Advancing the distributed Multi-GPU ChASE library through algorithm optimization and NCCL library