Wow this tessellation thing is pretty cool

[disappoint]

http://www.youtube.com/watch?v=EfrSaIY0YQA&feature=related

Nvidia water and terrain demo showing the increased detail tessellation can do. I used to think tessellation was just a gimmick, and I was like, why not just increase the polygon count, same thing right? But this shows it really adds a lot of polygons while improving frame rates.

Anyway I can't wait to see this level of detail in some games already.

 

[MJinZ]

It's too bad that games which use these types of technology are all fluff and no substance.

If Diablo III supported these kinds of things............ drool..........

 

[SHAQ]

Too bad Just Cause 2 doesn't have that level of water realism. Is Tesselation a better approach than CUDA?

 

[BenSkywalker]

Is Tesselation a better approach than CUDA?

It is a different approach. I think ideally using tesselation for surface detail then a CUDA based solution for translucency effects would be the best of both worlds, but extremely intense on a computational basis. The big advantage for CUDA over tesselation for a game like JC2 is that there are ~6Million DX11 parts and ~150Million CUDA parts.

 

[happy medium]

Too bad Just Cause 2 doesn't have that level of water realism. Is Tesselation a better approach than CUDA?

it looks very similar to me .

Watch this video.

http://www.youtube.com/watch?v=_pBLtZUJ9OA&NR=1

 

[zerocool84]

Too bad Just Cause 2 doesn't have that level of water realism. Is Tesselation a better approach than CUDA?

Different cus tessellation anyone can use. CUDA only Nvidia can use so tessellation will get adopted faster.

 

[Rubycon]

Where are these demos for download?

I have the rocket sled demo and actually got a sub 18s score but could not upload due to stupid network restrictions.

I don't see these demos on nvidia.com for download but have all the ones already listed.

 

[Sylvanas]

Too bad Just Cause 2 doesn't have that level of water realism. Is Tesselation a better approach than CUDA?

Tesselation and 'CUDA' are two very different things. While not related to Tesselation, after reading this DirectCompute vs OpenCL vs CUDA test it's clear CUDA is the slower approach when computing Geometry on the GPU which is exactly what PhysX does. So....perhaps if PhysX was done in OpenCL (thus making it an open standard) we may infact have better performance in PhysX games than we do now (massive penalty when enabled).

 

[evolucion8]

It is a different approach. I think ideally using tesselation for surface detail then a CUDA based solution for translucency effects would be the best of both worlds, but extremely intense on a computational basis. The big advantage for CUDA over tesselation for a game like JC2 is that there are ~6Million DX11 parts and ~150Million CUDA parts.

Unfortunately, CUDA is the slower approach compared to OpenCL like Sylvanas stated, and I doubt that the 8800 series will run that decently, and being DX10 only, pitiful. After all, most of those 150 Millon of CUDA capable cards are from low and midrange 9600/8800/9800 series in which the HD 4800 series outperformed in market share (Steam Data), even though none of the cards mentioned have relevance in the DX11 market share, only the GTX 4x0 are capable of using Tessellation the way it supposed to be like the HD 5x00 series, period.

 

[StinkyPinky]

http://www.youtube.com/watch?v=EfrSaIY0YQA&feature=related

Nvidia water and terrain demo showing the increased detail tessellation can do. I used to think tessellation was just a gimmick, and I was like, why not just increase the polygon count, same thing right? But this shows it really adds a lot of polygons while improving frame rates.

Anyway I can't wait to see this level of detail in some games already.

That water looks awesome.

I think because of freaking consoles a lot of games wont bother. When the PS4/New Xbox come out I think there will be a huge jump in the number of games with these types of features.

 

[alfa147x]

Why don't more mainstream games use features like this?

 

[Cookie Monster]

Tesselation and 'CUDA' are two very different things. While not related to Tesselation, after reading this DirectCompute vs OpenCL vs CUDA test it's clear CUDA is the slower approach when computing Geometry on the GPU which is exactly what PhysX does. So....perhaps if PhysX was done in OpenCL (thus making it an open standard) we may infact have better performance in PhysX games than we do now (massive penalty when enabled).

Well, its performance conclusion is based on a GTX260. I dont think one can assume that this will be the same for the GTX4x0 series since the latter is more heavily geared toward CUDA. So until there are numbers for Fermi based cards, I think its too premature to say "A is better than B" at this point in time.

