Hardware
Scaling Decision Optimization to 100 Million Variables and Beyond with mPDLP in NVIDIA cuOpt
NVIDIA cuOpt’s Multi‑GPU Primal‑Dual Hybrid Gradient (mPDLP) solver distributes linear programming problems across NVLink‑connected GPUs, cutting per‑GPU peak memory by up to six times and handling up
NVIDIA cuOpt’s Multi‑GPU Primal‑Dual Hybrid Gradient (mPDLP) solver distributes linear programming problems across NVLink‑connected GPUs, cutting per‑GPU peak memory by up to six times and handling up to 2.1 billion nonzeros. Benchmark tests on DGX B200 platforms show speedups that increase with problem size—up to 11.4× for PDLP steps alone and 3–5× overall for instances exceeding 10⁷ nonzeros, such as supply‑chain and stochastic energy models with 135–185 million variables. The approach employs min‑cut partitioning of a bipartite dependency graph to minimize cross‑GPU communication, leading to consistent performance gains that vary with sparsity structure.
Related
- Optimize Supply Chain Decision Systems Using NVIDIA cuOpt Agent Skills
- How NVIDIA NVLink 6 Delivers Multi-Layer Resiliency for AI Factories
- NVIDIA NVLink Fusion Brings NVHBM to Next-Generation AI Infrastructure
Source: NVIDIA Developer | 2026-10-07