search

NVlabs / instant-ngp

Instant neural graphics primitives: lightning fast NeRF and more
https://nvlabs.github.io/instant-ngp
Other
15.57k stars 1.89k forks source link

Error running with higher level, features per level (F, L) #1538

Open keythinking opened 2 months ago

keythinking commented 2 months ago

image

Hi, I'm training LEGO dataset using instant-ngp, and trying to reproduce the result from your paper.

But train doesn't work and get error message from cuda_graph,h , when I set parameters as F=8 T=2^17 L=16 F=4 T=2^17 L=32 which are F T L = 2^24.

I'm using RTX 3090 which is same with paper used. And It works well and can get same result as paper when I decrease F or L.

Is there something to fix to run training with higher F or L?

Thanks in advance.

--------------------------------------------------ERROR MESSAGE---------------------------------------------------

terminate called after throwing an instance of 'std::runtime_error' what(): /????/????/instant-ngp/dependencies/tiny-cuda-nn/include/tiny-cuda-nn/cuda_graph.h:139 cudaGraphExecUpdate(m_graph_instance, m_graph, &error_node, &update_result) failed: too many resources requested for launch Aborted (core dumped)