Distributed Linear Programming on GPU Clusters at Extreme Scale
arXiv preprint, 2026
SHARDLP is a distributed GPU solver for linear programs that exceed a single compute node’s memory. It keeps the matrix and primal-dual state partitioned from sharded input through solution output. Separately checked multi-node solves reach up to 13.604 billion variables and 40.807 billion nonzeros. On the largest Google PDLP benchmark, eight H200 GPUs solve a 1.185-billion-variable problem in 9.9 minutes.
Deza, A., Dey, S., & Van Hentenryck, P. (2026). "Distributed Linear Programming on GPU Clusters at Extreme Scale." arXiv preprint arXiv:2609.09108.