On the Horizon: Physics Processing Unit (PPU)

#1 My guess is that if any of the major studios adds support for this in their next-gen engine this might work.

And I bet you that Creative (or a similar add-on producer) will jump on this one.. :)

#2 Creative Physics Blaster.

#3 I heard Epic will be supporting it in Unreal Engine 3. Interesting idea but I don't think I would buy one as a seperate component until it's supported by most engines. Who knows maybe having it for Unreal Engine 3 will be enough for it to hit critical mass.

#4 Won't PPUs be obsolete pretty soon as CPUs continue increasing in power ?

the third-party FPUs (floating point units) of the early 90s* were, yes, pretty cool because they gave you several-fold improvement in power, but also sucked because they were a hassle for programmers to work with, and users had to be aware of which software supported them etc.

* i'm thinking of the motorola 68881 in the amiga, specifically.

#5 It seems like CPUs have hit a point where they're not improving as fast as they were. Something like this to handle the physics part would help a lot.

#6 >Won't PPUs be obsolete pretty soon as CPUs continue
>increasing in power ?
Of course, and who needs more than 640k of memory... (old joke at the expense of Bill Gates) As power increases so do the standards. Supporting more physical objects, modeling them with more precision and greater depth, heck why not have nuclear criticality calculations on that ball of plutonium in your game? As our machines get more powerful, I'm certain that we'll find ways to use that power.

>sucked because they were a hassle for programmers to work with
GPUs provide hardware support for the most common graphics libraries, OpenGL and DirectX. Developers on these libraries don't need to worry about how or whether a GPU is accelerating the library functions. Obviously a PPU will be a huge failure if its makers can't follow such an example, but I get the vague impression that they understand that:
"the physics middleware had already well established for some 3 years prior"
-page 2 of ageia link-

>users had to be aware of which software supported them etc.
This is a significant issue affecting how quickly (and whether) this new technology takes off. If major engines support it, though, I have no doubt that it will take off.

Floating point units specialized in only a few simple operations, so it makes sense that their functionality was eventually absorbed into CPUs. Physics processing is much more open-ended though. I really like the idea of pushing the envelope in hardware-accelerated physics calculations, and I don't think they will run out of new physics calculations to support.

#7 I'm still holding out for neural net addon cards to imporve game AI.
Creative Intelligence Blaster has a nice ring to it.

#8 My thoughts....Organic processors/RAM, etc. T h a t would be "truly revolutionary". This is more hardware... blah!! I'm still recovering from the last $500 I spent on the latest revolutionary video card...

#9 Good points.

I certainly support using PPUs if they're available and well-conditioned etc,
i just have a feeling they might be a technology which will only be relevant for five years or so.

long bets anyone ? :D

#10 I remember posting something about indie hardware sites, and asking where are all the indie hardware guys - and basically got laughed outta the forums when I mentioned PPUs (what I called World Processors), sound processors, AI processors, etc. Anyway, if I can find my designs from like 98-99 on this very type of processor Ill post them on my site just for craps and grins.
I put my money on PPUs, whether theyre a seperate unit on a board or integrated onto the CPU die. The fact is that chipmakers need to get away from putting everything on one huge chip and distribute tasks among several specialized processors, each with their own memory space. Yes, we have GPUs and to an extent sound processors, but the PPU will boost performance because it is adapted to doing that. I think its a good thing, maybe the big chipmakers will pull their heads out and see that more simple processors are better than one or two complex ones.

Just my .005 cents

Now that I think of it, guys at work also thought the idea was absurd.

#11 I tend to agree with those saying that PPU's will work along the same evolution path as GPUs, and not FPU's: when the FPU got absorbed into the CPU itself (and for quite a while, it simply was co-located on the same chips), it was mostly because the peak capabilities of CPU's hadn't been achieved until then.

Now that CPU's really are reaching the peaks, or at least near long term usability peaks, when it comes to capability and performance, I think that GPU's and things like PPU's actually are going to be viable for quite a while. I don't see GPU's going away any time soon at all, and I personally think that the usefulness of something like a PPU would be the same.

Now, if/when 4-8 CPU motherboards and OS's are the common gaming PC platform, that would change, but hey--not sure if/when that will be available for the $3,500-ish a top bleeding edge gaming platform normally is priced.

#12 I've got a strange feeling this might already be intergrated in some cool nex-gen console.

Especially considering that MS already decided to use U3 for all their internal work ....

hmm

#13 Ageia bought novodex, them supporting the xbox and xbox next has nothing to do with their endeavours in building a PPU.
For one, it's probably too late to integrate it for the planned launch in six months.

#14 Except from www.edharriss.com/xsi/version5.htm

Quote:
Softimage has added a new Dynamics engine (Ageia PhySx, previously called NovodeX) which is far superior to the previous one. Ageia PhySx physics simulation allows for complex Hard body simulations using complex polygonal shapes. At Siggraph, Softimage demonstrated this by showing hundreds of chairs bouncing off of each other and accumulating in a sink at the bottom of the scene. Performance in this engine is boasted to be up to 100 times faster than in previous XSI releases. However, they did not remove the old (ODE) engine. You can choose which one you want to use from a dropdown list.

