*rename* Balancing Work between Xenon & Xenos

and that is a problem. I think tesselation would have taken off faster if developers could tap into the large amounts of dx 9 hardware that supported it. I believe its everything from the radeon hd 2800 and up that supported it along with the low end verisons of the 2800s ? Thats alot of card many more than dx 11 has out right now.

They could have simply left a profile for dx 9 tesselation that may have been a step down but would have gotten the ball rolling . Right now and for most likely the next year or two your leaving out more hardware with some form of tesselation than there are dx 11 parts in the market.

By having the dx 9 tesselation as a path you coujld have gotten more developers on the 360 using it and thus more pc games using it. Even if dx 11's version is much better.
As far as I know the only chips that have a hardware tesselator like or better than the one in the Xenos are Direct X 10 and 11 cards. Direct X didn't include support for tesselation until 11 but I am on the same page with you that it's a shame that there isn't support for ATI's pre dx11 tesselation.
 
As far as I know the only chips that have a hardware tesselator like or better than the one in the Xenos are Direct X 10 and 11 cards. Direct X didn't include support for tesselation until 11 but I am on the same page with you that it's a shame that there isn't support for ATI's pre dx11 tesselation.

aye your right they are dx 10 cards. My bad !

But yup , more hardware = more support. They could have supported it with advanced profiles for the dx 11 verison. That way we could have seen tesselation sooner.
 
and that is a problem. I think tesselation would have taken off faster if developers could tap into the large amounts of dx 9 hardware that supported it. I believe its everything from the radeon hd 2800 and up that supported it along with the low end verisons of the 2800s ? Thats alot of card many more than dx 11 has out right now.

Developers are welcome to use it if they really want to. ATI released an SDK a long time ago.
 
and that is a problem. I think tesselation would have taken off faster if developers could tap into the large amounts of dx 9 hardware that supported it. I believe its everything from the radeon hd 2800 and up that supported it along with the low end verisons of the 2800s ? Thats alot of card many more than dx 11 has out right now.

In general, developers have less incentive to support optional hardware features.

But things which are really necessary, and which are offered in DX9 as a hacky workaround, do get used a lot (http://aras-p.info/texts/D3D9GPUHacks.html); tesselation doesn't seem to be one of them.
 
I have a small query (er.. two?) why haven't we seen more tesselator work from the 360 to date?

The second question is: the tesselator on the 360 is a hardware solution right? It's physically on the die? Does it cost system resources to use it? (I.E. RAM, GPU\CPU ?).

I guess i've always been lost as to why many titles (even first party) simply don't mess with it.

Does anybody know if AVP is? (Given that it was designed for DX11 and tessellation in mind.)
 
The second question is: the tesselator on the 360 is a hardware solution right? It's physically on the die? Does it cost system resources to use it? (I.E. RAM, GPU\CPU ?).

I'm not an Xbox 360 developer, but AFAIK it is indeed a hardware unit. However it only does one thing: it generates domain coordinates from your control points. Your vertex shader still has to perform the extra work of calculating actual positions for the tessellated vertices, and also any displacement mapping if you're doing that. Plus if you want to do skinning you can't just skin the control points like you can in DX11, which means you instead have to apply skinning for each tessellated vertex. The domain coordinates aren't stored in memory so you don't get extra memory overhead, I'd imagine it's mostly extra shader overhead.

There's also the more important issue, which is that using it for any geometry outside of terrain would typically require big changes in the art/tool pipeline.
 
I know that Alan Wake uses tesselation, but is there any particular reason that they use software tesselation instead of the hardware unit's? Is it because it is too weak, or not flexible enough?
 
Last edited by a moderator:
I'm not an Xbox 360 developer, but AFAIK it is indeed a hardware unit. However it only does one thing: it generates domain coordinates from your control points. Your vertex shader still has to perform the extra work of calculating actual positions for the tessellated vertices, and also any displacement mapping if you're doing that. Plus if you want to do skinning you can't just skin the control points like you can in DX11, which means you instead have to apply skinning for each tessellated vertex. The domain coordinates aren't stored in memory so you don't get extra memory overhead, I'd imagine it's mostly extra shader overhead.

There's also the more important issue, which is that using it for any geometry outside of terrain would typically require big changes in the art/tool pipeline.
You can skin before tessellation with Xenos it's just not automatic like with DX11. Execute the vertex shader and export skinned vertices and tessellation factors. In a second pass tessellate and run the equivalent of DX11's domain shader.
 
I know that Alan Wake uses tesselation, but is there any particular reason that they use software tesselation instead of the hardware unit's? Is it because it is too weak, or not flexible enough?
They do it on the Xbox 360's CPU? Do you have a link?
 
They do it on the Xbox 360's CPU? Do you have a link?

I believe they've only ever mentioned CPU tessellation once, at IDF '06:
http://www.anandtech.com/tradeshows/showdoc.aspx?i=2841&p=2

The game will actually spawn five independent threads: one for rendering, audio, streaming, physics and terrain tessellation. The rendering thread is the same as it would be in any game, simply preparing vertices and data to be sent to the GPU for rendering. The audio thread will obviously be used for all audio in the game, although Remedy indicates that it is far from a CPU intensive thread.
 
Considering the tessellation comment was made at an Intel event and referred to the apparently canceled PC version I don't think anything can be concluded with respect to the suitability of hardware tessellation for Alan Wake. Plus, a lot can change in 3+ years.
 
hehe, certainly. :)

Probably irrelevant for a trailer, but the one from X10 shows no LOD popping on the mountains as the camera passes over them. Some of the earlier vids did show it to some extent, usually on the outer edges of the screen.

http://ve3d.ign.com/videos/play/51557/PC/Alan-Wake/Trailer/E3-2009-Trailer/Flash-Video

^ Near the beginning around 13 seconds just underneath the lighthouse, and later on at around... 25s on the far left edge and heck, even the middle of the screen! Mountains can grow :p
 
You can skin before tessellation with Xenos it's just not automatic like with DX11. Execute the vertex shader and export skinned vertices and tessellation factors. In a second pass tessellate and run the equivalent of DX11's domain shader.

Never thought of that, but it makes sense. Not quite as nice as doing it in single pass with a domain shader, but definitely better than skinning the tessellated verts.
 
hehe, certainly. :)

Probably irrelevant for a trailer, but the one from X10 shows no LOD popping on the mountains as the camera passes over them. Some of the earlier vids did show it to some extent, usually on the outer edges of the screen.

http://ve3d.ign.com/videos/play/51557/PC/Alan-Wake/Trailer/E3-2009-Trailer/Flash-Video

^ Near the beginning around 13 seconds just underneath the lighthouse, and later on at around... 25s on the far left edge and heck, even the middle of the screen! Mountains can grow :p

Yeah I noticed that too, at E3 09 it was the biggest thing to point out that the trailer was actually running on 360 (prior trailers post E306 didn't have any pop in either)
 
Back
Top