Optimizing CUDA Memory Allocations Using NVIDIA Nsight Systems