Forum

Author Topic: Depth map generation performance  (Read 893 times)

andysfd

  • Newbie
  • *
  • Posts: 13
    • View Profile
Depth map generation performance
« on: December 15, 2020, 09:51:29 AM »
Hello,

i'm struggling with the depth map generation performance (v 1.6.5),
currently processing a 10k x 20mp corridor mission, splitted into several chunk.
The cpu workload is around 25% I7 8700K and the GPU Quadro RTX 4000 is around 5%, both with peaks up to higher percentages but only 1/10 of processing time.

why is the depth map generation so slow? i feel like it was faster in previous versions. (very high, mild filtering, no tweaks)

Alexey Pasumansky

  • Agisoft Technical Support
  • Hero Member
  • *****
  • Posts: 13729
    • View Profile
Re: Depth map generation performance
« Reply #1 on: December 15, 2020, 01:10:19 PM »
Hello andysfd,

Please make sure that your Quadro card is enabled in the GPU Preferences tab, I also recommend to uncheck "use CPU" option in the same dialog.

If you are monitoring the GPU activity via Windows 10 Task Manager, switch one of the graphs on GPU page to "Cuda" option to check the GPU load. Or use dedicated utilities, such as GPU-Z or NIVIDA-SMI.
Also check that in the processing log there are no errors related to GPU-processing and fallbacks to CPU.
Best regards,
Alexey Pasumansky,
Agisoft LLC

andysfd

  • Newbie
  • *
  • Posts: 13
    • View Profile
Re: Depth map generation performance
« Reply #2 on: December 15, 2020, 04:57:56 PM »
Thank you,

monitoring the CUDA performance i see 100% workload.