GAMS

Most optimization people are trained to build models. Not many are trained to build something that actually survives production. This conversation between Borja Menéndez Moreno and Dr. Tim Varelmann from the Advent of OR gets into that. The gap between open source and commercial tools isn’t really about features. It’s more about what happens when things get bigger. Can it actually handle the scale, or does it start to slow down? And then you realise it’s not just the solver, but everything around it. Model structure, data vs logic, how you deal with sparsity. That’s the gap GAMSPy is bridging, bringing that battle-tested GAMS logic into a Python workflow. Part of our curated Advent of OR playlist 👉 https://lnkd.in/dBUM33Uj #OperationsResearch #Optimization #AdventofOR

In Mixed-Integer Programming, the optimality gap is the gold standard but it doesn’t tell the whole story. In the real world, an equally critical question is often: “How fast can this solver find a high-quality feasible solution?” Our latest blog dives into this via the MIPfeas benchmark, utilizing the Primal Integral metric. Unlike static end-of-run results, this metric tracks how solution quality evolves over time. Some highlights: 🔶 The benchmark includes well known open-source MIP solvers Cbc, HiGHS, and SCIP.  🔶 GPU-based newcomer NVIDIA cuOpt with innovative GPU based primal heuristics is included in the benchmark. 🔶To navigate restrictions on benchmarking commercial codes, we’ve synthesized COPT, CPLEX, and Xpress into the Virtual Best, Mean, and Worst commercial solvers. 🔶 The benchmark covers 233 MIPLIB 2017 instances and uses a fully reproducible pipeline built around the GAMS solvetrace facility. Read the full blog and explore the results here: https://lnkd.in/dfbuxRU2 #Optimization #OperationsResearch #MIP #GAMS

I’m excited to share a new benchmark, MIPfeas, which shifts the focus in Mixed-Integer Programming from "final optimality" to "early feasibility." Using the Primal Integral metric, we evaluated how solvers like HiGHS, SCIP, and NVIDIA cuOpt deliver high-quality feasible solutions within a short time window. While direct comparison with commercial MIP codes remains difficult, we have grouped the results of three major commercial solvers—COPT, CPLEX, and Xpress—into three synthetic baselines: the Virtual Best, Mean, and Worst Commercial Solvers in this benchmark. A huge thank you to Hans Mittelmann, the NVIDIA team (Burcin Bozkaya, Akif Çördük, and Chris Maes), and ZIB Zuse Institute Berlin for the technical collaboration. Special thanks also to Timo Berthold, Gerald Gamrath, and Ferenc Katai for their expert insights. Check out the full results here: https://lnkd.in/dSfNc58v #Optimization #MIP #GAMS

How far can GPUs push large-scale optimization? That question was front and center this week at the 110th German Operations Research Society Workshop on High Performance Computing in Bad Honnef. In a joint session, NVIDIA’s Akif Çördük gave an introduction to CUDA for Linear Optimization and Frederik Fiand presented results from a computational study on 50 large-scale LP instances from energy system models, some reaching hundreds of millions of non-zeros. Rather than framing the discussion as CPU vs GPU, the focus was more practical: Which algorithms and architectures perform best for which optimization problems? A few highlights from the session: 🔶 First-order methods at scale Methods like PDLP/PDHG are showing promising performance on extremely large LPs on GPU architectures. 🔶 Evaluating the trade-offs GPUs offer huge potential, but hardware cost and solution accuracy still matter when comparing with traditional CPU barrier algorithms. 🔶 A rapidly evolving ecosystem Great discussions with researchers from the DLR Institute of Networked Energy Systems, Zuse Institute Berlin, and MOSEK ApS about where large-scale optimization is heading. And it wasn’t all matrices and benchmarks, the workshop also included a scenic Drachenfels railway trip, check out the stunning views!  Thanks to everyone who joined the discussion, to the organisers and especially to Fred and Akif for presenting the study. Learn more about GPU Accelerated Optmization with GAMS and NVIDIA cuOpt  https://lnkd.in/eZBkzCzW #OR #Optimization #HPC #EnergyModeling #GAMS

GPUs power everything from AI to climate modeling at incredible scale. But large-scale optimization is still largely a CPU domain. We’re at the Physikzentrum Bad Honnef this week for the German Operations Research Society Working Group on High Performance Computing where today at 1pm, Frederik Fiand and NVIDIA’s Akif Çördük are running a joint session on CUDA programming and large-scale optimization benchmarks: 🔶 Akif will give a primer on CUDA programming for optimization 🔶 Fred will present benchmark results on large-scale energy system LPs The key question: can first-order methods like PDLP make GPUs practical for optimization at this scale? Early results are promising as PDLP can outperform CPU barrier solvers on very large models, but there are still trade-offs around accuracy and hardware costs. Looking forward to the discussion and to catching up with colleagues from the PEREGRINE project (Manuel Wetzel, Thorsten Koch and Nils-Christian Kempke) and the QuSol project (Laurin Demmler) as well as the wider OR community. Huge thanks to Jens Schulz, Josef Kallrath and Thorsten Koch for organising. cc Zuse Institute Berlin, MOSEK ApS, DLR Institute of Networked Energy Systems

