O
OSQP
DEVOPS
Velocity0.0
Operator splitting solver for convex quadratic programming problems
QP solver that rebuilt its linear algebra layer for GPUs, then went quiet in beta.
convex-optimizationquadratic-programminggpu-accelerationembedded-solversbeta
◆Current state
OSQP is an operator-splitting quadratic program solver used widely as an embedded optimization backend. Its visible release history stops at the v1.0 beta from May 2023, which was a substantial rework: a pluggable linear algebra backend chosen at compile time, the separate cuOSQP GPU project folded into the main tree, and an Intel MKL backend added. That beta has not been followed by a stable v1.0 in this feed.
◆Where it's heading
The direction the beta set is clear — OSQP was becoming a solver core that could be retargeted at different compute hardware without changing the algorithm, with CUDA and MKL as the first two targets. Code generation and adjoint derivative computation moving into the C API point the same way, toward embedding OSQP inside other systems including differentiable pipelines.
◆Prediction
The stable v1.0 is the obvious next milestone, but nothing in these entries indicates when or whether it landed — treat the backend rework as available only in beta form.
◆Recent moves
- 3y ago
Beta re-tag adding CI-generated build artifacts
Adds CI-generated build artifacts for several platforms on top of the beta0 release published hours earlier. Packaging convenience with no change to the solver.
View source ↗ - 3y ago
Pluggable linear algebra backends, cuOSQP merged in, MKL backend added
⚡ SPARKThe release that defines OSQP's current shape and the last substantive one in this feed. It turns a single-implementation solver into one with swappable compute backends, and pulls the GPU fork back into the main project.
View source ↗ - 5y ago
Development version bump to 0.6.2.dev0
A bare version bump from 0.6.0 to 0.6.2.dev0 with no accompanying content. Feed noise rather than a release.
View source ↗