of business within 1 year.

Why? Your average Joe gamer ain't buying this as a stand-alone product EVER.

Also, soon, as dual-core tech becomes more common, the extra CPU will probably provide enough extra processing to make a dedicated physics chip's benifit negligible.

I think the only hope for them is to license this tech to be on chip on a graphics card, or on a CPU as extra instructions or something.

Flame away.

No flame. I pretty much agreed with you.

[RMan]

The idea that a faster cpu is a replacement for specialty processors is wildly silly, and refuted by a large number of examples, the biggest of which is obviously graphics cards. Although the ability to sell an add-on card remains to be seen, I feel it’ll be a little slower than graphics hardware, but the bottom line is that it will be in the next-next gen consoles no matter what the sell-through on the PC is. If they’re positioned to be the front runner for Xbox 720 or PS4 then they’ll be just fine.

[mindstrike]

Well here is hoping that with the higher price point of this card... maybe we can get away with those FX 5200s....... ;) For most of the Processing will be offloaded....

(Yes, I know there are "factors", just being humourous)

Still, if this "catches" on, Nvidia or Ati will suck them up... and add onto their cards....

[Thenetcase]

I would buy one at that price.
I believe that nothing makes a game more believable than good physics... Graphics take second place to that.

[mister_slim]

Wouldn't the Cell coprocessors, being designed for floating-point calculations, be pretty easy to use for physics?

[ElectricMonk]

they should stick this in arcade games or something. otherwise nobody is going to care when they can't see the difference it makes in games.

[Tyrant]

Gee...

It seems that for the added capacity and gaming experience that a physics card would offer, $250 - $299 doesn't seem like all that much to spend. Not when you have high end graphics cards selling for between $400 - $800 dollars.

At this point in time, I'd imagine the only people that would pay $250-$299 for a first generation physics processor would be the same people that pay $400-$800 for a graphics card.

[Furious Wang]

One more thing:

this could be a BIG advantage of PC gaming. After all you won't get this stuff in the next-gen consoles. Imagine HL3 with full physics accelerator support that simply can't be done on PS3/ Xbox 360.

Actually, you will. That's the whole point of the PS3s 7 ppes. To be used for thing like this.

Anyway, if ATI or nVidia are smart, they'd buy this company up and integrate it onto their video cards. Whichever gets this tech will have a pretty big head start as this will be in demand in a few years. Just not as a stand alone.

[Liquidize105]

Just more reason to play console. PC has way too much crap to buy, to tweak, to go wrong.

[Morratut]

I agree with XenonCJ. The best strategy for Ageia is to license the tech and chip to ATI or Nvidia.
If they integrate this into a graphics card with a big label saying 'NOW WITH PHYSICS!' and games start coming out with it then yeah it could significant advantage over the next gen consoles.

2 years from now i can see PC owners laughing at the 'software physics' in consoles :D

[Carnifex]

If this PPU is useful for more than just game physics (and it should be especially for scientific applications), I'd rather see it licensed to CPU makers as an integrated or multi-core solution. It all depends on what kind of calculations it does, the PPU might be too specialized.

[netcraazzy]

Imagine that you not only have to pay attention to which graphics cards a game supports but also which physics cards. I'm of the opinion that PC gaming is becoming more and more irrelevant as the cost of entry gets higher and higher. If the day comes where you not only need an expensive, specialized graphics card but also a specialized physics card to play the latest games then how many people do you think will make the investment? Many of us probably would but how many people who don't consider themselves enthusiasts would? As the gap between the high-end and the low end system gets wider and wider it'll be much more difficult for game developers too.

[RMan]

Wouldn't the Cell coprocessors, being designed for floating-point calculations, be pretty easy to use for physics?
Perhaps they could do a passable job, of course the Cell is more marketing than reality right now, so judging it's capabilities for such things is premature. Again, there’s a big difference between what specialized and general purpose processors can do, for instance, graphics hardware commonly operates on 4 float values per instruction because 3 or 4 component computations are extremely common with vector and color computations. This results in a big win because it’s streamlined for this purpose, rather than simply faster floating point processing. I don’t know what the PPU will do architecturally, but you can expect similar optimizations for common physics operations that will result in a greater efficiency for the specialized task.

[bobbler]

250 price point is pretty bad for a product like this, that won't see much use until mass market has them -- thats when gameplay could be based around this kind of stuff. Currently the best you can do is get a small framerate jump, since what the CPU would be doing the PPU can take over. But it wouldn't even be near stressing the PPU. If a game was made with the idea that people have PPUs then they could do so much more in terms of gameplay. Physics really is a gameplay changer more than just a superficial addition like graphics.

Ageia's PPU is pretty close to a GPU without shaders. Its supidly parallel with an extremely long pipeline (I'm assuming here since graphics and physics are pretty close, GPUs have pipelines that make the P4 look like a baby), it would need to be -- its the reason a dedicated piece of hardware can do so much better than a CPU for things like physics and graphics. You'd probably need all 7-8 of the SPEs on the cell to get the kind of throughput you can on a dedicated piece of hardware. Theres probably some sort of hardware collision detection also, which is a huge pain to do otherwise (you have to backtrack in code once a collision is detected, if it'll do it in hardware that'd save a lot of trouble right there).

I don't really see video card makers adding another ~100++ million transistor chip onto their cards until they can just combine it with their current chips (most of them run pretty hot already and suck up quite a bit of power). Another bank of GDDR3 would be needed also (or at least more ram and bandwidth to make up for two things sucking it down). It would add more cost to video cards. Who would buy a 500-700 dollar video card? I know some nuts do, but most don't have the cash for that; $500 wouldn't even get you the highest end card then also. I can see a day when 'shaders' get to the point where they will take any type of code (hell even now you can write a pac-man game inside a vertex shader if you wanted to), you could do physics calculations while doing graphics. It would make sense to incorporate both into one chip, being able to check for collisions and such at the same time as you draw them would save actually quite a bit (they could use normal maps for the collision data to make the collision detection even better). A chip like that would be insane though, not something we can expect to see anytime soon.

It'll either catch on and become standard (which I hope, I don't think another company will try to add a dedicated 'physics' chip again for quite a while if this fails), or fail miserably. The $250 price point isn't helping them. However if they sell a cheaper card thats ~100 and still easily more powerful in physics than a CPU then they can catch some market. The $250 card is several magnitudes better than a normal CPU at it (~1000 objects for CPU vs 30-40k objects for the $250 card). They need a cheaper card with maybe 32-64mb of GDDR3 (that'll help cut costs pretty nicely). They just need to be smart and market accordingly, they have a lot of developer support (UE3 and a few other big names; their SDK is getting all over). Start small and go up, theres no reason for cards that can handle 30-40x the physics load as a CPU now, nothing is going to take advantage of it -- we need a card that will maybe handle 10x and go from there.

[1FSTCAT]

Physics processors WILL go mainstream, one way or the other. It IS the "wave of the future".

I don't see it as an option to go on a graphics card though, since most manufacturers push the bandwidth limits of the bus, as it is. (Although I don't know the numbers as far as PCIE goes)

I would think it would make a good choice as an integrated unit on a motherboard, or in a processor, though.

Supposedly these processors have been used in collision modeling for auto and airplane manufacturers for several years, already.

--Ed

[RMan]

Actually, bandwidth would be improved if the PPU was integrated with the GPU, since object positions would not have to go to the CPU then back out to the GPU. This could be a big win for fluid and particle systems, since the bandwidth could be quite high (in fact, it wouldn't take long before this seemingly small reliance on the CPU would become the bottleneck) and of course result in greater parallel processing. This would of course require tight integration with Direct3D, but architecturally integrating with GPU seems to me the best approach.

[bobbler]

Actually, bandwidth would be improved if the PPU was integrated with the GPU, since object positions would not have to go to the CPU then back out to the GPU. This could be a big win for fluid and particle systems, since the bandwidth could be quite high (in fact, it wouldn't take long before this seemingly small reliance on the CPU would become the bottleneck) and of course result in greater parallel processing. This would of course require tight integration with Direct3D, but architecturally integrating with GPU seems to me the best approach.

That's true. As I mentioned it would save a lot of trouble by having the GPU/PPU use the same assets (I just didn't think about it bandwidth wise, but since they should be able to use the same assets, technically, it wouldn't really need extra bandwidth necessarily). Being able to use a 2mil poly normal map as a 'collision map' would be nice. No more swords going through heads or feet going through walls, or fists going through other people, or any other silly clipping problem. It would be a monster chip though... 400+ mil transistors easy, and it would only go up from there. Hell they have 500 million to 2 billion transistors in some of Intel's Itaniums, so its possible, but that just gets downright expensive and hot (yields also skyrocket downard when you 500+ mil transistors now days, even Intel had horrible yields on the simpler Itaniums and they have the best fabs in the world). So... it probably won't be happening any time soon, but it's certainly nice to think about.

[phillycheese81]

Some PhysX videos are located at http://airtightgames.com/currentproject.html

Looks like you could do some fun stuff with it. I would love to see a Death Match level with that many objects. Something where the level could change because walls are blown out ... etc.