I have followed the instructions from people on this post and have tried every archived version of nvidia drivers available from 441.28 - 436.02
I made sure that my cache was set to a drive with plenty of space
I have restarted the process and created a new scene several times to see if it was a bad file
in settings I have messed with vertex distance, decimation threshold and force single part mode
I have changed the reconstruction area anywhere from the whole model to just bits and pieces
Things to note about the setup
Reconstruction Settings
have set both GPUs to use and every combination inbetween
Normal Model Image downscale: 5
Mesh calculations: anywhere between 0 - 10
I have left macimal vertex count per part at: 5,000,000
Detail decimation factor: anywhere between 0 - 100
I have not changed or messed with anything in the Advanced settings.
I hope this give you enough information about the issue and what I’ve done to try and make a reconstruction happen. If there is anyone with more information on what I should try next or has a setting I should change I would love to hear from you!
I have tried with one card and that did not change the result. Also want to be clear, do you mean take the graphics card out of the machine or make sure the settings are set to only use one? I have changed the settings and got that same result.
Got the same problem today after updating to a version fixing an issue with 2D view of laserscan pointclouds. No difference using just one or both GPUs.
Exactly there is no difference if you will use one or two GPU units and support only advice o use one graphic card. Besides all my scans was done with high detail on two GPU units until installation of new version of reality capture so definitely it’s a software issue in new version
Physically remove all GPU cards, except one - single card configuration.
or
Install old GPU driver. The version 419.67 is working for us.
Reason:
The issue is with the CUDA memory de-allocation function, that has stopped working properly with latest NVIDIA GPU drivers.
More specifically the function CUDAFreeHost() resulted with success code, but the memory was not de-allocated and therefore after some time, the GPU pinned memory was filled up and the SW ended up with the message “CUDA error : 2 : Out of memory”.
It seems that this bug causes problems also to another CUDA developers:
With the latest update, you should be able to run with the newest drivers. If you anyways experience any of such issues try rolling back to the 419.67s, I do believe you will not have to though.
Hi, I have this CUDA error popping up once in a while without any clear logic - is sometimes appears, sometimes not -no matter if the project has 40 or 900 images - but when it does, it is usually during the texturing process and it keeps reappearing repeatedly even after canceling the error window many times - so as the result the texturing process could not be finished…
I have tried the newest Nvidia drivers, and the older one recommended above (419.67) but it still keeps appearing, can you help me solve this, please???
PS I do not have SLI - just one older GeForce GTX 750 card…
Dear user,
this error is not RC related and aims directly towards Nvidia as your GPU, it probably has some issues with memory leaking or use. Sorry to see that, but I don’t think I can help you here.
Sorry for posting in this old thread, I just came across the same memory problem and didn’t want to open a new one just to post what helped me fix it:
I’ve got a GTX 1050Ti and a GT 710 that I only use for extra monitors. I’ve reconstructed parts of the same mesh multiple times before but when I started the full reconstruction in high detail I got the same “low device … memory” error message. It looks like RC ran out of memory because it was using the 710 instead of the 1050 (which should be the main one).
You can choose the GPU RC uses in:
Reconstruction - Process / Settings - Image depth map calculation / GPUs to use
I set the 710 to “False” and it looks like it’s working fine now.