AvP Tessellation o.0 AMAZING

[Cenelder]

im not sure if this was already posted, but just WOW
I cannot wait to try this game out on my dx11 compatible machine at home!

http://www.youtube.com/watch?v=dVo_5VCWIzM

I dont even quite understand how this is done, but i love it. He said there are MORE POLYGONS than pixels on his screen *mmmmmm*

[crazyfool]

thats crazy, especially on the environment when it tessellates before your very eyes. Is this gonna be possible on current gen consoles not just pcs?

[Cenelder]

this is PC exclusive. PC's are 2 generations ahead of consoles right now (consoles are using a form of DX9 while PCs are now onto DX11)

So i did a bit more reasearch on how they achieve this and here is another video. It seems they use paralax mapping WITH tesselation to create true 3D which is actually FASTER than straight paralax... just... wow....

http://www.youtube.com/watch?v=6Wp4Y-u8-Qw

i love dx11!!!

[JacqueChoi]

fuh KYEA!!!!

[Ritorian]

Ya i think they're using that shit with CRYSIS 2. which looks frigin crazy.

like i said before, games r gettin frigin nutz. still... I want my virtual reality helmet, with an FPS game, still waiting on that :) hopefully within the next 10 years. (fingers crossed)

[cookepuss]

Don't wanna sound like the pessimist here, but the idea of dynamic tessellation is nothing new. To a lesser extent, this was done in Quake 3's curved surfaces. To a larger extent, it was done in Shiny's Messiah. Dave Perry was trumpeting this sort of technology as far back as 1998. (http://www.dperry.com/archives/articles/dp_speeches/messiah_what_yo/) In-between, games like Republic & Jurassic Park Trespasser took on the issue of scalability from different angles.

Every couple of years, somebody comes along and bangs the tessellation drum. AvP is just the latest.

As far as this being truly real-time...... I'll believe it when I see it. There's no such thing as magic. More likely than not, there's a lot of pre-calculation, caching, & other such trickery. Just take a look at ZBrush's Decimation Master. Try decimating a 14mil poly mesh. The pre-calculation phase alone is going to take a couple of minutes. Now, imagine doing that per frame. No programmer in his right mind is going to brute force the tessellation. Chance are high that stuff is going to be broken down long before it ever hits the screen.

I've programmed stuff like this before. There's a lot of calculus derived math & algorithms that make a brute force implementation prohibitive.

Naturally, the distinction between brute force and using a decimation cache is trivial once you see the per frame results on screen. For speed purposes though, there's a lot of programmer voodoo going on to make it work.

Like the saying goes, it doesn't have to be correct. It just has to look correct.

Anyway, let's see how this all pans out. AvP isn't the first on the scalability bandwagon and it certainly won't be the last. Seeing as how ZBrush & Mudbox artists typically deal with insane poly counts and have to ultimately compromise on the low poly retopo, this sort of tech should be more mainstream.

BTW, there's no reason why scalable technology can't be implemented on current consoles. That's the reason why scalability is so desirable. As long as you've got the basic RAM and math operations in place, you could even port this sort of technology all the way down to the PSP. The real challenge would be data storage of the high res assets, making delta compression a prerequisite.

[blankslatejoe]

Cooke, I think the difference here is that this is done on the graphics card itself. I'm sure you could do something with brute force before, but with the advent of geometry shaders in Shader model 4.0 there's been more and more of this sort of stuff popping up.

I had a chat with our graphics engineer here about it. From what I could decipher from the conversation, Directx 11 is further parrallelizing what's being handled by the graphics card and is taking advantage of dedicated spots on the card for vertex deformation. CPU and the GPU don't have to sit and chat about vertex data before rendering...which probably helps a lot toward making this sort of thing feasible in realtime...though it's probably still not going to do great things to your framerate.

Then again, AVP doesn't seem to have huge vistas.. If there's ANYWHERE you'd get away with this stuff, it's in a situation where you can only see 90 meters away down a tight corridor...you know?


Anyway, I was skeptical of Crysis too, but it did what they promised (with their geometry/vertex shader fun)...it just took another two years for PCs to come out that could run it!

[Maph]

Looks absolutely amazing. Unigine had a spiffy DX11 demo as well. Gorgeous stuff!

When it comes out, I may consider replacing my quadro with a dx11 consumer card as a matter of fact. Just so I can play with it!

Still waiting for the day that realtime raytracing is feasable though. :)

[zeke3d]

nice, thanks for sharing. I am seeing graphics just using dx9, the test bed for a lot of this stuff, just not optimized, begin taken to levels I could not imagine and as, if not more jaw dropping than this. its more than just how much crazy tech and memory your video card is capable of, like Cooke mentioned there are a lot of of parallel technologies that in the end will make the whole thing come together as I am beginning to see happen in the current gen, I can only begin to dream about what a fully decked out dx11 game could look like.

[JacqueChoi]

This seems to hit the nail on the head of something I think will define Next-Next-Gen (PS4 XBox720).

PRODUCTION EFFICIENCY.

Getting rid of normal maps as we currently use them. Great for screenshots, horrible when deformed.

The next step would be to get per-poly collision working at a level, where we don't have to Zbrush clothing anymore, and let cloth deformers be driven by physics.


If our industry is going to continue to grow, we are really going to need to somehow scale character production times back to under a week per character (like it was back on Ps2).




I'd like to see Motion Blur, 8x Anti-Aliasing be handled by the hardware as well. DX12 I suppose.

^_^

[zeke3d]

yeah, I don't see it working like that anytime soon, before actual cloth deformation, I would think we would see fluid simulation like normal maps being computed, where an artist specifies a tension map, a simple map that limits the flow of the movement of the cloth, it would then take into account the speed, thickness, and gravity to create the look of flowing clothing on a per pixel level(similar to older generation water simulation). the geo would deform on a per vertex level based on the map in a similar fashion, just to give it some give and subtle secondary motion, nothing to crazy. everything in the next gen seems to be pointing to things taking longer and requiring more people, similar to the auto industry, only the biggest fish will survive by eating the smaller fish, to where creating a studio from scratch would be difficult because of the start up costs, and given peoples tastes will only heighten even in the casual gaming market, outside of developing nations. projects with teams of 1000 or more. who knows, just speculation.

[Cenelder]

This thread is becoming boring... more talk about how awesome DX11 is and how it blew your mind plz

too much math... pffft i thought this was an artists thread!

[Beartastic]

This looks totally sick, new shaders that aren't one-vert-in-one-vert-out will be mad! Maths maths in your thread!

[Maph]

...new shaders that aren't one-vert-in-one-vert-out ...

Magic words.:excite: :excite: :excite: