Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[question]Double graphics card CUDA... useful or not #644

Closed
lord666 opened this issue Sep 18, 2019 · 8 comments
Closed

[question]Double graphics card CUDA... useful or not #644

lord666 opened this issue Sep 18, 2019 · 8 comments

Comments

@lord666
Copy link

lord666 commented Sep 18, 2019

Use CUDA graphics card, okay ... but, own 2 CUDA graphics cards (ex GTX 1070 and old GT730) does it speed up rebuilding in Meshroom or is it the same result?

@natowi
Copy link
Member

natowi commented Sep 19, 2019

Multi GPUs are supported, but DepthMap is still not fully optimized.
Support for multiple GPUs is useful, if you want to utilize older graphics cards like two GT730.
Your GTX 1070 alone is good enough. Adding the older GPU will not speed up rebuilding.
#250
#244

@natowi natowi closed this as completed Sep 19, 2019
@lord666
Copy link
Author

lord666 commented Sep 20, 2019

Thank you for your reply. I was hesitating because, precisely, my GPU didn't seem to me fully used ... but, the RAM ARGGGGG, I think invest in 16GB extra ... 32GB is the limit for a basic projects

@fabiencastan
Copy link
Member

If you have 2 graphic cards, both will be used but the work is split equally between them, so if you have 2 very different configurations it could also decrease the overall performance...

Regarding RAM, you can reduce the number of points without major quality impact:
https://github.com/alicevision/meshroom/wiki/Crashed-at-Meshing

@lord666
Copy link
Author

lord666 commented Sep 20, 2019

I wanted to test this option by reducing the value between 2000000 and 3000000 and indeed it greatly simplifies the calculations without there being big visual differences.
I will test add photos to the project to increase the reconstruction.
Is it possible to "lock" the 1st "structureFromMotion", so that when adding a new one, doesn't touch it. I thought that it was already the case but comparing the "StructureFromMotion" there are small differences

@fabiencastan
Copy link
Member

Is it possible to "lock" the 1st "structureFromMotion", so that when adding a new one, doesn't touch it.

Yes, there is an option on the 2nd SfM to "Lock Scene Previously Reconstructed" to avoid any modification to the previously reconstructed cameras of the input SfM.
The best option IMO is to allow to make modifications to the previous scene and then realign with the previous scene using SfMAlignment.

@lord666
Copy link
Author

lord666 commented Sep 22, 2019

oh..Thanks.. indeed, I had seen this option but I wasn't sure if that was it.
I have also tested "SFMalignment" which works very well, but the meshroom interface makes it..hmm not very user-friendly for multi alignment. Mushroom is a great genius software with a very responsive community, but his nodal interface makes it difficult to access for a novice user, I'm not new in photogrammetry, but here, I'm lost like a young chick looking lol... I dream of a "nodal graph interface" like "Substance Designer" not necessarily with a lot of node, but, better management of links and especially a possibility to save node and graph presets...

@fabiencastan
Copy link
Member

That's on its way, see some preliminary steps in this direction:
#632 with --pipeline option (so you can save a scene and use it as the default graph) and some ui improvements to the graph editor #612.

@lord666
Copy link
Author

lord666 commented Sep 22, 2019

Oh oh!! it's christmas before the hour lol
I am really excited about the idea to test it, But, before that, I must continue to understand already existing option... and there are many héhé

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

No branches or pull requests

3 participants