Xenos tessellation unit - what have we learned?

Dr. Nick

Veteran
Well so far I know that it's there and that it isn't as flexible as a geometry shader. You could use it to particle instancing but that is the only use that has been hinted at.

What I'm really wondering about is using it to do tessellation in games. We have seen a supposed example of it's use before the 360 launched and the use of displacement maps to accomplish it but we haven't really heard much about this outside of a more recent gamefest presentation. Microsoft has brought it up again so they must think that it can be used in games. I would like to ask if you guys think it will ever be used the way it was "intended" in games or is it likely a feature that will never really be used this generation?
 
Last edited by a moderator:
Well so far I know that it's there and that it isn't very flexible.
In what manner do you think its not flexible?

Its also already being employed in current titles - Viva Pinata implemented it last year for the terrain rendering system.
 
If you actually want to do tesselation then the tesselator is a much, much more performance oriented mechanism than doing it in the GS.

You may want to take a look at the following link that discusses the tesselators capabilities, and the implementation in Viva Pinata. Although this is talking about Radeon HD 2000 Series implementation the capabilities are pretty much translatable.

http://ati.amd.com/developer/techpapers.html
 
If you actually want to do tesselation then the tesselator is a much, much more performance oriented mechanism than doing it in the GS.

You may want to take a look at the following link that discusses the tesselators capabilities, and the implementation in Viva Pinata. Although this is talking about Radeon HD 2000 Series implementation the capabilities are pretty much translatable.

http://ati.amd.com/developer/techpapers.html

Care to speculate whay happens on the G8x series in Viva Pinata (in terms of tesselation)?

Does it simply not get the effect? Or will it try to pull it off in the GS?
 
Based on the Viva presentation, it looks like the tesselator could have been put to good use in Sega Rally Revo for the deformable road.

You'd think some of the 360 exclusive developers would have put it to more use by now though. Couldn't it have been used in Forza 2 or PGR4 to get rid of the polygon edges in wheel wells, guages, headlights... etc?
 
I looked at that and other tessellation related PDFs and I have to say, what on earth is holding devs back from leaping on this? What are the gotchas?

At the most basic level, adaptive tessellation looks like a way to replace LOD systems entirely.

As far as I can tell the key requirement is to create low-poly meshes that are parameterised. For animated stuff, particularly, this is presumably quite a change in the art pipeline - requiring entirely new tools. I've got no idea how much of a deterrent that is...

Perhaps the key problem here is that it will only work on Xenos and R6xx or newer - so it's unfriendly for cross-platform developers. The fallback for other GPUs would appear to be non-existent - the developers would be forced to create a traditional LOD-based system etc.

Jawed
 
I looked at that and other tessellation related PDFs and I have to say, what on earth is holding devs back from leaping on this? What are the gotchas?

Maybe it has to do with current pipelines, workflow, existing in-house libs or fresh (read buggy) API's?

We'll see what Rare and Lionhead come up with next year...
 
Maybe it has to do with current pipelines, workflow, existing in-house libs or fresh (read buggy) API's?

We'll see what Rare and Lionhead come up with next year...

Standard..

As a development house art is what costs you the most nowadays and, especially for multiplatform, you're not going to be particularly interested in further extending your art budget by getting your team to re-create all the assets & extra content (not to mention training them to use the new tools, which you also need to pay for, to generate the parameterised data) just for a single SKU & for something that the vast majority of users will never even notice the difference wrt using conventional LOD..

Unless you have Halo3's or Killzone 2's budget & platform exclusivity that is..

& even then you have to invest a considerable amount of time/resources into R&D, building your new production pipelines & tools, training etc..
 
Wasn't it laos not possible to do AA, or atleast tiling with tessellation? But since most devs don't seem to use tiling all that much anyway I don't think that it keeps them from not using tessellation, atleast not as much as having to change the art pipeline...
 
I looked at that and other tessellation related PDFs and I have to say, what on earth is holding devs back from leaping on this? What are the gotchas?

I don't really know it, but I'd blindly bet on performance issues.
 
Wasn't it laos not possible to do AA, or atleast tiling with tessellation? But since most devs don't seem to use tiling all that much anyway I don't think that it keeps them from not using tessellation, atleast not as much as having to change the art pipeline...

Where in the world did you ever get that idea?

:LOL:
 
Where in the world did you ever get that idea?

:LOL:

I don't know. I just remember that there was some discussion long time ago that when using some capability of the xgpu you could not do tiling or it would cost to much or something, but don't realy remember what it was about so I though it was tessellation, so I take it that is wrong...
 
I don't really know it, but I'd blindly bet on performance issues.
LOL. No.

The demonstrations we've done for HD 2900 have this pushing over 1 Billion poly's per second, thats actually at the peak rate of the setup engine (you have to net off backface culled poly's). Xenos's throughput rate will be no different, just about 1/3 less due to the lower clockspeed.
 

I'm guessing it will actually be PC developers that will really put this tech to good use. I'd imagine that Nvidia will eventually have one as will if they don't already. I will say that I'm more interested seeing it used on character models and objects that are the center of attention like cars, maybe Turn 10 will make use of it. I certainly hope to see more post about this aspect of these "next gen GPUs".

I don't know. I just remember that there was some discussion long time ago that when using some capability of the xgpu you could not do tiling or it would cost to much or something, but don't realy remember what it was about so I though it was tessellation, so I take it that is wrong...
You were probably thinking of Mem Export which is also something that we don't see talked about that much...even though PC cards should now support a feature that is like it or is exactly the same.
 
I'm guessing it will actually be PC developers that will really put this tech to good use...
Take a look at Natasha's presentation that was in the link earlier. It includes a "potential future DirectX pipeline" (as documented by MS) that includes a Tesselator...
 
You were probably thinking of Mem Export which is also something that we don't see talked about that much...even though PC cards should now support a feature that is like it or is exactly the same.


Yeah, that is right, that was the one. Two features that we don't really hear that much about and are not used so I forgot which was which, cheers...
 
It would be nice if some Devs chimed in about this feature, and how it could be (or is) used.

Unfortunately, I'm guessing that only 360 exclusive titles would even consider using it, and given how a lot of the exclusive titles aren't tiling I feel there is not much hope for the 360's tessellation unit.
 
What we have learned about tesselation over all the years that it's been available in various pieces of hardware, is that nobody gives a damn about it. Now only if IHVs learned the same...
 
Back
Top