OptiHPC by IgnitionComputing · since 2020
Smarter simulations
Before it exists, it's simulated. OptiHPC optimizes every layer of simulating on HPC, so your results are trustworthy and your runs keep up with your engineering speed — and every run makes the next one smarter.
“Tuned” is a state, not an achievement
Keeping simulations correct and quick requires the concerted effort of several specialists, supporting your business — and the gains you hand-tune decay with every new machine, new regime, and code release. The hard part is staying tuned, but the common strategy of relying on procedures does not work. Discipline loses to automation, so we instead focus on minimizing reliance on human effort.
One objective, four levers
Every layer of simulating on HPC has levers. OptiHPC optimizes the entire ladder, data-driven: Every level records metadata, models train and optimize, and predictions return for the next runs. You focus on the simulations that matter to your business, and your scientists and engineers stay at the controls.
“Where should the work run?”
Wraps your run script and decides how the job should be sized for this case, on this machine — based on domain expertise combined with learned knowledge about your runs, instead of folklore core counts.
“Are we solving the equations right?”
Rides your runs and sets solver, preconditioner, and initial guess per solve. As it does not touch discretization, convergence criteria, or physics, and therefore works with zero risk, it is the building block that our higher-level products rely on.
“How converged is converged enough?”
Sets tolerances and relaxation to the digit budget your result actually needs, avoiding paying for every unneeded digit, many times per run. Provides you with information required for trade-off between accuracy and computational cost.
“How much work needs to exist?”
Time step and mesh decide how much work exists and are therefore the biggest lever. OptiMesh adapts them under quantified error budget.
Delivery through set of secure and user-friendly adapters that require minimal adaptation or recompilation, plus a local library and a datastore with models — on our servers or fully on-premise. Your data is never shared with anyone else.
Proof, auditable in your logs
OptiHPC products ride your runs and report their decisions, so you can audit them. The predictions come with confidence intervals, and the models are trained on your own runs. Some results for OptiSolve, measured against the baseline:
Speed-ups on the solver stage — typically 20–35% of total simulation time — measured per case.
Built in Eindhoven, since 2020
OptiHPC is developed by Ignition Computing — experts in nuclear fusion, physics, mathematics, and computer science. The products encode that expertise as data-driven software; the consultancy applies it directly to your codes with development, integrations, and support.
Book a first meeting
Have a chat with us to discuss your needs and how we can help you.
Our business model keeps us honest: we are compensated only for making your simulations correct and quick.
Nothing is sent from this page — the button opens a pre-addressed message in your own mail app.
