- Feb 7, 2008
- 30
- 0
- 0
Simultaneous Multiprojection. Looks like they are years ahead of AMD.
I'll update my Twitter Twitter.com/doingandy
Stencil will help with decreasing pixel shader work, while multi-projection decreases vertex shading work. You want to use both for best performance!I would still like to see just how much of a performance improvement Nvidia's MRS/multi-projection stuff brings in VR compared to simply using a properly implemented stencil mesh.
It seems like all of their comparison have been to full frame rendering, which is somewhat misleading.
Stencil will help with decreasing pixel shader work, while multi-projection decreases vertex shading work. You want to use both for best performance!
Simultaneous Multiprojection. Looks like they are years ahead of AMD.
It's completely different. The stencil mesh trick avoids rendering pixels you don't see because they are simply outside of the HMD field of view. This is something its supported by any GPU out there. Lens matched shading, as described by NVIDIA, modifies the way the GPU shades pixels so that the shading density is constant throughout the field of view. Right now the GPU uniform shading rate across the screen becomes very uneven after lens distortion, with lots of pixels at the center of the image and very few at the periphery, which forces to shade more everywhere (i.e. at higher resolution) just to avoid having gigantic pixels in the image. With lens matched shading this problem is gone and you get performance back. So overall you want to use all 3 techniques on NVIDIA HW:It's probably worth noting here that nvidia's multi projection tech enables multiple new features. One of them is single pass stereo (what you're talking about), another one is lens matched shading (what I'm talking about). Lens matched shading doesn't really seem that different from simply using a stencil mesh to me.
Single pass stereo is definitely quite awesome, although a slightly ironic thing is that it probably leads to subpar SLI scaling (since you would be forced to go back to two passes with SLI).
Are you saying that this technique will be in use at once or soon? Will games, etc, need to be programmed for this to work? How long will this take?It's completely different. The stencil mesh trick avoids rendering pixels you don't see because they are simply outside of the HMD field of view. This is something its supported by any GPU out there. Lens matched shading, as described by NVIDIA, modifies the way the GPU shades pixels so that the shading density is constant throughout the field of view. Right now the GPU uniform shading rate across the screen becomes very uneven after lens distortion, with lots of pixels at the center of the image and very few at the periphery, which forces to shade more everywhere (i.e. at higher resolution) just to avoid having gigantic pixels in the image. With lens matched shading this problem is gone and you get performance back. So overall you want to use all 3 techniques on NVIDIA HW:
1) single pass stereo to send geometry to the GPU for both eyes only once to save on bandwidth and vertex shading.
2) stencil mesh to avoid shading pixels in the far periphery of the image that are not visible in VR
3) lens matched shading to avoid over shading the image to compensate for lens distortion.
On current AMD HW only the stencil mesh trick can be used, so NVIDIA is making a big leap forward in VR rendering here. Of course Polaris might change everything again.
You can still use it to render the left half of both views on a GPU and the right half of both views on the other GPU![]()
Are you saying that this technique will be in use at once or soon? Will games, etc, need to be programmed for this to work? How long will this take?