Forget Havok 2! I want NovodeX + PPU!

http://www.ageia.com/pr_03082005c.html

Who's with me?!

From the sounds of it, PPU sounds way more advanced than Havok 2. But according to the website, it says that PPU is a hardware based physics system while Havok 2 is software based which is easier to implement into SL... i think >_<.

Well actually NovodeX is a physics API much like Havok except it's multi-threaded and capable of using multiple processors. This will allow it to take advantage of hardware based physics processing once PPU's become available near the end of this year. To really take advantage of it you would need the hardware/PPU, but sign me up! The advantages in physics quality and speed this would offer Second Life are amazing!

Rathe you bastard... you're gonna make Andrew cry.

Ohmygod, I think I jumped the gun.

I just posted a question about this in Feature suggestions, before moseying over to the General forum.

either way. if it gets me vechicles with more than 31 prims i'll sacrifice a goat and pack my computer with Parmesan. after i get the cat hair out.

Sweet!

("ah screw you guys", can be heard somewhere in the llabs background..)

Well I spose now that they have announced it publicly I can say something

I'm currently working on the demo for this card and this physics engine. I can say several things about it but I have to get clarification from my boss about it first. But suffice it to say, this thing is going to rock the gaming world. Its amazing. We are doing things with it that havent been seen in realtime applications. Novodex is an amazing system, very fast and very flexible. The number of physical objects it can handle is insane. Software acceleration only I have seen scenes with better than 1,500 physical objects. Hardware accelleration is supposed to push that way farther. It is a standard physics api, no different from Havok only imho after using both havok 2 and Novodex, Novodex is FAR superior. I am not sure how hard it is to integrate it since we are using an engine that already had it integrated. Havok 2 was a pain in the ass to get in to the last project I worked on, the developers had months of aggrevation before it was stable and running at a decent speed. The thought of Novodex in SL makes my mouth water however.

Cheers

Neil, I understand that these cards would be necessary as a plug-in for games which make such heavy use of physics.

But, for an application like SecondLife, would the add-in hardware be required for people's home computers, or just for the servers? My understanding of Secondlife is that physics are processed, entirely, on the server side, and that only a person's home graphics card would make a difference in the quality of the visuals; the rendering.

Is this correct, and if the technology proves stable, would it mean that an all-around upgrade to the hundreds of servers in SL (instead of 10,000 to 20,000 end-user computers) could drastically improve the physics abilities in-world?

Or am I mistaken about the way the technology functions, and would this require upgrades on the part of each of the thousands of end-users?

That is an excellent question!

Based on the way the servers work with physics now it would be my assumption that the physics cards would be used as a server side solution. It would reduce the overall load of the server cpu's since those calculations are done server side on the cpu now. Though inversely there might be more information going to the client regarding physics.

Physics are just a sideline of what I do, I havent done the necessary research to give you a completely informed view of the situation as it pertains to sl. I am not completely certain the methods they are using now for client updates of the state of the phsycis simulation. I am not certain how much of an impact on the client side this would have.

In summation, I do believe it would be a server side upgrade if they were to use the PPU card. From what I understand once the hardware card is available it will be a matter of setting a flag in the engine constants to look for it, so pretty simple once the cards have been installed in the servers. However I also believe the card is pretty power hungry which may or may not require a power supply upgrade in the servers as well.

For what its worth...

Ohmygod though, that's why I was asking about it. How awesome would it be, to be able to put in a physics upgrade of that magnitude, and only require that the servers be upgraded? I imagine that a big determining factor in any new hardware like this, is whether the end-user is willing to pay for it? People playing SL already require both a highspeed connection, a decent amount of RAM and a graphics card above certain specs. I think that many players would balk about a significant improvement to SL, if it also required them to spend another US $100 or $200 in hardware upgrades in order to benefit.

If this is something that could be attained by some (pricey) upgrades, even to the hundreds of servers at LL, it would be a single centralized move, and therefore grant an immediate, drastic improvement in the SL experience to everyone, regardless of whether they themselves could afford a PPU card.

If the technology proves stable and achieves widespread adoption, perhaps it would be something to look at for SecondLife 2.0?

The improvements (if the talk about the upcoming PPU cards is to be believed) would be a significant enough jump to warrant an inclusion in the Big Two-Point-Oh, but beyond that, by the time we reach that point, it might be time for LL to consider changing their subscription policies. Whatever changes come about with 2.0, we know that they're going to have to be significant enough to deserve the numbered point release. Perhaps at that time, LL might also look and say, "Look, we've got a hardware migration scheme underway already, in order to constantly improve our server statistics overall. We have something in the works which would be attainable, would drastically improve the experience for everyone, but which would require a massive one-time hardware upgrade across the board. Venture capital isn't cutting it right now; would the community be in favoer of abandoning this move until a later date, or paying a one-time $10 (or $20) per account fee in order to upgrade our world to 2.0?"

Like Andrew said, who knows how many users or servers we'll have at that point? The pricing scheme would have to be low enough that people would pay for it, but high enough that it would cover or subsidize the cost of the upgrade.

Neat stuff though, Neil. If your boss allows you to, I'd be interested in hearing what else you can about this technology.

