high performance computing on graphics processing units: hgpu.org

We present a multi-GPU graph processing library that allows programmers to easily extend single-GPU graph algorithms to achieve scalable performance on large graph datasets with billions of edges. Our design only requires users to specify a few algorithm-dependent blocks, hiding most multi-GPU related implementation details. Our design effectively overlaps computation and data transfer and implements a just-enough memory allocation scheme that allows memory usage to scale with more GPUs. We achieve ~20 GTEPS peak performance for BFS, demonstrating a ~6X speed-up with ~2X total GPU memory consumption on 8 GPUs. We identify synchronization/data communication patterns, graph topologies, and partitioning algorithms as limiting factors to further scalability.