NVIDIA Released New Multi-GPU Linear Solver

The platform now handles massive optimization models with enhanced speed and reduced memory usage.

Updated on Oct. 8, 2026 in Quantum Computing

Isometric editorial illustration of a stack of computing processing units with interconnected bridges, representing advanced multi-GPU computational power. — NVIDIA has launched a new multi-GPU linear solver within its cuOpt platform, enabling faster processing for large-scale optimization models in supply chain and energy systems. AI Illustration. Upload story photo >

Live Poll

Do you believe new high-speed computing tools will make supply chains more reliable for your household?

NVIDIA has introduced a multi-GPU solver within its cuOpt platform designed to process complex linear-programming models. The technology allows for calculations involving over 100 million variables and billions of matrix entries.

Why it matters

Linear programming is critical for organizations managing constrained supply chains or complex energy systems. This solver enables faster decision-making by distributing heavy workloads across multiple GPUs.

The solver manages models with over 100 million variables and 2.1 billion nonzero matrix entries. It utilizes min-cut partitioning to distribute calculations, resulting in up to 6x lower peak memory usage per GPU.

The players

NVIDIA

NVIDIA is a multinational technology company that designs graphics processing units and hardware for artificial intelligence and high-performance computing.

Kinaxis

Kinaxis is a supply chain management software company that provides cloud-based solutions for integrated business planning.

PSR

PSR is a firm specializing in energy and power sector consulting, modeling, and market analysis.

The details

By leveraging NVLink to connect multiple GPUs, the system effectively partitions sparse matrix calculations to minimize communication overhead. Testing on NVIDIA DGX B200 systems demonstrated significant speed improvements for supply-chain and energy-expansion models.

October 8, 2026: NVIDIA introduced the new multi-GPU solver.

The Tech Race

This development marks a significant advancement in the NVIDIA cuOpt platform, extending its reach into complex linear-programming optimization. By scaling these models across multiple GPUs, NVIDIA is directly competing to reduce the computational bottlenecks that legacy single-GPU systems face.

Organizations relying on these models can expect faster processing times and reduced hardware memory constraints for large-scale supply chain or energy planning. This allows data scientists to iterate on complex optimization problems more efficiently than previous methods permitted.

The takeaway

Advanced linear-programming solvers are moving toward parallelized, multi-processor architectures to keep pace with modern data volume. Businesses can leverage these tools to significantly shorten the time required to solve complex resource allocation and distribution challenges.