Moved to Technical Issues for further discussion

Ran across this article on Gamasutra, not much additional info but still nice to see.

http://www.gamasutra.com/php-bin/news_index.php?story=5080

I cannot post any additional information as of yet, however after this milestone is complete (4 weeks) I should be able to post some stuff

Cool, sounds like the physics equivilent of something else I've heard about regarding 3D rendering. But that would require definite hardware changes on every client...(think: a tiny processor for EVERY pixel on your screen to help your main GPU do 3D math). This supposedly will allow ray-traced quality (i.e. equal to or almost picture quality) in real time. If I recall correctly. Of course, the artists and modellers would have alot of work to do.

But with how long it's taking to get to Havok 2, I don't think we want to try and get LL to switch gears again!

Has Andrew commented about this anywhere yet? I'd absolutely love to see this in SL. It sounds so great!

"Forget about Havok 2! Now we want the P.P.U.!"

This is not a thread about technical issues.

I agree. Please move this to Feature Suggestions.

Show your support for this and vote for Prop 46!

http://secondlife.com/vote/get_feature.php?get_id=46

Folks, LL can't even sort itself out sufficiently to use the hardware that is already present on our graphics cards (+followup), despite the very graphically intense SL needing it so desperately.

Yet you expect them to adapt the code to totally new extra external hardware? This century?

I suggest that you add your votes to the calls for open-sourcing SL, because it's just not going to happen inside LL. With the best will in the world (and they do have plenty of will and vision), they simply don't have the manpower for anything substantial. 1.6 proved that in shedloads.

Incidentally, those very same hardware shaders that they're not using on our GPUs can be used to speed up physics engine calculations too, or pretty much any other highly parallel computations.

They're really just generic programmable vector units after all ... they just happen to be optiimized for vertex and pixel processing, but that's not the limit of their application. There's a lot of material out there on how to use them on generic scientific problems, and computing physics is a classic example of that.

Yes actually, hooking into this card is a flag in the Novodex engine, its already implimented in the sdk. LL would not have to do any more work than they would have integrating Havok 2. Integration of the sdk takes roughly a week or two in a standard game development environment, depending on the depth of the integration and the engine setup. In LL's case it would be a bit longer since its a vastly different environment than a typical game. Though the PPU and this SDK lend itself far better to SL's situation than does Havok 2. I have used both professionally and can attest to the strength of the Novodex engine over the Havok 2 engine.

I have seen things I never expected to see in games using this engine. It can handle things in realtime that are decently difficult to do in your average 3d package like Maya.

The PhysX card (PPU) is designed to greatly increase calculation speed of....

Rigid body dynamics
Universal collision detection
Finite element analysis
Soft body dynamics
Fluid dynamics
Hair simulation
Clothing simulation

Furthermore all the physics calculations (which are considerable) are now done on hardware freeing up the CPU to do more stuff. Like many more avatars in a sim than could be supported with Havok 2 since the cpu is completely free of physics calcs and the physics card is giving a 100x processing boost. So instead of the perhaps 64 avatars in a sim, think closer to 200 or so (assuming the physics calcs are the bottleneck).

All in all I have found working with the Novodex engine to be far more enjoyable than working with the Havok 2 engine. It is far more forgiving of mistakes than the Havok engine is.


Novodex Engine information and Demos

One more time - on the servers or on the clients?

I just want to make absolutely sure. Second Life has folks computer budgets pushed to the max as it is.

But if it's server side I would say go with the future instead of the past.

I'll answer it as straight as I can too so that nobody else is confused.

YES, THIS IS A SERVER SIDE SOLUTION.

In a few years, when the hardware is as common as 3D graphic cards are today, things might change. But since physic is already handled by the server, it would only speed that up. Users wouldn't even notice (except for the lag-free servers of course).

Remember that LL are using commodity boxes server-side to keep costs down, and because they have a ton of them they must be getting huge discounts on this hardware. Anyone want to guess at the cost of a grid server, which is just barebones CPU, memory, disk and Ethernet? No Microsoft Tax, no graphics cards, no CD or DVD drives (prolly network boot). Hey, it could be a *lot* less than $1000.

If the price of one of these PPU cards significantly pushes up the cost of a grid server, that's going to be very bad for LL finances. These won't be cheap cards if they're coming out of a new startup ... can they be cheap enough to not be signifficant on top of the (small) cost of a server?

If PPUs are costly, it might be worth mentioning that most highly parallel numerical algorthms can be implemented fairly easily on the vertex and pixel shaders of the lastest generation of GPUs, and these can be bought for well under $100. So that might be a viable alternative way for LL to accelerate their physics.

(And of course, the same could be done in our clients.)

Remember that LL are using commodity boxes server-side to keep costs down, and because they have a ton of them they must be getting huge discounts on this hardware. Anyone want to guess at the cost of a grid server, which is just barebones CPU, memory, disk and Ethernet? No Microsoft Tax, no graphics cards, no monitor, no mouse, no keyboard, no CD or DVD drives (prolly network boot). Hey, it could be a *lot* less than $1000.

If the price of one of these PPU cards significantly pushes up the cost of a grid server, that's going to be very bad for LL finances. These won't be cheap cards if they're coming out of a new startup ... can they be cheap enough to not be signifficant on top of the (small) cost of a server?

If PPUs are costly, it might be worth mentioning that most highly parallel numerical algorthms can be implemented fairly easily on the vertex and pixel shaders of the lastest generation of GPUs, and these can be bought for well under $100. So that might be a viable alternative way for LL to accelerate their physics.

(And of course, the same could be done in our clients.)