Page 2 of 2 FirstFirst 12
Results 16 to 21 of 21

Thread: LW 2018 .0.5 transparency not visible in viewport

  1. #16
    Electron wrangler jwiede's Avatar
    Join Date
    Aug 2007
    Location
    San Jose, CA
    Posts
    6,507
    That answer doesn't make sense. Whether the geometry is transferred within VBOs (via DMA), or streamed via sending primitives down the pipe (typically PIO), is orthogonal to how the surfaces (shaders) applied to the geometry sets are handled.

    It also fails to acknowledge that VBO-hosted geometry offers major performance benefits over streaming geometry in modern graphics hardware. Being forced to use streaming geometry transfers for cases when multiple transparency states are present has a significant impact when dealing with large, complex geometry data sets.

    I've attached a (private) link to a video of Layout 2018.0.5 manipulating a 3M-poly object using both VBO and streamed geometry modes, and the difference is fairly stark. Though the video may not show it clearly, using VBO-hosted geometry the model rotates smooth as silk (unsurprisingly), but using streamed geometry it's a slideshow (and not a particularly fast slideshow, either).

    Forcing us to use streamed geometry with multi-transparency surfacing cases is a significant issue that needs to be fixed. There's no intrinsic OpenGL reason VBOs can't deal with surfacing like that any different from streamed geometry, the shaders don't really care how the geometry data gets into VRAM. The current limitation just makes working in LW even less pleasant for folks working with large/complex geometry data sets.

    https://vimeo.com/279596795/b0ac2467f2
    Last edited by jwiede; 07-11-2018 at 10:54 PM.
    John W.
    LW2015.3UB/2019.1.4 on MacPro(12C/24T/10.13.6),32GB RAM, NV 980ti

  2. #17
    Registered User
    Join Date
    Aug 2006
    Location
    UK
    Posts
    200
    Quote Originally Posted by jwiede View Post
    That answer doesn't make sense. Whether the geometry is transferred within VBOs (via DMA), or streamed via sending primitives down the pipe (typically PIO), is orthogonal to how the surfaces (shaders) applied to the geometry sets are handled.

    It also fails to acknowledge that VBO-hosted geometry offers major performance benefits over streaming geometry in modern graphics hardware. Being forced to use streaming geometry transfers for cases when multiple transparency states are present has a significant impact when dealing with large, complex geometry data sets.

    I've attached a (private) link to a video of Layout 2018.0.5 manipulating a 3M-poly object using both VBO and streamed geometry modes, and the difference is fairly stark. Though the video may not show it clearly, using VBO-hosted geometry the model rotates smooth as silk (unsurprisingly), but using streamed geometry it's a slideshow (and not a particularly fast slideshow, either).

    Forcing us to use streamed geometry with multi-transparency surfacing cases is a significant issue that needs to be fixed. There's no intrinsic OpenGL reason VBOs can't deal with surfacing like that any different from streamed geometry, the shaders don't really care how the geometry data gets into VRAM. The current limitation just makes working in LW even less pleasant for folks working with large/complex geometry data sets.

    https://vimeo.com/279596795/b0ac2467f2
    Yes I have to agree. I noticed a considerable impact on viewport performance for streaming. VBO is faster and smoother. Also the explenation of this only happening on a poly heavy scene is not true.
    In the example given above of a simple subpatched sphere this still occurs.
    This is a major bug that needs fixing. Switching to streaming maybe a temporary fix for now but not a satisfactory one long term.
    All in all at the moment this new Lightwave is proving to be a slower version than before. Takes longer to set up shaders, takes longer to render and vpr takes longer.
    The model I am using at present renders in about 3 seconds in vpr in LW 11.6.3 but takes around a minute in LW2018. I'm also having to fiddle with node shaders a lot to get rid of unwanted speckles in LW2018.
    I know I have to get used to this new lighting and shading system but at present I'm not seeing much benefit. I finding I'm jumping back to 11.6.3 for getting things done quicker at the moment which is a bit dissapointing.
    Maybe as I learn more about LW2018 this will change, hopefully.

  3. #18
    Registered User
    Join Date
    Jan 2005
    Location
    Colorado Springs
    Posts
    1,795
    Yes, this is a regression and should be addressed. I re-created the reproducer object in LW11.6.3 and Layout does NOT demonstrate the flaw. Load the scene in LW2015, the flaw is not present.

    I added the information to the Case and requested that it be re-opened as a regression, indicating that VBO acceleration is essential in complex models so the regression should be addressed.

    mTp

    P.S. the case has been re-opened and will be "assigned to the proper developer for assessment".
    Last edited by MonroePoteet; 07-12-2018 at 08:20 AM.

  4. #19
    Registered User
    Join Date
    Aug 2006
    Location
    UK
    Posts
    200
    Quote Originally Posted by MonroePoteet View Post
    Yes, this is a regression and should be addressed. I re-created the reproducer object in LW11.6.3 and Layout does NOT demonstrate the flaw. Load the scene in LW2015, the flaw is not present.

    I added the information to the Case and requested that it be re-opened as a regression, indicating that VBO acceleration is essential in complex models so the regression should be addressed.

    mTp

    P.S. the case has been re-opened and will be "assigned to the proper developer for assessment".
    Glad to hear this is being looked at again. VBO is essential in viewport as you say and have to agree this is a regression.
    Lightwave is always advertised as a product for getting things done quickly and this is only adding to the thought that it is getting slower in its production.

  5. #20
    Electron wrangler jwiede's Avatar
    Join Date
    Aug 2007
    Location
    San Jose, CA
    Posts
    6,507
    Quote Originally Posted by MonroePoteet View Post
    P.S. the case has been re-opened and will be "assigned to the proper developer for assessment".
    I just hope that hasn't become the Newtek equivalent of "Real Soon Now".

    On the plus side, it shouldn't be that difficult to reconfigure the code that assembles the viewport OGL to handle either VBO-hosted or streamed geometry equally well. I'd be more than happy to help them reconfigure the code, as needed.

    It'd be even better if we could get back the ability to use CgFX shaders again as well (perhaps as custom OpenGL Material?).
    Last edited by jwiede; 07-12-2018 at 01:08 PM.
    John W.
    LW2015.3UB/2019.1.4 on MacPro(12C/24T/10.13.6),32GB RAM, NV 980ti

  6. #21
    Registered User
    Join Date
    Aug 2006
    Location
    UK
    Posts
    200
    Has there been anymore word on resolving this by Newtek?

Page 2 of 2 FirstFirst 12

Bookmarks

Posting Permissions

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