Ageia, the PhySx company, is also the creator of the PhysX Chip, a PPU - Physics Processing Unit - unit, a GPU for physics if you like. An XSI compatible accelerator board should be out in the near future which will speed your simulations up even faster than they already are.

This chip not only accelerates RBD, but also Soft Body, Fluids, Hair, wire and Cloth. The PhySx SDK also includes a Particles API . While XSI takes advantage of just the Rigid Body Dynamics portion of this technology, one can only assume that other areas of XSI will utilize it in the future. Its already there, why not use it?

The PhySx engine includes new Convex Hull collisions, which is an intermediate step between primitive collision and actual shape collision and which provide better performance than actual shape. New PMap Actual Shape collision type. Options for actual shape resolutions. UI changes to collision layers. Setting initial states automatically at the first frame and UI changes to collision layers.

Also of note is that Sony and AGEIA have entered into a sublicensing agreement. Through this agreement, SCEI obtains sublicensing rights of the AGEIA PhysX SDK (I.E. NovodeX), the physics library developed by AGEIA. The licensing agreement allows for SCEI and AGEIA to optimize the AGEIA PhysX SDK for utilization in PLAYSTATION3, and will be included in the Software Development Kit for PLAYSTATION3. Frontline support will be provided by SCEI. It's nice to see that XSI's native RBD system is the exact same one that the PS3 will be using. This should save a few game programmers and artists some time down the road. For more information check out Ageia's PhysX site and watch this XSI Simulation video.

#15 So both the XBox360 & PS3 will have PPUs. How Very Interesting.

I just hope they get used for more than "better looking cloth/hair/water". Physics-based gameplay has started to appear in the indie scene... lets hope it shows up on the consoles soon.

www.ragdollkungfu.com
www.wikgame.com

There's also a bridge building game here on gg.com. But I don't recall seeing much actual physics GAMEPLAY in AAA titles... the major exception being HL2 with the grabby-throwy-gun thing... which was a lot of fun... particularly when people started leaving circular saw blades laying around. >:]

But even in HL2, it was mostly just a couple puzzles. Knock this into that so you can continue on your way. Picking up and throwing things was cool and all, but really just another form of ammo if you think about it... though I suppose you could use things as shields too. Hmm. Need to revise my opinion there.

--Mark

#16 Deciding what to do when one player has a PPU and the other doesn't is going to cause headaches in multiplayer gaming.

What happens when you cause that avalanche to go rushing towards the other player? Do the hundreds of perfectly modeled bouncing rocks just move down en masse. Do they cull half the rocks so the player without the PPU can still see them bouncing but fewer rocks means the CPU can deal with it (and obviously increase the chance of the player dodging out of the way).

It's going to be a tough call.

#17 That kind of physics calculation isn't going to work very in a network environment anyway... particularly if you throw Mac & Intel machines into the same network. Different FPUs (or PPUs possibly) means different floating point errors way out on that last digit. Those little differences can add up in a hurry.

IN THEORY, you could just send out an initial "impulse" over the network, and everyone's machine should be able to calculate the interactions identically (given that they all have the same floating point hardware). In practice, as soon as one of those rocks bounces off a player, it changes everything. And in a networked environment, everyones idea of where that player is can be a little different... so everyone gets different versions of events.

Physics over a network is still largely a pipe dream. Oh, some people around here have been messing around with OGL on torque multiplayer... but I'll be suprised if they've overcome the problem I mentioned above. And once you're clients are out of sync, you can't promise the initial "give everyone the same impulse and it'll all work out" thing any more.

#18 Which kind of defeats the point really. Why spend a couple of hundred quid on a product that you'll only be able to use about 50% of the time?

Personally I'm not convinced about them, even in a single player setting. All it means is that we'll get more games that feature you having to knock down a big wall with tumbling blocks in order to progress further into game.

#19 HULK SMASH PUNY SOFTWARE PHYSICS

#20 They can't be all that expensive if they're going to be in the PS3 and XBox360. The base console is what, $200 in both cases? Economies of scale, sure... loss leader, sure... Even so, I can't see them being all that pricey, particularly when one of the above articles mentions them being mounted directly on motherboards.


Network games may end up being limited to "better looking particle systems" (hair/water/cloth again)... but anything designed to be played without that network should be golden.

Multiplayer console games often are just multiple people sitting at the same box, so you might be able to dodge that bullet too. The only other thing you could do would be to have physical events that players couldn't interfere with once started (back to the single impulse starting a chain reaction that can stay in synch across multiple clients)... leaving players watching a pachinco game through an unbreakable window (or whatever).

That or you can find was to simply gloss over some differences in a given environment... grabbing something out of a pile of freshly-fallen rubble will always do X, that sort of thing.

Meh.