Search papers, labs, and topics across Lattice.
Universidad de Cantabria, Barcelona Supercomputing Center
2
0
2
Non-convex resource allocation strategies can surprisingly outperform traditional convex approaches in HyperX networks, offering a new path to optimize HPC system performance.
MRLS networks offer a 50-100% speedup over traditional Fat-Tree and Dragonfly topologies for large-scale collective communication in data centers.