Page 830 of 830 FirstFirst ... 330730780820828829830
Results 16,581 to 16,594 of 16594
  1. #16581
    I agree mipmap for secondary rays is much better(although perfect mirrors or transparent surfaces should still sample the original).
    Regarding memory usage, mipmaps take more memory in the first place. Or do you do a kind of precomputation like for adaptive subdiv and only load the mipmaps you really need?
    If it takes more memory, do you relly on OS/driver to layout the memory according to the frequency at which each tile/mipmap is accessed (fequently used tiles in dedicated gpu memory, less frequently used in system, rarely used in pagefile)? Or do you implement your own memory management?



  2. #16582
    Mip levels are never picked explicitly, you do this by estimating the ray's footprint (for example with ray differentials. Mip mapping makes sense for perfect mirrors too, when they're far away or convex.

    Texture tiling and caching is done entirely via OpenImageIO's TextureSys, which to my knowledge should be the same code that's being used by SPI and Arnold. This is not using any system paging but a custom cache. The main work inside Cycles is to add ray differentials in all the missing places, calculate texture differentials from ray differentials and to manage mip map paths and color spaces.

    No GPU.



  3. #16583
    With No GPU you mean that it won’t support GPU?

    Then how do Vray does it? I think Vray have mi al support in their GPU version and that was a leverage specially because it saves GPU memory, but I may be wrong.

    Cheers.



  4. #16584
    Originally Posted by juang3d View Post
    With No GPU you mean that it won’t support GPU?
    Correct. GPUs can't read from hard drives. The problem with GPU rendering is that there you need to have all necessary data in place before launching the render kernel, where the CPU can fetch from the hard drive at any time in the render process.
    Then how do Vray does it?
    I don't know. They're not sharing details, only very high level descriptions: http://on-demand.gputechconf.com/gtc...-rendering.pdf

    The aforementioned deferred shading approach could work for final frame rendering (not interactive), but that would be work I can only look into after I wrap up what I'm doing currently. One step at a time.
    Last edited by skw; 16-Aug-17 at 05:18.



  5. #16585
    Can’t you process the mipmap before the render starts? In the scene preparation?

    This may not work in real time, but it would be the same as with displacement, the displacement is prepared for the point of view used to start the render, if you move then the displace could be at lower resolutoin or you could even see the camera fustrum, I suspect in Vray they something like that, again I may be wrong.

    Pleas bear in mind this is a humble idea of a person with zero knowledge on GPU programming

    Cheers.



  6. #16586
    Well, you only need to load the textures that are being hit by a ray. The problem is that you don't know what that is until after you have traced that ray and ran the shader, which is, you've rendered it. The camera frustrum can tell you that only for direct visibility, but not for shadow rays, reflections, volumetric, SSS, multi-bounce diffuse, etc. In path tracing, any pixel can interact with any part of the scene, and you won't know which it is until you render it.



  7. #16587
    Understood, it's way more complex than what it seems, thanks for the explanation!

    Cheers



  8. #16588
    @ skw
    Since you have been looking at an Embree integration for Cycles have you seen this (these) paper(s) before?
    http://rapt.technology/
    They all claim a performance plus over Embree once AVX512 comes into play.



  9. #16589
    I've seen those papers, haven't looked at them too closely. Since Embree is under active development, my hope is that once it's integrated, future improvements like this are done for us for free by Intel



  10. #16590
    Member Ace Dragon's Avatar
    Join Date
    Feb 2006
    Location
    Wichita Kansas (USA)
    Posts
    28,012
    Improved triangle sampling for meshlights (by Stefan Werner)
    https://lists.blender.org/pipermail/...st/099537.html

    This will increase rendertime, but it will be offset by better sampling in regions near the emission surface (especially in volumes as Brecht notes in the patch discussion). Users should only see an improvement as areas further from the light will still sample with the old method.
    Last edited by Ace Dragon; 17-Aug-17 at 13:05.
    Sweet Dragon dreams, lovely Dragon kisses, gorgeous Dragon hugs. How sweet would life be to romp with Dragons, teasing you with their fire and you being in their games, perhaps they can even turn you into one as well.
    Adventures in Cycles; My official sketchbook



  11. #16591
    Excuse my ignorance question.
    Can cycles at any point in their development be as fast and equal when compared to Arnold when it comes to refinement and accuracy ?
    Can Cycles go from 'adequate' to 'stellar' performance with an awesome end result ?
    Or are there limits to source code that will make it impossible to achieve that development ?

    Thx in advance.



  12. #16592
    @carlosan

    Can you put some examples about what you are referring to?

    Why do you think the end results you can get from Arnold are so different from what you can get from Cycles?

    Honest question here, no pun intended or anything similar.

    Cheers.



  13. #16593
    Member Ace Dragon's Avatar
    Join Date
    Feb 2006
    Location
    Wichita Kansas (USA)
    Posts
    28,012
    Arnold is also a pure pathtracer, so nothing it can do should be impossible to implement in Cycles (the only possible exception is if they have any specific methods that are patented).
    Sweet Dragon dreams, lovely Dragon kisses, gorgeous Dragon hugs. How sweet would life be to romp with Dragons, teasing you with their fire and you being in their games, perhaps they can even turn you into one as well.
    Adventures in Cycles; My official sketchbook



  14. #16594
    Originally Posted by Ace Dragon View Post
    Arnold is also a pure pathtracer, so nothing it can do should be impossible to implement in Cycles (the only possible exception is if they have any specific methods that are patented).
    But in Europe you don't have software patents.



Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •