Hiltscher, JohannesCunningham, Douglas W.Hofstedt, PetraMeer, KlausSchmitt, Ingo2017-06-302017-06-302015978-3-88579-640-4Accelerators have gained an important role in recent years. While being used primarily in the scientific community in the beginning, they are now employed in a wide range of every day applications. Accelerators can hence be viewed in the focus of machine consolidation and virtualization, offering new opportunities for cost saving and services. Although these opportunities have been discussed intensively, today's commercially available accelerators are not well suited for such scenarios. More specifically, they lack support for partitioning, i.e. the concurrent assignment of subgroups of their computation elements to different applications. Mechanisms for indirect partitioning support exist but have inherent limitations. This paper presents a hardware supported approach for efficient accelerator partitioning which overcomes the limitations of existing solutions. SR-IOV is used to create an accelerator able to enforce guarantees on parameters like memory bandwidth.enText/Conference Paper1617-5468