VIDEO: https://vimeo.com/323361180 Variable rate rendering Always - - PowerPoint PPT Presentation

video https vimeo com 323361180 variable rate rendering
SMART_READER_LITE
LIVE PREVIEW

VIDEO: https://vimeo.com/323361180 Variable rate rendering Always - - PowerPoint PPT Presentation

VIDEO: https://vimeo.com/323361180 Variable rate rendering Always grouping similar work items No rasterization Real-time rates (50 ms or less per frame) Groups similar work items Enables efficient implementation of:


slide-1
SLIDE 1
slide-2
SLIDE 2

VIDEO: https://vimeo.com/323361180

slide-3
SLIDE 3
  • Variable rate rendering
  • Always grouping similar work items
  • No rasterization
  • Real-time rates (50 ms or less per frame)
slide-4
SLIDE 4
  • Groups similar work items
  • Enables efficient implementation of:
  • Variable rate rendering
  • Foveated rendering
  • Checkerboard rendering
  • Any analytic or random pattern
slide-5
SLIDE 5

FOVEATED VRR (Variable Res. Rendering)

slide-6
SLIDE 6
slide-7
SLIDE 7
slide-8
SLIDE 8
slide-9
SLIDE 9

TIP: as NN input, find scene properties that can be mostly represented with a continuous function.

slide-10
SLIDE 10
slide-11
SLIDE 11
slide-12
SLIDE 12

Performance impact BVH vs. SMG

BVH – Linear SMG – Sublinear

slide-13
SLIDE 13
  • Spatial denoise
  • NN approximate energy at surface
slide-14
SLIDE 14

GOOD FOR:

  • Static scenes
  • Can compliment lightmaps;

by vectorizing soft shadow regions.

BAD FOR:

  • Dynamic scenes
  • Very small primitives
slide-15
SLIDE 15

A viable high performance substitute for:

  • Bidirectional PT
  • Metropolis light transport

Finds up to 70% more paths than unidirectional path tracing.

slide-16
SLIDE 16
slide-17
SLIDE 17
slide-18
SLIDE 18

GPU 1 GPU 2 … GPU N

  • Offline voxel data interpolation
  • A bit of overfitting is welcome
  • Each voxel can be processed by a

different GPU, training scales linearly!

slide-19
SLIDE 19
slide-20
SLIDE 20
slide-21
SLIDE 21

Rage, Megatextures | Id Software

slide-22
SLIDE 22
slide-23
SLIDE 23
slide-24
SLIDE 24
slide-25
SLIDE 25
slide-26
SLIDE 26
slide-27
SLIDE 27

Animation guided by NN inputs QNM model size: ~5 KB QNM primitives: 9 Polygonal model size: ~1 MB

(vertices, normals, texture coordinates)

Polygonal primitives (triangles): 31 415

slide-28
SLIDE 28
slide-29
SLIDE 29
slide-30
SLIDE 30

Questions

slide-31
SLIDE 31

Email: nejc@lightmass-dynamics.com Twitter: @nejclesek