Slow pointcloud generation on nVidia Tesla K80 with Metashape v.1.5.0

[adam_s]

Thank you for a quick response.

Yes, "enable CPU" flag was disabled and "fine-level task subdivision" option was enabled for both runs.

Project was present on the local disk, but the images were on the network disk. At this moment, i'm copying files to the local disk and i will perform some additional tests.

Best regards,
Adam

[Alexey Pasumansky]

Hello Adam,

The longer processing time could be related to the saturated network connection, when both cards are receiving the image data via network, so the results of local processing would be useful.

[adam_s]

Dear Alexey,

I've performed additional tests after i've copied image files to the local disk.

I've achieved following results:

2x GPU   - depth maps generation ~~ 14h , point cloud generation ~~ 5h

1x GPU (first GPU enabled only) - depth maps generation ~~ 14h , point cloud generation ~~ 7h

1x GPU (second GPU enabled only) - depth maps generation ~~ 13h , point cloud generation ~~ 6h. Thanks to that, i wanted to rule out the impact of GPU  failure.

As you can see, I expected some other results in case of both GPU enabled any advice?

Best regards,
Adam

[Alexey Pasumansky]

Hello Adam,

Do you have logs saved from these runs? If they are available, please send them to support@agisoft.com.