Biedert, TimWerner, KilianHentschel, BerndGarth, ChristophAlexandru Telea and Janine Bennett2017-06-122017-06-122017978-3-03868-034-51727-348Xhttps://doi.org/10.2312/pgv.20171094https://diglib.eg.org:443/handle/10.2312/pgv20171094An increasingly heterogeneous system landscape in modern high performance computing requires the efficient and portable adaption of performant algorithms to diverse architectures. However, classic hybrid shared-memory/distributed systems are designed and tuned towards specific platforms, thus impeding development, usage and optimization of these approaches with respect to portability. We demonstrate a flexible parallel rendering framework built upon a task-based dynamic runtime environment enabling adaptable performance-oriented deployment on various platform configurations. Our task definition represents an effective and easy-to-control trade-off between sort-first and sort-last image compositing, enabling good scalability in combination with inherent dynamic load balancing. We conduct comprehensive benchmarks to verify the characteristics and potential of our novel task-based system design for high-performance visualization.I.3.2 [Computer Graphics]Graphics SystemsDistributed/network graphics10.2312/pgv.2017109463-71