Performance Optimization

Adjusting SURE and hardware conditions to optimal performance and throughput

Step-by-step guide

In order to achieve an optimal throughput, please consider the following suggestions:

  1. Optimize hardware
  2. Choose processing and source imagery directory
  3. Optimize parameters
  4. Distributed processing

Optimize Hardware

Optimize Hardware

  • Meet Hardware Requirements - in particular sufficient main memory (RAM)
    small images (<40MP) >=16GB RAM, larger images  >=32GB RAM - more is beneficial
  • Decide whether your graphics card (GPU) or main processor (CPU) is faster through test
    in user interface Preferences - use GPU; or commandline -gpu; consider CUDA Driver Timeout
  • Enable fast hard disk access
    if possible use a SSD or a RAID storage for large projects
  • Avoid network access if possible to avoid connection loss and to benefit from caching
  • Avoid external harddrives since caching is not possible (process continue while file writing)
  • Avoid other applications running in parallel consuming RAM or CPU/GPU power 
  • Evaluate whether Antivirus file-checks are slowing the process down and turn off if needed

Choose Processing Directory

Processing Directory

  • Assign processing directory (basepath) to fast disk storage (SSD or RAID if available)
  • Store source images on fast storage - in particular if given in raw or tif format
  • Avoid network access if possible to avoid connection loss and to benefit from caching
  • Avoid external harddrives since caching is not possible (process continue while file writing)

Optimize Parameters

Parameter Optimization

  • Use predefined Scenarios - particularly when processing large frame airborne imagery
  • Choose processing resolution / quality according to your needs
    for default "High" recommended (close range, UAV, Oblique), for large frame Nadir flights "Ultra" 
  • Define Region of Interest / Area to save processing time
  • Increase tile size if more than the minimum amount of memory is available (see table on the right)

  • Use LAS (default) point cloud format for speed, LAZ (optional) to compress (speed loss)

  • Deactivate LOD (Level of Detail) mesh generation or texturing if not needed to save time
  • For DSM / 2.5D Workflow and if RAID system is available:
    • increase thread count - testing necessary, since only powerful hardware can handle it
    • reduce unneeded output in GUI or with -dsm_output_minimal / -true_ortho_output_minimal

Distributed Processing

Distributed Processing

  • If more than one PC/Server and license are available consider distributed processing
  • Blocks can be subdivided using the Region of Interest / Area (commandline: -area)
  • Consider using overlap tile if needed
  • Approach the nFrames sales team for distributed processing framework options

Retrieve support

If you would further like to improve performance, please send a log file of your process, your system setup and the amount of images+resolution being processed to our support team.



























RAM

2.5D Process

Tilewidth (Pixels)

3D Process

Tilewidth (Pixels)

16GB1000500
32GB2000 1000
64GB4000 2000
128GB50002000

Suggested tile size (GSD * pixels)