A customized approach to Pseudo Random Number Generation (PRNG) is developed specifically for the highly parallelizable sensor models in the ground vehicle autonomy application domain. The work considers three desirable attributes (namely quality, efficiency and determinism). Furthermore, the application demands high fanout (1:1Million+) seeding of traditional PRNGs. An approach using hash functions to generate the seeds for the PRNGs, each of which generates a small (i.e. 20) run of numbers, to handle determinism is investigated. Quality and efficiency are evaluated for multiple combinations of hash functions and PRNGs and a pareto front is created. Quality assessments were performed using industry standard testing suites (TestU01 and PractRand) and efficiency of various hash, PRNG, and batch size combinations was benchmarked on Windows/x64, ARM and NVIDIA/CUDA architectures.