# AMD introduces rocOpt for GPU-accelerated decision optimization

> **Key takeaway:** rocOpt is an open-source engine that brings GPU acceleration to large-scale mathematical programming problems on AMD Instinct hardware.

**Published:** 2026-10-20T00:00:00+00:00 
**Source:** AMD ROCm Official Blog 
**Category:** ai-local-edge 
**Canonical URL:** https://fosswire.org/news/amd-introduces-rocopt-for-gpu-accelerated-decision-optimizat.html 

## Summary
AMD has launched rocOpt, a new open-source engine designed to accelerate decision optimization workloads on AMD Instinct GPUs. The engine targets large-scale mathematical programming tasks such as linear programming (LP), mixed-integer linear programming (MILP), and vehicle routing problems. By leveraging the ROCm software stack, rocOpt allows users to run complex optimization models on the same hardware used for AI and analytics. It is currently available for MI300X and MI355X GPUs running ROCm 7.2.3.

## Technical breakdown of rocOpt solvers
- **Solver Families:** The engine includes four specialized families: PDLP (a first-order method for massive, sparse LPs), Dual Simplex (CPU-based for small-to-medium LPs), a hybrid GPU/CPU MILP solver, and a metaheuristic solver for Vehicle Routing (TSP/VRP).
- **Software Integration:** Built on the AMD Data Science stack, it utilizes HIP for device runtime, hipBLAS/hipSPARSE for linear algebra, and hipSOLVER for direct linear solvers.
- **Hardware Support:** Optimized specifically for AMD Instinct accelerators, with immediate availability on MI300X and MI355X architectures.
- **Deployment Options:** Users can build from source via GitHub or utilize prebuilt Docker containers for easier deployment.

## Why it matters
rocOpt expands the utility of AMD Instinct GPUs beyond AI/ML, enabling high-performance mathematical optimization within the open-source ROCm ecosystem. This provides a competitive, non-proprietary alternative for industries like logistics and energy dispatch that require massive parallel processing for decision-making.