Geometry Shaders
- Thread starter rwolf
- Start date
This is just too easy to resist....
1. Open up your favourite web browser.
2. Go to http://www.google.com
3. Type in Geometry Shader or D3D10 Geometry Shader into the search field.
4. Start clicking and reading.
1. Open up your favourite web browser.
2. Go to http://www.google.com
3. Type in Geometry Shader or D3D10 Geometry Shader into the search field.
4. Start clicking and reading.
Neeyik said:This is just too easy to resist....
1. Open up your favourite web browser.
2. Go to http://www.google.com
3. Type in Geometry Shader or D3D10 Geometry Shader into the search field.
4. Start clicking and reading.
Yes, I deserved that. Thank you very much.
I knew I was going to get nailed when I posted the thread.I did google the topic by the way and there wasn't much to go on.
Last edited by a moderator:
I've working on some D3D10 code featuring geometry shaders in between browsing forums today...
I'm not looking at creating cutting edge graphics yet - I'm more interested in the greater level of information available throughout the pipeline. My current GS's have just been computing properties about the geometry and using that to influence rendering/rasterization.
You could always check out the SDK for examples of fancy GS usage. The object space motion blurring sample (new in February) is very neat - and it's documentation has a lot of detail.
hth
Jack
I'm not looking at creating cutting edge graphics yet - I'm more interested in the greater level of information available throughout the pipeline. My current GS's have just been computing properties about the geometry and using that to influence rendering/rasterization.
You could always check out the SDK for examples of fancy GS usage. The object space motion blurring sample (new in February) is very neat - and it's documentation has a lot of detail.
hth
Jack
I'll try to discuss this topic anyway...
I was just looking at an NVidia demo for the 7800 called luna. The demo contains an extremely realistic looking female character with animated facial features. When you switch wireframe mode on, you can see that hundreds of thousands of polygons are required in the facial area to do this. However, with geometry shading, most of these polygons could be generated on the fly.
A big problem for realistic character animation (especially facial expressions) comes when the radius of a curved surface changes - in current animation this causes the surface to deform unrealistically due to the fixed polygon count used. With geometry shading you could adjust the number of polygons generated for the surface to compensate for this.
In summary, we can expect to see much more realistic character animation using geometry shading
I was just looking at an NVidia demo for the 7800 called luna. The demo contains an extremely realistic looking female character with animated facial features. When you switch wireframe mode on, you can see that hundreds of thousands of polygons are required in the facial area to do this. However, with geometry shading, most of these polygons could be generated on the fly.
A big problem for realistic character animation (especially facial expressions) comes when the radius of a curved surface changes - in current animation this causes the surface to deform unrealistically due to the fixed polygon count used. With geometry shading you could adjust the number of polygons generated for the surface to compensate for this.
In summary, we can expect to see much more realistic character animation using geometry shading
Yes, I suppose that will happen...
BUT as I mentioned before, the general consensus seems to be that the GS won't be best suited as a progammable tesselator for HOS. Yes, it can do that, but No, "first gen" hardware probably won't have the performance for it. Rather, the GS allows for topological information to either be retrieved or generated - something that is much less "wow" but still very cool.
Then again, I suppose we'll have to wait for some real hardware samples to see
Jack
BUT as I mentioned before, the general consensus seems to be that the GS won't be best suited as a progammable tesselator for HOS. Yes, it can do that, but No, "first gen" hardware probably won't have the performance for it. Rather, the GS allows for topological information to either be retrieved or generated - something that is much less "wow" but still very cool.
Then again, I suppose we'll have to wait for some real hardware samples to see
Jack
http://www.gamedev.net/reference/programming/features/d3d10overview/
Looks like the geometry shader is going to be more significant then I thought.
Looks like the geometry shader is going to be more significant then I thought.
Well the single-pass cube map rendering should be both an interesting quality improvement as well as the obvious performance increase.rwolf said:
It's very common for cubemapped reflection in games to be rendered at much lower resolution and lower detail and less often (e.g. every 3rd frame). Partly just because you can (and people won't notice) as well as trying to offset the pain of re-rendering the scene 6 times.
So far I've seen a lot of people talk about single-pass cubemap rendering as a neat performance optimization. I wonder if it'll also translate into some improved visuals/reflections due to it being a less painful effect...
It's the same thing this line (from the WGF2 list) that I was referring to in my GDNet article about GS's (potentially) being able to do on-the-fly material setup.
Cheers,
Jack
