Slow reconstruction

Collapse
X
 
  • Time
  • Show
Clear All
new posts
  • MPP33
    Blossoming 3Dflower
    • Oct 2026
    • 2

    #1

    Slow reconstruction

    I`m using Zephyr Lite (updated tp 9.013) now for a while.
    As one of my first projects I made a 3D model of our house for garden planning which worked great (Pictures from Canon EOS550D, 698 pictures, ~5GB)
    When I remember right it took roughly a bit under one day (mostly over night) to calculate the model.

    After a little break with switch to Win11 on my laptop I started again doing some tryouts and recalculating some models where I had the feeling that it is way slower than before.

    Hardware:
    Fujitsu Celsuis H760 mobile CAD workstation
    32GB RAM
    NVIDIA Quadro M2000M (4GB)
    Intel Core i7-6820HQ CPU @2,7GHz

    Now I have a new project, again the same house but this time I have used a drone.
    6,3GB, 312 pictures so I expected to be completed over night.

    It feels like it`s an endless process and after roughly 24h I stopped because it seems that it is not using the GPU at some stage.
    I tried to get the latest drivers from NVIDIA and install it but while using the GPU at some earlier steps there is again the step "Computing photoconsistency" where it seems to use only the CPU

    I have uploaded the current log file where I find one point where it seems that it has an issue using the GPU and then it switches to CPU only:
    Task manager tells me 100% CPU, 0% GPU (at least it shows that Zephyr is using the NVIDIA GPU)

    [14:03:10] Level 1/1 - Block 1/1 - [OpenCL 0] +++ ERROR +++ <kernel>:1:8971: error: passing '__generic r64 *' (aka '__generic double *') to parameter of type 'r64 *' (aka 'double *') changes address space of pointer
    [14:03:10] Level 1/1 - Block 1/1 - +++ ERROR +++ (cl build program failure [launch_kernel '_openclBarycentricMapCalculator2DKernel'])
    [14:03:10] Level 1/1 - Block 1/1 - [OpenCL 0] (compute) enable reserved CPU device
    [14:03:10] Level 1/1 - Block 1/1 - [OpenCL 0] (compute) run same jobs on CPU

    What is the issue here? Should it use the GPU? And if yes - what can I do that it will use it?


    As I`m currently in the process of trying if I can use Zephyr also for work so I tried to run the same task on my company workstation (Zephyr full version 30 day test license)
    There I also recognized that a smaller task took longer than I initially expected according to earlier projects but now I wanted to have the direct comparison.

    Hardware:
    Dell Precision 3591
    32GB RAM
    NVIDIA RTX 2000 Ada
    Intel Core Ultra 7 165H 1,4GHz

    It has currently reached the same step, it shows exactly the same error and then it starts running the CPU on 100%
    So I would not expect that this more modern DELL workstation will be faster than my 10year old private workstation.
    I would also rule out that this is simply an issue because my old workstation isn`t made for Win11 (everything else runs fine on it).

    What can I do to get the process faster?
    Attached Files
    Last edited by MPP33; Yesterday, 05:36 PM.
  • cam3d
    3Dflover
    • Sep 2017
    • 705

    #2
    Hey MPP33 -

    Here are my thoughts:

    - A 10 year old 4GB VRAM GPU is considered very weak, and for serious 3D production work, close to unusable. The RTX2000 is 16GB VRAM and much more powerful and suitable for this kind of work.

    - 32GB RAM is at the low end of the spec scale with oversaturated RAM you will end up with a SWAP slowdown - Best to upgrade to 64GB if possible to give yourself some processing headroom.

    - As to this specific error, I'm not sure, but it could be a CUDA compatibility issue with your old GPU. The provided log seems to be truncated. Can you please provide the full session logs so we can do a proper analysis? Thanks!

    If you want faster processing there are two primary options: Lower processing settings and/or better hardware - You will also get better at image budgeting, there is a sweet spot where you can take just enough photos for a good reconstruction, and excessive overshooting will slow down processing for very little gain.



    Comment

    Working...