Before the numbers
Most published comparisons are dishonest in the same way: they set a one-time build cost against years of licensing, and omit maintenance, validation and knowledge retention. This one includes them, which makes the case for building narrower — and more credible.
Licence renewal season prompts this question in a lot of engineering organisations. It deserves a proper model rather than frustration-driven arithmetic.
What each option actually costs
| Commercial licence | Owned solver |
|---|---|
| Annual licence, scaling with cores | One-time development |
| HPC pack costs for parallel runs | Annual maintenance (15–20% of build) |
| Additional modules for extra physics | Cost per new physics model, plus validation |
| Vendor support included | Your engineers are the support |
| Training on the tool | Knowledge retention and documentation |
| Price rises at vendor discretion | Compute cost only, which trends down |
The variables that decide it
- Core-hours per year The single biggest factor. Licensing that scales with cores punishes exactly the large simulations you want to run.
- Breadth of physics needed A narrow, repeated analysis is cheap to own. Broad, varied physics is expensive to own and is what commercial packages are for.
- Time horizon Ownership is a capital investment that pays back over years. If your programme is 18 months, licensing almost certainly wins.
- Whether you must embed or distribute If simulation ships inside your product, commercial licences typically prohibit it and the comparison collapses to one option.
- In-house capability Owning a solver you cannot maintain is worse than licensing one you can use.
The variable people underweight is breadth. Teams model the cost of replacing their current analysis and forget that next year someone will ask for conjugate heat transfer, or a multiphase case, or a new turbulence model. With a licence that is a module. With an owned solver it is a development project plus fresh validation.
Structuring an honest comparison
Rather than quoting figures that will not match your situation, build the model with these inputs:
- Annual licence and HPC pack cost at your actual core count
- Expected annual licence escalation
- Estimated build cost for the specific, narrowed physics scope you need
- Annual maintenance at 15–20% of build
- Cost of the validation campaign, and of re-validation when physics changes
- Compute cost, which both options pay
- A realistic figure for engineering time lost to owning and supporting the code
Sum both over five years. If the answer is close, licence — because the owned option carries execution risk that a spreadsheet does not capture.
If ownership only wins by a small margin on paper, licensing wins in reality. Build only when the gap is large enough to absorb the project going 40% over.
When ownership clearly wins
- Very high sustained core counts where per-core licensing is the binding constraint on simulation size.
- One narrow analysis run constantly — a specialised solver can be dramatically faster than a general one, compounding the saving.
- Embedding simulation in a product you sell.
- Export control or data residency ruling out the commercial option entirely.
- Physics the package does not implement and cannot be extended to.
When licensing clearly wins
- Varied, exploratory analysis across many physics types.
- Modest core counts where licence cost is not the constraint.
- No in-house numerical software capability to maintain a solver.
- Short programme horizons.
- Regulatory or customer requirements naming specific commercial tools.
The hybrid most large engineering teams settle on: a small number of commercial seats for exploration and unusual cases, plus an owned solver for the single repeated analysis that consumes the bulk of core-hours. You capture most of the licensing saving on the workload that drives it, without taking on breadth you cannot maintain.
A staged way to de-risk the decision
- Measure your actual core-hours by analysis type. The distribution is usually far more concentrated than people expect.
- Identify the dominant repeated case. That is your candidate for ownership, not your whole workload.
- Prototype in a framework — a custom OpenFOAM solver proves the physics at a fraction of from-scratch cost.
- Benchmark honestly against your tuned commercial setup on the same cases.
- Only then commit, with validation cases and acceptance criteria agreed up front.
Want this modelled against your real usage, including an honest answer if licensing wins? Tell us your core counts and analysis mix. See our CFD solver service and the build cost breakdown.