%env CUDA_PATH=/usr/local/cuda
 
May set the CUDA verison to an incorrect version in some cases.
Plus this is non-portable if running on Windows.
Only set this, if it is not already set. Ideally set it relative to a CUDA version which is on the PATH already.
Alterantively we could modify FLAMEGPU2 to try finding nvcc on the path and getting the CUDA directory that way instead.