We are proud to announce that senior Emilie Lesinski is the 2nd place winner of the annual GAMSPy student competition! GAMS, one of the leading tool providers for the optimization industry, hosts this annual competition as part of the GAMS Academic Program. For her project, Emilie tackled the challenge of optimizing her post-graduation trip across Europe. She built a multi-objective model to balance travel costs, transit time, and the number of cities visited. The result was a data-driven, efficient and robust plan to execute her adventure - and capture second place in the competition! Congratulations, Emilie! UW-Madison College of Engineering University of Wisconsin-Madison https://lnkd.in/gDweHK5J

The latest edition of the GAMS Update is out and its a doozy. This month we’re digging into: 🔶 The winning GAMSPy student project that designed a severe-weather radar network for Texas, improving coverage and delivering real infrastructure impact 🔶 Supercomputer-scale modelling projects helping energy systems and supply chains navigate uncertainty 🔶 Solver engineering updates that push optimisation performance to new limits 🔶 A lot more! If you’re interested in optimisation, industrial modelling, or the technology shaping decision-making at scale, it’s worth a read.

Last week, Mateo Gallia & Vaibhavnath J. ran a hands-on GAMSPy workshop for 30+ students at the Aristotle University of Thessaloniki (AUTH). A huge thanks to Georgios Georgiadis for hosting us, we’re excited to keep working with AUTH to bring optimization and Python-powered modeling into the curriculum and support the next generation of industrial engineers. The Highlights: 🔸Theory to Practice: A deep-dive session bridging the gap between optimization theory and modern Python tools. 🔸GAMSPy in Action: Writing and solving the Capacitated Lot Sizing Problem (CLSP) with the power of algebraic modeling. 🔸 Academic Program: Supporting students research and learning through our free academic resources Great to see the back-and-forth on practical modeling skills that translate directly to industry! ◆ Want the slides or a classroom demo? Comment below or DM us. ◆ Explore our Academic Program: https://lnkd.in/dWu_5wSY #GAMS #GAMSPy #Optimization #Python #IndustrialEngineering #HigherEd #STEM

Great to visit University of Zagreb / Sveučilište u Zagrebu this week for the Third General Meeting of ROAR-NET. 🇭🇷 Our very own Hamdi Burak U. took the stage to talk about a challenge many of us face in optimization: Finding the right balance between heuristic methods and exact solvers. While heuristics are often the go-to for robust performance in the field, complex problems still demand the precision of an exact mathematical approach. Burak’s presentation focused on how we can bridge that gap using #GAMSPy. He shared some specific workflows on: 🔸Integrating heuristics as standalone components within a larger GAMS framework. 🔸How the GAMS approach to modeling compares with emerging ROAR-NET methodologies. 🔸Using Python to facilitate these hybrid workflows for better precision. It was a productive few days comparing notes with peers from Hexaly, ESTECO, Dots & Lines Ltd. and the broader research community. A big thanks to ROAR-NET COST Action for the hospitality! Get started with GAMSPy 👉 https://lnkd.in/dpQKUMnY 📸 credit to: Barbara Blečić #GAMS #GAMSPy #Optimization #ROARNET #OperationsResearch #MathematicalModeling #Python

We've updated our blog on GPU-Accelerated Optimization with GAMS and NVIDIA cuOpt to cover the February 2026 release of the GAMS/cuOpt solver link (v0.0.5d) for NVIDIA cuOpt 26.02. This comprehensive update introduces major architectural and algorithmic enhancements for both Large-Scale LPs and Mixed-Integer Programming (MIP): 🔶New PSLP Presolver: Architected specifically for very large-scale LPs, this robust, dual-preserving presolve replaces the PaPILO presolver to prevent memory exhaustion and time-outs on massive instances. 🔶 Expanded Hardware Support: The solver link now supports arm64 CPU architectures alongside x86_64. 🔶Advanced MIP Capabilities: We have exposed numerous new MIP-related options, including mip_trace for detailed solve insights and mip_start to utilize initial variable levels as a starting MIP solution. 🔶Initial Solutions: Support has been added for passing initial primal and dual solutions for both LP and MIP models. 🔶Solution Verification: For users leveraging first-order GPU methods like PDLP, we detail how to use the GAMS/Examiner tool to independently verify primal feasibility, dual feasibility, and true optimality. Read the updated technical breakdown and deployment guide here: https://lnkd.in/eZBkzCzW Explore the potential of cuOpt with GAMS and GAMSPy for your most challenging optimization tasks. To get started with cuOpt (or any other solver) on GAMS Engine SaaS, request a free test account by contacting sales@gams.com. #GAMS #NVIDIA #cuOpt #Optimization #OperationsResearch #MixedIntegerProgramming #LinearProgramming #HPC