Tobias Achterberg

What's next for GPU-accelerated optimization? We’ve seen how GPU-enabled algorithms can speed up solve times for some very large linear programs, but what does this mean for MIPs? https://ow.ly/5kNr50Y8uNX

Hmm... somewhere is a dot missing in this post. I wish the number was true. :)

Backing PDHG into a Corner Recent enhancements to the Primal-Dual Hybrid Gradient (PDHG) algorithm have enabled GPUs to efficiently solve large linear programming problems, often faster than the long-dominant simplex and ...

Another article by Ed Rothberg on PDHG.
arxiv.org/abs/2511.13894

Concurrent Crossover for PDHG First-order methods based on the PDHG algorithm have recently emerged as a viable option for efficiently solving large-scale linear programming problems. One highly desirable property of these methods...

Nice and short article by Ed Rothberg on concurrent crossover for PDHG (PDHG on GPU, crossover on CPU).
arxiv.org/abs/2510.24429

Nice report about the Gurobi Summit in Vienna from Tim Varelmann, winner of the tuning challenge:

www.bluebirdoptimization.com/blog-posts-e...

Gurobi is committed to investing in the next generation of #optimization talent—which is why we’re proud to announce the launch of #GEMS, the Gurobi #Engineering #MIP School.

Applications for the 2026 program are now open. Learn more and apply here: www.gurobi.com/lp/all/gems/...

I guess you are not surprised by the general approach that a company trains talent (in the hope that they will stay in the company), but more by the fact that we feel the need to do so for MILP.
Yes, we do. Yes, it's a significant investment.

A MIP Development Scholar is a full position at Gurobi with salary and benefits.

Our goal is to finish this year's interview process for the MIP Development Scholar positions end of March. A full interview cycle consists of one session with the hiring manager, two with R&D team members, and a final session with HR. Therefore, please apply soon to get a chance to be considered!

Joins us for a two year education program with the goal of becoming a full member of Gurobi's R&D, Expert or TAM teams! There will be lectures, seminars, interactive classes, C coding exercises and projects, customer and user interaction and lots of fun.

I am excited to announce the Gurobi Engineering MIP School: jobs.lever.co/GurobiOptimi...

We are looking for passionate programmers who like efficient algorithms, data structures, math and optimization and want to learn the ins and outs of mixed integer programming in theory and practice.

The #mip2025 computational competition is open! This year’s topic is finding feasible solutions for MIPs with quadratic functions. Student participation is encouraged and we may separately award outstanding student submissions. For more details, visit www.mixedinteger.org/2025/competi... #orms

Discover what’s new in #Gurobi12! Join one of our webinars this week to explore the latest features and enhancements including new #MINLP capabilities, revamped, user-friendly documentation, and upgrades to Cluster Manager and Compute Server. Save your seat today: www.gurobi.com/events/webin...

Discover how #GenerativeAI + Mathematical #Optimization are transforming industries and redefining what’s possible on Dec. 17th at 11am EST. Register Today: www.gurobi.com/events/the-i...

Don’t miss the chance to explore the future of GenAI + optimization during this live panel discussion!

There is now gurobi-ai-modeling.readthedocs.io which is aimed at the non-mathematician crowd for building optimization models from human-language problem descriptions.

2025 Mixed Integer Programming logo, designed by Julia Silbermann

Save the date! The 2025 Mixed Integer Programming Workshop will be held June 3–6, 2025 at the University of Minnesota in Minneapolis, MN. #mip2025 features a wonderful single-track program, poster session, and computational competition. Stay tuned for more details.

www.mixedinteger.org/2025/

#orms

We take the shifted geometric mean (with a shift of 1 second) over the run times and report the ratio(s) of these means.

Usually, we provide ratios for all non-trivial instances (at least one version needed at least 1 second) and for "hard" instances (at least 100 seconds). 3/3

We only consider instances for the comparison that can be solved to proven optimality (MIP gap 0.01%) by at least one of the Gurobi versions, and for which all versions agree on the optimal objective value. 2/3

What we do is the following: we run the old and new versions on our benchmark set (for MIP this is >8000 instances) with a 10000 second time limit on a cluster of about 185 identical (pretty old) 4 core Intel machines. 1/3