On GPU crash error

When trying to reconstruct an asset I get the following error:

and this message on the console (sorry for the low res)

A bit of background, there are 11,639 images drone images with 1,349 images being 48 MP stills and 10,290 images being 8 MP frames extracted from videos. The alignment works well but it fails on the normal reconstruction. I am running a machine that has 2 NVIDIA 4090s, but based on the task manager tracing only one of them seems to be getting used. I’d like to understand the source of the issue a little better and if there are solution suggestions that would be great!

Thank you,
DB

Hello Derek,
this could be a HW issue, as it is CUDA error.
Also, it is strange, that there is a dollar sign after the image name. Have you changed the images somehow or moved them? Is this also happening after application reset? Which version are you using? Do you have enabled both GPUs in the Advanced Reconstruction settings? Are you using default Normal detail reconstruction settings?

Hi Ondrej,

Thank you for the timely reply. I’ll try and itemize my points efficiently here.

Q: Could this be a hardware issue?
A: Of course it’s possible, but I’ve been using RC on this machine for many months and have used it successfully since so I think it’s unlikely.

Q: Also, it is strange, that there is a dollar sign after the image name. Have you changed the images somehow or moved them?
A: Agreed that the dollar sign is unusual, but when I look in the Images directory it isn’t there. I thought that was part of RCs internal workflow. I’ve been very careful not to modify the images or their locations during the construction.

Q: Is this also happening after application reset?
A: I haven’t tried on the same set. That will be something I should try though and report if it works.

Q: Which version are you using?
A: 1.4.0.117414RC

Q: Do you have enabled both GPUs in the Advanced Reconstruction settings?
A: Yes.

Q: Are you using default Normal detail reconstruction settings?
A: Yes.

I will try again on the set that had issues and report back, but given the size it takes a long time and I likely won’t be able to try until next week. In the meantime, are there any other suggestions you have?

Best,
DB