Kostas Anagnostou

@KostasAAA

Senior Rendering Engineer at . DMs open for gfx/rendering questions or for mentoring people wanting to get into the industry. Views my own.

Leamington Spa, UK
Vrijeme pridruživanja: svibanj 2009.

Tweetovi

Blokirali ste korisnika/cu @KostasAAA

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @KostasAAA

  1. Prikvačeni tweet
    27. lip 2019.

    I am refreshing the pin as a lot of people DM me to ask if the invitation is still open. It is a joy interacting with people on Twitter and I will keep it open for as long as I can.

    Poništi
  2. 2. velj

    Switched the toy engine to physical light units, not much has changed visually but it is fun to set light intensities in lumens and tweak actual camera settings. Also remember to add back the division by pi that you may have removed from the lighting calculations as a speedup.

    Poništi
  3. 29. sij

    Filament's documentation provides an extensive overview of a wide range of graphics topics including PBR, physical lights and camera, clustered shading, image based lighting, and importance sampling. It is a great learning resource.

    Poništi
  4. 26. sij

    This 2 level depth representation can speedup rendering a lot, but it is not a pyramid. We would still have to create a full Hierarchical-Z pyramid, successively downsampling the depth buffer in a compute or pixel shader, to use for occlusion culling purposes. (4/4)

    Prikaži ovu nit
    Poništi
  5. 26. sij

    If for example the min depth of all the pixels in the new tile is larger than the maximum depth stored in the corresponding Hi-Z pixel then the GPU rejects the whole tile. If not, then it updates the min/max value of the HiZ pixel and goes on to process the tile further. (3/4)

    Prikaži ovu nit
    Poništi
  6. 26. sij

    Every time a pixel tile, belonging to the same triangle, arrives, the GPU will use the min/max value in that buffer, that corresponds to the tile, to compare it with the min/max depth values of the pixel tile. (2/4)

    Prikaži ovu nit
    Poništi
  7. 26. sij

    Question from DMs: "So can the GPU automatically generate a Hi-Z pyramid?" The confusion comes from a GPU feature often called HiZ (esp for AMD GPUs): for every tile of pixels (say 4x4 or 8x8), the GPU stores a min and max depth value in a special buffer while rendering. (1/4)

    Prikaži ovu nit
    Poništi
  8. proslijedio/la je Tweet
    23. sij

    New blog post! In "Learning from data" we'll talk about vertex cache optimization again - starting from an algorithm built by a human, using analysis tools to understand the hardware behavior better, and using the machine to make the algorithm better.

    Poništi
  9. 23. sij

    This post describes a normal from depth technique that uses extrapolation of depth values to achieve higher quality reconstruction. Extrapolation of depth values can also be useful when conservatively downsampling depth buffers.

    Poništi
  10. 20. sij
    Poništi
  11. 19. sij

    It also mentions deep G-Buffers to store multiple layers of properties per pixel for reflection/refraction/transp and explains how to pronounce G-buffer, as "Ge" in the German alphabet to differentiate from another g-buffer introduced 3 years ago to store pixel coverage for AA.

    Prikaži ovu nit
    Poništi
  12. 19. sij

    This 1990 paper introduces G-buffers to cache the properties of a 3D geometry rendering pass (inc normals, depth, uv and id), on the CPU, so as to allow interactive image manipulation like shading, edge detection, and medical imaging

    Prikaži ovu nit
    Poništi
  13. 18. sij

    It's worth storing indices/loop counts you pass to shaders as ints instead of floats it'll save you some conversion instructions and maybe registers. Only make sure that you do it on both ends and not only in the shader, it'll save you some GPU hangs. Uhm, happened to a friend.

    Poništi
  14. 18. sij

    Ray-tracing the GI at 1/16th resolution and reducing the temporal feedback produces decent enough results on high detailed surfaces like the walls. Fixing a pesky division by PI fixes the light intensity as well. 😉

    Prikaži ovu nit
    Poništi
  15. 18. sij

    Testing raytraced GI after adding support for a structured buffer BVH. With such a big GPU load, power saving soon kicks in on the laptop and kills performance in the end. Using some temporal antialiasing the GI converges nicely though.

    Prikaži ovu nit
    Poništi
  16. 15. sij

    Tried to get the toy renderer to use dxc, but it fails horribly. I think I'm asking too much of the HD4000 and its ancient driver, I should be happy it runs DX12 at all 🙂. BTW 's post has good instructions on how to support dxc in your code

    Poništi
  17. 13. sij

    Recommended! We need conferences like these in Europe.

    Poništi
  18. 8. sij

    Tile classification can reduce divergence in a shader by grouping tiles of pixels requiring same shading and processing them with a set of simpler shaders. This post describes how it can be used in postprocessing, for DOF rendering.

    Poništi
  19. 8. sij

    This deck presents the current state of art in hair lighting, for single and multi-scattering, and summarises a geometric strand rendering approach. Includes a great visualisation of the 3 main components of single scattering on hair.

    Poništi
  20. 5. sij

    It is also a good idea to look for documentation on how compute shaders work, like , as they are "closer to the metal" than pixel/vertex shaders and also posts about GPU profiling, such as this , as they often expose how a GPU works.

    Prikaži ovu nit
    Poništi
  21. 5. sij
    Prikaži ovu nit
    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·