Announcement

Collapse
No announcement yet.

[GPU] VRAM availability

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • [GPU] VRAM availability

    Hello,

    I did some test with V-Ray GPU and watching the Event Log, I noticed that V-Ray can't access to all the - theoretically - VRAM available, 4Gb in this case.

    System -------- Time -------------------- Type -- Message -------------------------------------
    Log System Thu Sep 29 23:53:20 2016 Info Initializing CUDA renderer (single kernel version)...
    Log System Thu Sep 29 23:53:20 2016 Info Querying for CUDA devices...
    Log System Thu Sep 29 23:53:20 2016 Warning Devices query yielded no results; using all available devices!
    Log System Thu Sep 29 23:53:20 2016 Info Device GeForce GTX 980 has compute capability 5.2
    Log System Thu Sep 29 23:53:21 2016 Info Setting CUDA stack size to 14000
    Log System Thu Sep 29 23:53:21 2016 Info Building optimized CUDA kernel (0) for comp cap 5.2 for device 0...
    Log System Thu Sep 29 23:53:21 2016 Info PTX file generated with CUDA Toolkit v7.5 for CUDA compute capability 2.0
    Log System Thu Sep 29 23:53:21 2016 Info Optimized CUDA kernel assembled successfully
    Log System Thu Sep 29 23:53:21 2016 Info Total memory for device 0 : 4096 MB, free 2807 MB
    So something is eating around 1GB of memory. I supposed it's ok since the OS need some VRAM to draw GUI etc. but 1,2 GB ? It's tough, mostly with only 4Gb of VRAM.

    Is it something known ? Is there any workaround / solution ?


    One last thing, I can't render this scene since it doesn't fit in the GPU. But I'm a little bit surprised since there is only 206.451 faces visible, 5 materials enabled and the textures are resized to 512x512 maps.

    Fri Sep 30 00:00:29 2016 Info Mtls 434.30MB, Geo 2079.76MB, Lights 0.04MB, LightCache 8.07MB, Other 243.83MB (data index 4707)
    Fri Sep 30 00:00:29 2016 Info Building optimized CUDA kernel (0) for comp cap 5.2 for device 0...
    Fri Sep 30 00:00:29 2016 Info PTX file generated with CUDA Toolkit v7.5 for CUDA compute capability 2.0
    Fri Sep 30 00:00:30 2016 Info Optimized CUDA kernel assembled successfully
    Fri Sep 30 00:00:30 2016 Info Total memory for device 0 : 4096 MB, free 60 MB
    Fri Sep 30 00:00:30 2016 Failed ./src/ocl_tracedevice.cpp(3012) : CUDA error 2 : unable to allocate enough memory to perform the requested operation (out-of-mem)
    Is there something wrong or I shouldn't be surprised ?

    Thank you

  • #2
    Do you think MODO is holding on to an image or something from the scene? Preview uses the Vram buffer too right, though I doubt you had that open.

    Comment


    • #3
      I don't think, but I can be wrong.

      I have just tested an other very simple scene ( the glass from the other thread ), just after booting up my computer and with only one other program opened ( internet browser ), and the same number appear : free, 2847 MB

      Comment


      • #4
        Would be interesting to see what happens if you can get the GPU to render in standalone. That might help narrow it down.

        What's the texture resolution of that checker background you used and is it an EXR? I take it the scene doesn't start up with AVP (advance view port) or you don't have other scenes in the BG loaded running AVP right?

        Comment

        Working...
        X