Metashape Python-API slowing down during DEM generation step from depth maps

[geot]

Our script is slowing down heavily and seems to eventually stop during the following step:

chunk.buildDem(source_data=Metashape.DataSource.DepthMapsData,  progress=progress_print)

The same process used to take a few hours max, but now seems to take several days (plus continuously slowing down) before I kill it.

Metashape.version: 2.1.3.18946

GPU configuration is attached. Output of the above function is attached as well.

Thanks for any pointers that help resolve this issue. A new CUDA driver may have been installed since I processed earlier flights for the same perimeter successfully (and much faster), but I'm not sure the driver causes the issue. 

[Alexey Pasumansky]

Hello geot,

Do you observer same problem when running depth maps based mesh generation via GUI?

Please also check that chunk is properly georeferenced before DEM generation is started.

As for the GPU Preferences tab, it is recommended to disable integrated Intel graphics and to uncheck "use CPU" option, leaving only discrete GPU enabled.

[geot]

Yes, it did, see screenshot attached.

Notice that the Metashape version of the GUI is 2.2.0.19890

Is it possible to adapt GPU preferences in the script, too?

[geot]

Information regarding driver attached: 32.0.15.8115

[Alexey Pasumansky]

Hello geot,

Maybe you can share the log here or via email to support@agisoft.com together with the PDF report exported from the related project?

Maybe Quadro RTX 4000 is just not fast enough for large datasetes. If I am reading specification sheets properly, its estimated processing power for single precision computations in TFLOPs is several times lower compared to GeForce RTX 4070 Ti or RTX 4080, for example.

[geot]

Installed driver/runtime CUDA: 11070/10010 and things are working as they used to.

Using device: Quadro RTX 4000, 36 compute units, free memory: 7149/8191 MB, compute capability 7.5
  driver/runtime CUDA: 11070/10010