Graphics Programming weekly - Issue 41 — June 3, 2018


The Machinery Shader System (part 3)

  • shader authors can define variations of shaders with a number of systems
  • each system can inject code/resources/constants into the shader
  • a material allows specifications of which shaders belong together, which systems they use and allows to insert command at the correct time in the Frame
  • all of the resulting shaders get a shared resource binder and constant buffer to reduce


Adding texturing to a glsl path tracer

explanations how ray tracing shaders in OpenGL can access the necessary textures for all objects in the world using GL_ARB_bindless_texture


Pathtracer 13: GPU threadgroup memory is useful!

  • moving the scene information into group shared memory to speed up the ray tracing code a lot
  • problems on metal, slower unless passing data by value instead of by const reference


Pathtracer 14: iOS

  • look at the iOS implementation with an overview of performance tools for CPU and GPU


Hybrid Multi-Frustum Logarithmic Depth Buffer

  • now using Logarithmic depth buffer when available
  • writes custom depth in a pixel shader, this does disable early depth optimizations but still a performance win for their use-case


Witcher tricks - a bit about rendering the witcher 3

  • look at the implementation of Witcher 3 rendering from an outside perspective using RenderDoc
  • breakdown of rendering Frame structure
  • how normals are stored in the g-buffer
  • explanation of a number of techniques from the d3d disassembly
    • sun rendering
    • blinking stars
    • eye adaptation
    • tonemapping
    • vignette


Basis universal GPU texture format examples

  • discussion of tradeoffs for the basis file format
  • comparison of different images formats transcoded from the same source format


Denoising with Kernel Prediction and Asymmetric Loss Functions

  • extending previous work in CNN(Convolutional neural network) based denoising
  • using a modular architecture that improves temporal stability and detail preservation


Optimising path tracing with SIMD

  • how to optimize a Rust path tracer using SSE4.1


On Games’ Power Consumption and phones

  • the longer the CPU/GPU can idle, the less power is needed
  • optimizing a CPU bound game might cause more frames to be rendered (hitting 60 fps instead of 30). causing less GPU idle time and increasing power consumption
  • kernel heuristics might trigger higher frequency mode which increases battery usage significantly


Texture Distortion Faking Liquid

  • Tutorial about the implementation of water surface movement of using a flow map in unity
  • how to deal with deformation of normals using derivative maps


A Trio of New Nsight Tools That Empower Developers to Fully Optimize their CPU and GPU Performance

  • NVIDIA Nsight Systems visualizes system-wide application interactions across CPU and GPU
  • Nsight
    • Volta, Vulkan 1.1 and Cuda 9.2, are supported
    • user configurable memory view


Nsight Systems Exposes New GPU Optimization Opportunities

-how to investigate GPU starvation and detect CPU/GPU synchronizations and overview of other tools