nvcc version remains the same after updating the CUDA Toolkit version...

CudaMalloc not allocating to shared GPU memory on Ampere micro-architectures in opposite to Turing...