Trimble Business Center

We are trying to train a custom classification model in Trimble Business Center. The GPU is NVIDIA Quadro RTX 3000 with driver 32.0.15.9659. We also installed CUDA Toolkit, but the error remains. TBC loads the dataset and builds the model, but crashes immediately at training start in C:\ProgramData\Trimble\PythonDL\training-env\... with: RuntimeError: CUDA error: no kernel image is available for execution on the device. It appears that the LinNet/Pointcept CUDA extension may not be compiled for this GPU architecture. 

What could be the problem?

Can you confirm that you are NOT using a AMD processor (CPU)? Also, do you have at least 64 gigs of ram and are you ONLY using a SSD internal drive/drives? I would also make sure you have at least 100 gigs free on whatever drive you are using. 

Hi Robin,

We have had a couple of cases with people with this NVIDIA Graphics card with the same problem (not being able to use a Custom model for classification). The 6GB VRAM of the GPU is below the recommended 8GB. The Quadro GPU is quite old now too, so there may be an incompatibility of CUDA versions (e.g. we might be needing a version later than what is compatible with the Quadro series it's ~7 years old now). 

What happens when you complete the Analysis step? What values are returned for Epochs/Batches/Samples?

Reducing these down may help. 

------------------------------
Brendon Edge
Geospatial & BCFS Director of Support
------------------------------