With just over a month to go until the Game Developers Conference, we already know that Valve will once again give great prominence to virtual reality. We don’t know if they will show a hardware-level prototype, but they will talk at length about how to render for VR.
Valve has been creating advanced prototypes of HMDs since mid-2013, more advanced prototypes than many that other developers have access to, and thanks to this they have obtained a deep knowledge of rendering features focused on VR that they are going to share with developers who are working or planning to do so in the field of virtual reality. The talk will start with the basic requirements of VR rendering, and will progress to more advanced topics focused on both performance and visual quality.
It is expected that the first generation of consumer HMDs will have to render around 4 million pixels at a minimum of 90 fps. Due to the wide FOV of these HMDs, we will notice that each pixel has less resolution than the equivalent on a monitor, and therefore requires better shading algorithms than in 2D mode. The render at higher resolution, with higher quality pixels and at higher refresh rates than was done until now makes us take a step back and rethink many aspects of the render. Some topics to be covered include: efficient rendering in stereo, reducing render latency, GPU saturation despite synchronization points, reducing the rendering cost of less priority pixels, specular antialiasing, restricted anisotropic lighting and other tricks and tips directly related to the performance and quality of rendering in virtual reality.