Forget Havok 2! I want NovodeX + PPU!

'Course, if they open-sourced the client, they might actually have *time* to work on the servers.

hay guys don't wory i will right a sfotware emulater for this PPU!!!11oen AND it will run on a P8 no prbolem without ANY EXTRA HARDWARE!!!! OMG!!!11

...

Seriously, though, this kind of upgrade would be phys-tastic. I'm all for >31-prim vehicles... then, finally, top-secret project Mobile Party Unit-01 will be fully linkable.

/still listening to airwolf theme (SID-version)... going on 29 minutes...

Unless the gain in processing power allows to use less servers.

Besides, adding a top-notch video card to use it's vertex shaders can't be THAT much cheaper than a PPU card.

Considering that the physics in SL is done server side, a PPU could be very beneficial to the grid itself, a dedicated physics processor, as opposed to software physics (our current physics model) would allow more dedication of the main processor to other assets, and would be the server equivelent of a 3D card, considering the servers themselves wouldn't benefit from graphic cards, as they don't use them.

A client computer would not need this... though this would add to the total cost of a purchased sim.

Can we PLEASE get this moved to Feature Suggestions? It's not a Technical Issue and is important enough to be seen as a suggestion!

Don't discuss... Vote Vote Vote!

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

Rathe's proposal would be entirely awesome, and then we could see where the VC went!

/esc

Wow, Neil! *clapclapclap*

I also suggest this thread be moved to Feature Suggestions, as it points a way to a solution in the future, not a problem in the present!

NovodeX does not require the hardware accelerator, the whole thing is done in software. Hardware acceleration would be optional, something they could move to in the future. It makes sense to move something like physics to dedicated hardware for the same reasons it made sense to move graphics processing to dedicated hardware - first the custom blitters for moving 2D windows around and such, then it was 3D processors that would do 3dfx's custom thing and later OpenGL/DirectX. Gravis and one or two others made hardware-accelerated sound cards that would play tracked music natively so games wouldn't have to spend time on synthesizing music (or relying on shitty MIDI sound.)

So, sometime in the future, LL could move from Havok 2 to NovodeX. (They are going to implement Havok 2, no way some voting proposal is going to get them to dump all that hard work and make us put up with Havok 1 even longer.) NovodeX would be run software-only at first. In a few years, when the PPUs become good and cheap, perhaps they can install them in the simulators.

BTW, according to Philip, the interest list is the big bottleneck, not physics. The interest list is what simulators spend the most time working on. Havok 2/NovodeX will eventually allow us to do much cooler stuff more easily, like (eventually) having realistic hair/clothing, ropes, multijoint linked objects (think: vehicles with a REAL suspension), have some sort of realistic fluids (water that acts like water), and so forth. That all takes extra work though. The immediate reason to move to Havok 2 is to cut down on server crashes, reduce border crossing headaches, decrease the tolerance gap that makes physical objects not able to get very close together, eliminate "deep thinks," build vehicles and other physical objects that mix phantom and non-phantom prims in the same link set (so you could have a 60-prim vehicle but only have five or ten prims collidable), and probably a few other things I don't know about.

Moving to PPUs would just make all that cool stuff work better, but it isn't necessary in the near term since NovodeX can run 100% in software.

Ealier in the thread I promised to post more information when and if I could. I dont have much additional information to add however I do have a few links where you can see it in action. One of the companies I have been working for has released a video of the Novodex SDK in action.

http://www.airtightgames.com/currentproject.html

Three videos showing metal bending, massive numbers of physical objects, fluid simulation, linked joint dynamic impact seperation etc.
and...

http://www.gdhardware.com/files/ageia/Airtight/DivxPhysxAirtight720x400.avi


Enjoy!

This is an un educated guess but would this not allow the potential to run say 4 sims per server machine, effectively reduceing hardware count to 25% of whats currently needed to run the SL grid? (aka reduceing costs)

PS Neil the demo looks amazing.

I just watched a video from gamespot (its a bit old) but it contained a representitive for Agiea which stated that the price should be "around the same as mid range graphics cards" my guess would be $200 a pop at that price, in terms of corporate money, the upgrade would be more then afforadible and as previously mentioned, as easy as plug'n'play asueming there is a transistion to Agiea's engine in the medium future.

Just caught this post. Very cool physics engine and I am SO for it. I was thinking about the implementation of this though. For example, if someone blew up a tank of water in a building area and the water started spewing to everywhere and everyone, some people might not be too cool with this being in the middle of their build projects. Pieces of "physical" carnage nailing people and pushing them to other sims and such. You get the idea. Yea, it would be some funhouse but you have to set a limit to what you are capable of somehow or another to keep a level of sanity in the place. There's going to have to be some thought into this as far as how, what, who, when and where. If this were the case then it might not be as fun as one would hope it to be but at least you have the power of a better engine.

Thoughts on this?

I caught the Ageia demos at Quakecon this Saturday and I was very impressed with it. I've put most of my thoughts on what I saw in a response on Hiro's blog.

I saw the Castle demo and the Boulder demo in real time and both are fairly impressive. The Castle demo reminded me a lot of the Rampart SL, except it used hundreds of bricks. The Boulder demo was far more interesting: there was a thousand objects, probably more, rolling down a hill. Under normal software emulation with a good graphics card, the computer churned the scene around 5-10fps. When the physX and hardware acceleration, it jumped to 40-50fps.

The rep shot out that the hardware accelerator could handle 15K physical objects per second. Isn't that how many prims a sim can handle, and that's without scripting and such going on?

I think Jaynius is right. Judging from some of the conversation of it's limits in networking and such (which IIRC would require the server to be accelerated), I do think it's a great server-side solution, without having to have the client needing to upgrade.