Unigine Heaven Benchmark with DX11 Tessellation

It looks fantastic! But does it actually improve gamplay? To borrow an example from Elder Scrolls, a dungeon wall in Oblivion looks more detailed than a dungeon wall in Daggerfall but the impact on gameplay is zero. Don't get me wrong, I approve of better visuals.
 
DarrisBard said:
It looks fantastic! But does it actually improve gamplay? To borrow an example from Elder Scrolls, a dungeon wall in Oblivion looks more detailed than a dungeon wall in Daggerfall but the impact on gameplay is zero. Don't get me wrong, I approve of better visuals.
Click to expand...

You might get better performance on same visuals or better visuals. Don't think it will improve gameplay other then adding eyecandy and perhaps performance. Here's two official DX11 video's of the Dirt2 game. Doesn't give much comparison, so it won't say much. Nice to watch though. :)

Dirt 2 DX11 gameplay:
http://www.youtube.com/watch?v=XOLlgHM2a-k

Dirt 2 DX11 tech demo:
http://www.youtube.com/watch?v=WODI-Wvafqc
 
doesn't really matter since 90% of gamers are going to stick with their current cards that work just fine.

Just as a for instance, I can run batman, or resident evil 5 at 1920x1080 on high with my "old" 8800GTX. Maybe it was just a killer card way ahead of its time, but its going to take something more than just another FPS for me to shell out over 300 bucks for a vid card that can make things more realistic. Fry cry was neat and all, and the reason why I got my card, but games aren't improving that much.

Look at how many people are probably still running WoW on a VERY old vid card... These people have no reason to switch and this will keep the prices of these fancy vid cards high. Make some games the masses will enjoy and you'll sell more cards.

Sometimes I think people in marketing should take a cue from tech people instead of just calling them nerds.
 
The AvP is cool, but those models are retarded dense when tesselation is on. They're denser than a typical 3D-model they use for CG in a big-budget film.
 
Pyrolistical said:
This is not displacement maps! That's old tech.

Displacement maps is just a texture rendering trick to make it appear it has depth, when it in fact does not. You can't get extreme displacement without it looking wrong.
Click to expand...

Wrong - a quick google would have shown you that, eg.
http://download.blender.org/documentation/htmlI/ch11s03.html
http://www.zbrushcentral.com/showthread.php?t=20310
http://www.blogcdn.com/www.joystiq.com/media/2006/01/Unreal3engine-berserker_cg.jpg - famous ue3 image in which all that detail comes from displacement maps


Tamlin_WSGF said:
Here's the tessalation shown in Aliens vs. predator. Running on AMD hardware. Should give you some indication of the use in real games. :)

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

That's a better example then I've seen elsewhere. The way it dynamically increases detail in a smooth way is more impressive then just having a two or three levels of detail for an object. It would be good to see it in action to see if you get any funny graphical effects due to the shape of the object changing on the fly.
 
Last edited:
as fancy and good looking as tessellation is, i think it's 50% useless as the newly shaped surfaces are still going to be flat to the game, and people running up the new stairs will probably look weird :p it would be nice if the whole environment became aware when something becomes tessellated
 
Digital Viper-X- said:
as fancy and good looking as tessellation is, i think it's 50% useless as the newly shaped surfaces are still going to be flat to the game, and people running up the new stairs will probably look weird :p it would be nice if the whole environment became aware when something becomes tessellated
Click to expand...

Apart from anything else the game's collision engine, and physics engine won't have a clue about the tessellated geometry so in real games the non tessellated versions will have to look very close to the tessellated ones or it won't work. E.g. if the stairs are a slope and only made stair like by the tessellator then the physics engine would see them as a slope make a box slide smoothly down them not bump like real stairs. Another example would be the tessellated spikes on dragon - they wouldn't be collideable.
 
sbuckler said:
Apart from anything else the game's collision engine, and physics engine won't have a clue about the tessellated geometry so in real games the non tessellated versions will have to look very close to the tessellated ones or it won't work. E.g. if the stairs are a slope and only made stair like by the tessellator then the physics engine would see them as a slope make a box slide smoothly down them not bump like real stairs. Another example would be the tessellated spikes on dragon - they wouldn't be collideable.
Click to expand...

Why wouldn’t the physic engine know the difference? If you look at the DX 11 video for Dirt2 they are basically using DX11 tessellation to enhance they way physics look, IE the water movement and flag movement.
 
GabooN said:
Why wouldn’t the physic engine know the difference? If you look at the DX 11 video for Dirt2 they are basically using DX11 tessellation to enhance they way physics look, IE the water movement and flag movement.
Click to expand...

I agree the effect of the water or flag flapping is a physics effect - but if it's generated using tessellation then the polygonal mesh to simulate ripples in the water or the flag waving only exist in the graphics card. Hence physics in the traditional sense (e.g. object one hits object two) cannot take account of tessellated geometry because the only place it exists is in the graphics card - not in the main memory version of those objects which is what the physics engine has available to it to calculate the collision.
 
sbuckler said:
Apart from anything else the game's collision engine, and physics engine won't have a clue about the tessellated geometry so in real games the non tessellated versions will have to look very close to the tessellated ones or it won't work. E.g. if the stairs are a slope and only made stair like by the tessellator then the physics engine would see them as a slope make a box slide smoothly down them not bump like real stairs. Another example would be the tessellated spikes on dragon - they wouldn't be collideable.
Click to expand...

Not necessarily. In the case of unigine's stairs, you are correct, but you could easily have a very rough stair model that would look horrible visually, but contain plenty of information for a physics engine. The physics engine doesn't need to know how the stones bump, for example, and would likely just really slow things down and have objects get stuck on them. Same with the dragon, the dev would probably just surround it with an invisible wall anyway. Unigine was really showing what tessellation is capable of, not necessarily what would make it into games. I fully expect tessellation to be used as a high quality bump mapping alternative, which from the looks of DiRT 2 and AVP seems to be what devs are going for as well.
 
You need a benchmark that has similar output geometry detail for both code targets. If the DX11 tessellation creates 4M triangles from 1M then the DX10 version should be supplied with 4M triangles instead of the previous 1M.

It's nice knowing how much turning it on will impact performance and quality but is the method of increasing detail better than the previous brute force method.
 
It'd be interesting to see if GPU physics could be combined with tessellation to increase physics detail. Subdivision of collision boxes, for example, could be used to bounce particles off a model with greater accuracy as you come closer. Doing this by shunting geometry backwards and forwards between the CPU and GPU (or rather, between GPU and main memory) would usually prevent this, but if everything is done on the GPU with everything already resident in memory, it could work.
For example, you could bounce sparks off a rough or angular surface with proper reflection angles, rather than just a psuedo-random spray pattern of a flat collision surface.
 
I must admit to being a little bit confused. (It happens often, though (me being confused, not admitting to it)) :p

It's the rope that did me in. Lemme explain.

ATi has given us Truform since at least the Radeon 8500 days (I remember my 8500 had the "dolphin" truform demo). How is this "tessellation" any different from "truform"?

Tessellation (or should I say Truform) is easy to understand from a polygonal perspective - you simply break up existing polys and this allows for smoother, more detailed lighting and curvature without the 3D modeller having to do extra work.

However, how does DX11 tessellation know that the rope is supposed to be spiral? How does it divide a low number of polygons into a higher number that happen to be spiral?

You see what I'm getting at? The rope example merely looks like traditional LOD, and the modeller had to go and model the high-detail spiral in the rope ANYWAY.

What am I missing here, peeps?
 
Grimgor said:
ATi has given us Truform since at least the Radeon 8500 days (I remember my 8500 had the "dolphin" truform demo). How is this "tessellation" any different from "truform"?
Click to expand...

TruForm is a form of tessellation. Tessellation is the generic name, TruForm was ATI's name for the tech. :)

However, how does DX11 tessellation know that the rope is supposed to be spiral? How does it divide a low number of polygons into a higher number that happen to be spiral?

You see what I'm getting at? The rope example merely looks like traditional LOD, and the modeller had to go and model the high-detail spiral in the rope ANYWAY.

What am I missing here, peeps?
Click to expand...

As has been said about a bajillion times in this thread, there is a height map that tells DX11 how to shape the extra polygons. It uses the exact same resources as bump mapping (model + height map), only instead of faking depth using lighting tricks it actually creates polygons to make depth.
 
dx11 and tesselation would rock in rpg and mmog games.
simply not acceptable to have game makers as bioware making a game for 2009 that simly sucks grapichal wise. (dragon age)
why no dx11?

Tessealtion allows dynamic interaction where houses and ropes for example to look like those should look.

Love the heaven benchmark.
more fun to run than games.
 
kllrnohj said:
As has been said about a bajillion times in this thread, there is a height map that tells DX11 how to shape the extra polygons. It uses the exact same resources as bump mapping (model + height map), only instead of faking depth using lighting tricks it actually creates polygons to make depth.
Click to expand...

I know that - what I want to know is how you make a spiral heightmap? Looking at the non-tess wireframe, the rope hasn't been modelled in a spiral fashion in any way.

12573821526izM8p4LAl_1_10_l.png


How does it get from before to after using a hightmap? Something seems fishy if you ask me, and I'm not convinced.

ETA: Unless it's possible to layer a texture around a tube "primitive" in a spiral fashion, then I guess it would work. But I'm deffo no expert on this at all.

