OpenGL Geometry shaders: poor performance on NV GF9800 GT
- Thread starter const
- Start date
nbohr1more
Newcomer
So the only major geometry initiative that Nvidia has championed recently is Vertex Texturing and even that was awful on 1st gen products (has it improved?)... Why is Nvidia so against improving Geometry, I would think that now that they are the only major IHV without an x86 business that they would want to "cut the cord" a little more... I will admit that the Tessellation PPT they put out was actually more impressive than the "Old Man Face" video that AMD used to show off their tessellation...
nbohr1more
Newcomer
The R520 series did not have Vertex Texture Fetch while the NV40 does. Nvidia was promoting the feature as a reason not to get ATI's hardware... http://developer.nvidia.com/object/using_vertex_textures.html
I'd say the GS is the Microsoft approach. Hardware guys never really liked it.
The funny thing is that ATI 'software' implementation (pixel shaders + R2VB) was often faster than nvidia's approach.
nbohr1more
Newcomer
So VTF, regardless of how poorly implemented, appears to be nvidia's last push for something that affects geometry... Nvidia has been getting the word out early for DX11 features (including tessellation) so no matter how AMD centric things seem to have become Nvidia appears to be on board with it all... maybe they've been cowed into submission by Microsoft after struggling with Vista drivers (just a crazy conspiracy theory) 
...
...Microsoft works very closely with the IHVs and I don't think they can add anything to DirectX that is not backed by at least two IHVs.
It's not like GS is a bad idea for things like rendering to cubemaps and doing things like point sprites in a unified way. It's just bad for large geometry amplification. But that was never the intent.
nbohr1more
Newcomer
I'm going to push my luck and say that when a former ATI employee says something is a MS approach that there may be some fire to go with the smoke. But if you has any ties to "another IHV"
we might have some confirmation one way or another...
PhilTaylor
Newcomer
I'm going to push my luck and say that when a former ATI employee says something is a MS approach that there may be some fire to go with the smoke. But if you has any ties to "another IHV"
yes, GS was part of the original D3D10 spec proposal.
the feature pipeline is:
game developers inform MS they want features x,y,z
Microsoft makes spec proposal to IHVs
Microsoft and IHVs negotiate what stays in/out of the spec
Spec gets ratified
Work commences
New OS platform with new D3D ships
New IHV hardware becomes available.
the goal of this process is a sw feature set that works independent of individual hw or hw implementation. features must be implementable by all IHVs. Microsoft owns the API, Microsoft drives the spec process.
the original IA and GS proposal actually went quite a bit further, but got trimmed down due to "hw guy" input from both ATI and NV.
nbohr1more
Newcomer
Yikes! I hope most of those developers have a "reasonable" notion of what would be the most efficient hardware implementation of their intended result... :S ... Meh, I'm sure the IHV's veto a metric ton of suggestions...
PhilTaylor
Newcomer
yes, asks that are truly unreasonable get filtered out. the merely difficult is where the rub lies.