GPU setup assistance

ODM
#1

I have 2 nodes that have different issues with the same data but both are not running with the GPU

I have WebODM installed on a computer which has a Ryzen 5900X and a Radeon 6900XT. When running the dataset on this node/machine it gives me this:

[INFO] Running opensfm stage
[INFO] Maximum photo dimensions: 4000px
[INFO] Photo dimensions for feature extraction: 4000px
[INFO] No CUDA drivers detected, using CPU

The main docs advise GPU support is Open-CL based so not how to tell ODM to look for other processors besides cuda.

On the second node, (same WebODM from first machine but the second node is running nodeodm:gpu on docker on a second machine) it has a Threadripper 3960X with a GeForce 1080TI. When running the same dataset it gives me this:

[INFO] Running opensfm stage
[INFO] Maximum photo dimensions: 4000px
[INFO] Photo dimensions for feature extraction: 4000px
[WARNING] Image size (4000x3000px) would not fit in GPU memory, falling back to CPU

After checking the memory of the GPU, it is empty. I think this model has 11GB of memory so I was surprised a single photo wouldnt fit.

Is there anything I can do to improve this? I’m running large datasets and any improvement on time done would be awesome!

1 Like
#2

Can you link to where it says this? That is an error I’ll need to correct post-haste.

GPU processing is in fact based upon CUDA, so will require NVIDIA hardware and drivers compatible with CUDA Runtime 11.4 and above.

We’re still working on CUDA and memory usage. How many 12MP images do you have in this dataset?

#3

“The SIFT GPU implementation is OpenCL-based, so should work with most graphics card (not just NVIDIA).”

900+

Is ODM trying to load the entire dataset into memory? I guess that would go over the memory limit then.

2 Likes
#4

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