cuOpt

[Eugene Freuder]

What do you all think about this:

https://www.nvidia.com/en-gb/ai-data-science/products/cuopt/

[Fabio Tardivo]

It is nice to see NVIDIA working on this, especially considering that it is not LLM-related.

The underlying theory makes a lot of sense for exploiting the computational power of GPUs.

CP has a very rich and expressive representation, which does not naturally fit the regular, uniform style of computation that GPUs are designed for.

A number of recent papers have explored different approaches to addressing this challenge, either by making the representation more uniform or by accelerating specific propagation algorithms.

[steven kelk]

Hi everyone,

I think cuOpt is already a few years old. I believe they started incorporating GPU-based LP algorithms somewhere in 2024. Gurobi have already been experimenting with GPU-based

PDHG (Primal-Dual Hybrid Gradient) algorithms for solving LP for a similar length of time, see https://support.gurobi.com/hc/en-us/articles/360012237852-Does-Gurobi-support-GPUs . I don't know

whether the two projects are linked, but it looks like Gurobi were consulting with NVIDIA in the early days of this development.

Gurobi have always been open about the limitations of GPU to help with integer optimization (see https://www.gurobi.com/resources/blog/using-gpus-to-solve-lps-vs-mips-what-s-the-difference), which I assume

is the use case most relevant to this group --

Cheers,

Steven

Op ma 31 aug 2026 om 02:32 schreef Fabio Tardivo <tardiv...@gmail.com>:

--
You received this message because you are subscribed to the Google Groups "Constraints" group.
To unsubscribe from this group and stop receiving emails from it, send an email to constraints...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/constraints/2b72c952-2011-434b-b4cc-d6669576a30dn%40googlegroups.com.