Page 1 of 3 123 LastLast
Results 1 to 10 of 25

Thread: Unified sampling and distributed rendering

  1. #1
    Join Date
    Mar 2008
    Posts
    28

    Default Unified sampling and distributed rendering

    Is it possible to do a distributed render using unified sampling. My tests through Qube! to machines running mental ray standalone 3.9.1.39 haven't been successful.

    The error is RC 4.0 fatal 081501: interrupted by signal 11 or DB 2.7 fatal 041501: interrupted by signal 11 if that helps.

    I can get more specific on the setup but if it's just not possible to do DBR and unified then I'll shelve it for now.

    Thanks,
    Dave

  2. #2
    Join Date
    Oct 2007
    Location
    New York
    Posts
    366

    Default

    My understanding is it is currently not possible to do this with Unified sampling yet.

    Looking forward to when this feature becomes possible.

    Best,

    Richard

  3. #3
    Join Date
    Dec 2004
    Location
    Marina Del Rey, California
    Posts
    2,916

    Default

    Unified + progressive is a challenge, but are you using unified without progressive?

    Also, just to be clear, you mean distributed network rendering, sharing a single frame render across multiple machines on a network, right? Some call it DBR (distributed bucket rendering), where each tile (bucket) can be rendered on a given core of a given machine sharing the render.
    Barton Gawboy
    Training and Special Projects, NVIDIA ARC
    LAmrUG Forum Originator

  4. #4
    Join Date
    Mar 2008
    Posts
    28

    Default

    Thanks for the replies.

    I was using only Unified sampling which failed. After turning on progressive it did correctly distribute then render so that looks to be the way to go.

    Bart your correct in that I'm sharing a single frame across multiple machines. When you said 'Unified + progressive is a challenge' are you aware of any major pitfalls of this setup?
    I'm primarily looking at this workflow to enable me to add DOF to large print still images.

    Thanks,
    Dave

  5. #5
    Join Date
    Jan 2007
    Location
    Berlin, Germany
    Posts
    90

    Default

    Dave,

    Progressive is not designed to work with DBR. Slave/satellite machines do connect to the master, but do not participate in the final image rendering. If progressive is used in combination with finalgather/photons/etc, when slaves/satellites participate in the map precomputation stages.

    Unified sampling WAS supposed to work with DBR. Unfortunately it slipped out network regression testsuite and got broken. This is fixed now for mental ray 3.11 and 3.10.3.x. I am looking forward to the fix still being integrated to mental ray 3.10 OEM branches (Maya 2013).

    Thanks a lot for reporting the issue!

    Juri

  6. #6
    Join Date
    Oct 2007
    Location
    New York
    Posts
    366

    Default

    hmm interesting. seems like someone has unified sampling working with DBR http://forums.cgsociety.org/showpost...3&postcount=43

    I was under the assumption that it never worked, think Juri or Steve might have stated that in the forums so I never tried it.

    I will do a test tomorrow now.

    Best,

    Richard

  7. #7
    Join Date
    Mar 2008
    Posts
    28

    Default

    Thanks for the info.

    I'll keep my fingers crossed that the unified sampling fix makes it into Maya 2013.

    With regards to progressive not being designed for distributed rendering, I can see that from a 'distributed bucket rendering' point but it's a shame there isn't any method to distribute it.

    I was hoping that progressive rendering would give me the option to render 'in camera' depth of field, on very large print stills, with manageable render times.

    Anyway, thanks again for the info.

    Dave

  8. #8
    Join Date
    Jan 2011
    Posts
    10

    Default

    hi guys, I'm "m0z" from cgtalk...

    interesting topic, I'm gonna make some more tests with DBR & IBL & Unified Sampling but in my first test it worked.

    Although I have to mention that usually I don't use DBR because the render starts much slower than with Maya Batch (but this could be a problem with our network and render manager config).
    Usually I send it to the farm via RoyalRender and tell RR to "tile" the frame like 8times, so each of the 8 fast machines get one tile (vertical tile) to render with Maya 2012 batch.
    Then I have 8 own exr files, RR merges them to one but the passes aren't supported, so I just merge them together in Nuke.


    I'll update this post later... cya

  9. #9
    Join Date
    Oct 2007
    Location
    New York
    Posts
    366

    Default

    Hi m0z.

    Ah ok so you are not actually using "DBR" but using the tile render option within Royal Render?

    Cheers,

    Rich

  10. #10
    Join Date
    Jan 2011
    Posts
    10

    Default

    for the test image @cgtalk I used DBR! At the moment I'm rendering a complex scene with US, IBL and DBR.
    All blades are assigned but it seems that only the master blade uses all cores. But this could be a problem with our configuration, we have troubles at the moment and are replacing our servers soon because of this... so I don't know if it would work otherwise.

    But I don't get any errors whatsoever, it renders. At least on machine 1

    But usually I use tile rendering, because I have visual feedback much faster. Only problem is that with wrong FG settings you have seams. But there are workarounds (like using a map).

Posting Permissions

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