P.S Correct me if I'm wrong - assuming it's a hightmap, it seems like height maps can also act as "depth" maps. You'd expect the rope to expand outwards with a height map, but it is actually still the same diameter, which implies that the height map is applied as a "negative".
 
Last edited:
displacement map... it doesn't have to be flat.

It could be a cylinder, too...

a cylindrical displacement map...
 
Ohhhh I see what you mean. Just like the normal square texture would have diagonal lines on it, so would the height map!

CHA-CHING! That penny made a loud noise as it dropped! :eek:
 
sbuckler said:
That's a better example then I've seen elsewhere. The way it dynamically increases detail in a smooth way is more impressive then just having a two or three levels of detail for an object. It would be good to see it in action to see if you get any funny graphical effects due to the shape of the object changing on the fly.
Click to expand...

I noticed the dynamic increase of detail too. I wonder if that might boost the performance, when the increased detail doesn't seem to be there when not needed. Would be cool if the closer you get to an object, the stronger detail you'll get as well. :)
 
Any word on the average performance hit for most games? I can see how it is in the benchmark, but anything implemented yet?
 
kllrnohj said:
As has been said about a bajillion times in this thread, there is a height map that tells DX11 how to shape the extra polygons. It uses the exact same resources as bump mapping (model + height map), only instead of faking depth using lighting tricks it actually creates polygons to make depth.
Click to expand...

But why? What is the point in applying an extra processing step that can be achieved simply by having a high-res model in the first place. As far as i can see, the only advantage of this method is that it is degradable by having a convenient 'on/off' switch.

Besides the real beauty here is not how the geometry is obtained, but the means by which it is apparently morphic at a scene wide level. It must be quite a sight to behold in wireframe mode as polygons slip in and out of existence at the mystical whim of some unseen renderer :)
 
Last edited:
dragonico said:
But why? What is the point in applying an extra processing step that can be achieved simply by having a high-res model in the first place. As far as i can see, the only advantage of this method is that it is degradable by having a convenient 'on/off' switch.
Click to expand...

With tessellation you actually get a slider, not an on/off. So the number of polys for something like, say, a rope can be very few when far away, but when you pull out your sniper scope and zoom way in the polys can go through the roof. Think insane precision LOD rather than sudden model swaps. The other reason is that for someone to CREATE a high quality rope would be a massive waste of time and money. Making a texture and a height map and then letting DX11 do the work is much simpler and *cheaper*.

Or, think of a character. The character is going to have several LOD models for the console versions, and it is definitely going to have a height map for bump mapping. Developers can take the work that was already done and give PC gamers a huge jump in model quality with very little coding work and *no* work from the artists. It is entirely a win-win. Gamers get far more detailed models, and developers get to save boatloads of money by not having to actually work for the increase in IQ.
 
kllrnohj said:
The other reason is that for someone to CREATE a high quality rope would be a massive waste of time and money. Making a texture and a height map and then letting DX11 do the work is much simpler and *cheaper*.
Click to expand...
Erm, with current modelling techniques, the highest LOD model is created first (usually with detail FAR above what any card would be able to render), and then used to make the displacement/etc maps for a lower poly version thatis actually used in the game. This is the same technique that would be used to create tessellated models, so the amount of modelling work will not change (the process of turning a high-poly model into a mapped low-poly one isn't completely automated, but it's not exactly labour intensive).
Basically, the actual process of making a model will be pretty much identical to how they are made currently.
 
EdZ said:
Basically, the actual process of making a model will be pretty much identical to how they are made currently.
Click to expand...

I guess that's promising news? I mean if it takes the same amount of labour to use tessolation versus not using tessolation and using tessolation results in a better image qualtiy developer's won't really many time/budget or other reasons not to do this?
 
The stairs are sloped without DX11 turned on? I don't get it, steps aren't anything new. Why are they sloped without DX11?
 
This game engine is fucking jaw dropping awesome :eek:

Would love to see a good MMO like Ever Quest ]l[ or WoW-2 use this engine.
 
ew1075 said:
The stairs are sloped without DX11 turned on? I don't get it, steps aren't anything new. Why are they sloped without DX11?
Click to expand...

I was thinking the exact same thing.
There are plenty of realistic stairs in DX9 games so how come all of a sudden we need tessalation so the stairs don't look sloped?
 
Yea dunno, I am not buying fully into this, it feels like a con when you think about it.


Some objects clearly change shape between DX10 & DX11. There is no way the GPU can decided a square object can suddenly look round and in perportion etc, there must be substantial manual input.

Assuming there is manual input, surely the shapes in DX10 can be made to look the same although the process to get the result will be different and sure the performance "may" be down.

To me DX11 is a more efficient API that can run a similar level of eye candy faster than DX10 and more efficient way to render it, but not specifically more eye candy as this demo is trying to make us believe.
 
You must log in or register to reply here.
Back
Top