General
Multilevel Graph Partitioning on GPU Supercomputers
arXiv cs.DC0 views
arXiv:2611.24264v1 Announce Type: new Abstract: Graph partitioning is a key enabler in computational science, with applications to sparse linear algebra, load-balanced parallel computation, and graph computations. In recent years, graphics processing units (GPUs) have become the preferred accelerator on the world's top supercomputers. Despite this, most existing distributed graph partitioners do not utilize GPUs. In this work, we develop a new multi-GPU distributed partitioner based on Jet, a prior single-GPU partitioner. We seek to optimize scalability without sacrificing quality. Our partitioner utilizes CUDA-aware MPI, one-sided communication routines, and an overall bulk-synchronous parallel structure. We demonstrate solution quality comparable to Jet. Our experiments show excellent weak scaling up to 256 GPUs, enabling partitioning of instances significantly larger than a single-GPU partitioner can handle.
Source:https://arxiv.org/abs/2611.24264PreviouseFAIR: Fair LLM Serving under Hourly Electricity CommitmentsNextScaleBench: An Extensible and Reproducible Experimental Platform for Kubernetes Autoscaling Research
Related posts
Do not buy RAM, compress
0Hacker News
DeepMind Employees Are Unionising
0Hacker News
PHP 8.6 Release Announcement
0Lobsters
At Scout Motors’ New Plant, Validation of Terra and Traveler EVs Has Begun
0CNET News
Ford Designed Its Fathom EV Battery Pack to Bend in a Crash, for Safety
0CNET News
Supermarket refrigerators got hacked. Make sure your smart home devices are secure
0Phys.org Tech