 

[Udgnim]

Why don't more mainstream games use features like this?

developers do not want to spend too many resources on features that a majority of its customer does not have access to

think that's why tessellation support has been limited to character models (correct me if I'm wrong here) so far

 

[darckhart]

the motion of the water is a little too periodic. ruins realism for me. way better than fsx though. terrain was also very nice. but unless i sit there and watch this stuff passively, the hit in performance when gaming, is it worth it?

 

[Tsavo]

Not to mention all the hardcore gamer FPSer's that buy the absolute latest and greatest then run in in DX9 with all the textures set to low and no AA or AF so they can be more leet killing machines.

 

[Sylvanas]

Well, its performance conclusion is based on a GTX260. I dont think one can assume that this will be the same for the GTX4x0 series since the latter is more heavily geared toward CUDA. So until there are numbers for Fermi based cards, I think its too premature to say "A is better than B" at this point in time.

True, but in all likelihood OpenCL will scale just as much as CUDA will. The main GPGPU advances made in Fermi is perhaps concurrent Kernals which will benefit OpenCL just as much as CUDA. The results we see here could just be expanded to a larger scale.

 

[frostedflakes]

Pretty cool stuff, why didn't tesellation catch on sooner? ATI had hardware tessellation back in like 2001 (they called it TruForm) and I think a handful of games supported it. Hopefully we'll start to see it used in a lot more games.

 

[zerocool84]

Pretty cool stuff, why didn't tesellation catch on sooner? ATI had hardware tessellation back in like 2001 (they called it TruForm) and I think a handful of games supported it. Hopefully we'll start to see it used in a lot more games.

Cus Nvidia made sure it wasn't in DX10.

 

[Tsavo]

Pretty cool stuff, why didn't tesellation catch on sooner? ATI had hardware tessellation back in like 2001 (they called it TruForm) and I think a handful of games supported it. Hopefully we'll start to see it used in a lot more games.

ATI spent $500,000 putting into their hardware and $5 marketing it to devs.

 

[JRW]

Too bad Just Cause 2 doesn't have that level of water realism. Is Tesselation a better approach than CUDA?

JC2's water looks pretty good with gpu simulation enabled but its to bad it doesn't react to your presence.

 

[Sylvanas]

Pretty cool stuff, why didn't tesellation catch on sooner? ATI had hardware tessellation back in like 2001 (they called it TruForm) and I think a handful of games supported it. Hopefully we'll start to see it used in a lot more games.

Truform, and ATIs subsequent Tesselators in the R600 and beyond are not the same as the Tesselators present in todays modern Cypress and Fermi GPUs. The DX11 spec implemented a programmable geometry pipeline with Hull and Domain shaders, also new instructions were added with the capability for more complex geometry (Catmull-Clark algorithms). That's why todays DX11 games can't 'run Tesselation' on previous ATI GPUs like the 4800 series.

Not to mention OpenGl 4.0 now has support for programmable Tesselation. In the future we will see it take off much more than Truform did.

 

[The Milkman]

Can somebody link me to an article explaining the performance cost VS just using lod's?
I was under the impression tessellation just made life easier for devs, rather then allow for more poly's at a lower cost to performance.

 

[gorobei]

there's a bit of confusion and conflation. cuda and tessellation are 2 entirely separate things.

dx11 tessellation(aka subdivsion/limit surfaces) just breaks larger polygons into smaller ones based on certain math functions. So instead of the exploding oil drum being made out of a 12 sided cylinder, it is rendered as a 24 or 48 sided cylinder. ALL dx11 cards can do this.

cuda just enables many parallel calculations of non render things.

the water in JustCuz is a tessellated grid of polygons, and a displacement map is applied to make wave crests/troughs. On cuda enabled cards, the gpu is used to calculate the displacement map as a quasi-simulation. Old school game programmers would just make a greyscale bitmap to define the displacement or use a fractal generator function. The cuda method just does a lot of extra busywork to arrive at roughly the same result. If it actually calculated interaction between the water and the boat, then it would be significant. But if it did so, it would most likely bring the framerate to a snail's pace.

 

[evolucion8]

The Tessellation model under DX11 isn't totally programmable AFAIK, its a fixed function geometry pipeline which uses Hull and Domain shaders to adjust the geometry, but that's it. The Tessellation can't be accessed or programmed directly like they can with pixel/vertex shaders. I don't know if I'm totally wrong, but I think is the same approach as the R2VB did in the DX9 era.

 

[Seero]

There are 3 parts to the test