Less 'GPU not used' issues with v 2.8.1?

Comparing releases 2.8.1 and 2.7.1 I see mentions about the GPU issue on Github…

I have already read threads on this forum about this problem, I just want to know if the situation has changed with the latest version, since I get that all the time with my (older) laptop with 2.7.1 (Warning: image size would not fit in GPU memory, falling back to CPU)

Using a lower feature-quality (0.25x) the GPU is used, but the processing fails (while it doesn’t fail with the same set of photos running on the CPU only:

[ERROR] Uh oh! Processing stopped because of strange values in the reconstruction. This is often a sign that the input data has some issues or the software cannot deal with it…

I’m curious to know if those who had this issue notice a difference with ODM 2.8.1?

Running ODM console on Win 10 Pro, but I don’t think it makes a difference here?

1 Like

Using Windows native version here, which has ODM release 2.8.0, and the errors when using GPU are still occurring. In my case it seems to be the race condition problem, where the normal order of events gets out of sync- features files are not produced in time for pair matching.

This means for many datasets I have to use full size images (M2P) and ultra to force CPU feature extraction, in order to not fail.


Thanks, I wouldn’t have thought of that, it’s an interesting point.This may explain why I have more fails on my gaming laptop with GPU than on my tablet… although from what I see in the console (I have no idea of the internal processes/files in ODM) the GPU is not used with my usual settings.

Will ‘ultra’ feature-quality slow the process a lot, compared to ‘high’ ? I’m already setting time consuming parameters, low gsd, high features number, etc…

1 Like

Yes, it does take significantly longer, in the order of 4X over high I think.

1 Like

I don’t know that we’ve made any GPU-processing specific improvements in v2.8.2 over v2.8.0, but there’s always the hope that other fixes improved things in this arena.

If you get around to testing, any feedback is always welcome.

1 Like

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.