Increasing complexity is one of the biggest challenges facing industry today. Large scale computations required by industrial simulations need ever more effective strategies to be adopted. Parallel computing may play a significant role in this field, especially relying on Graphics Processing Units (GPUs) based architectures; however, algorithms optimization and design of hardware-tailored software still represent a demanding challenge. Particle-In-Cell (PIC) methods such as Material Point Method (MPM) fits particularly well the fine-grained parallelism required by GPU based architectures. Even though several algorithms based on PIC and MPM have been developed in various fields and ported to GPU architectures, only few works have been published in the compressible computational fluid dynamics (CFD) domain since late 90s, not allowing to highlight and exploit their High-Performance Computing (HPC) potential in this area of application. This contribution aims to show results obtained in developing GPUs targeted software for MPM methods in compressible CFD simulations. The focus will be on the strategies adopted in order to obtain high levels of parallel scalability on GPUs architectures in industrial-grade clusters. All the results will be compared to some standard CFD test cases which are of interest in industrial applications.