Linear solvers are often a bottleneck in simulation runtimes, and static, hand-tuned solver settings rarely deliver optimal performance across every solve. PreconNet by Ignition Computing is a user-friendly software product that eliminates this bottleneck by optimising your matrix solves at runtime.
Whether you are running complex CFD simulations (DNS, RANS, or LES) using FVM or FEM discretizations, or leveraging simulation frameworks built on solver libraries like PETSc or HYPRE, PreconNet does the heavy lifting for you without requiring changes to your existing code.
Think of PreconNet as a real-time co-pilot for your simulation software: always watching, always adapting, and choosing the fastest path to convergence.
Are you developing or using simulation codes? Let's explore how PreconNet can speed up your simulations.
At every linear solve, PreconNet dynamically analyses your matrix and instantly selects and configures the best combination of solver, preconditioner, and initial guess, including all associated parameters, to minimise solve time while keeping convergence fully under your control.
Dynamically choose the best solver–preconditioner combination and optimize all associated parameters at runtime.
Using the prediction of the solution as input for a linear solver ensures any errors are still corrected up to defined error tolerances.
Generate and reuse preconditioners to further reduce solve time across successive linear systems.
PreconNet integrates with your simulation software through a lightweight adapter that communicates with the PreconNet API. At each linear solve, the adapter sends a compact set of 0D matrix characteristics to PreconNet. These scalar descriptors are non-identifiable: they do not expose the matrix entries or any problem-specific data, and the original matrix cannot be reconstructed from them. Based on these characteristics and continuously updated performance data from the server, PreconNet returns a recommended solver configuration. The figure on the right illustrates this process for a typical local deployment.
Free up local workstations and reduce expensive cloud HPC billing by letting PreconNet find the fastest path to convergence at every solve.
Run more complex, higher-resolution models within your existing time budgets to catch design flaws earlier and iterate with confidence.
Eliminate the need for your engineering team to spend valuable hours manually benchmarking and tweaking solver parameters, PreconNet handles it automatically.
PreconNet embeds directly into your existing simulation pipeline, delivering immediate speedups without requiring you to overhaul your current workflows or rewrite your core simulation code.
Are you developing or using simulation codes? Let's explore how PreconNet can speed up your simulations.