Running FDS in the cloud: 4 ways to get more computing power

fdsrun editorial team · 6 November 2026 · 6 min read

Your own server, a cloud VM, a service integrated with PyroSim or an FDS computing portal — how they work and what to look at: FDS version, cost forecast, billing, data in the EU.

A stairwell model or a car park with jet fans can run for days on a laptop. When you run out of computing power, there are in practice four ways forward. Below is a description of each and a list of things worth checking before you pay for the first hour of computing.

1. Your own workstation or server

A machine with many cores and plenty of RAM on which you run FDS with MPI (one process per mesh). Full control and no hourly fees, but an up-front purchase, and the machine sits idle between projects. At peak times — several variants at once before a deadline — you still run them one after another.

2. A cloud machine you set up yourself

A virtual server from a cloud provider (e.g. AWS, Azure, Hetzner): you install the NIST FDS bundle for Linux, copy the files, run mpiexec, watch the run and download the results. AWS has even described a reference setup for fire simulations on a cluster (ParallelCluster with the EFA fast network and the FSx for Lustre file system). You pay for the machine's running time, but you do all the administration yourself — and a forgotten server keeps billing, weekends included.

3. An FDS cloud service integrated with PyroSim

PyroSim has an Analysis → Run FDS Cloud command that sends the model to CFD FEA Service's Cloud HPC (currently the only service PyroSim integrates with). You need an account with that service and an API key; progress and results are followed in the provider's dashboard. Convenient if your whole workflow is in PyroSim.

4. fdsrun — a portal for FDS runs

You upload an .fds file (from PyroSim, BlenderFDS or written by hand) in the browser. The portal checks the file for errors, shows the forecast time and cost on several servers in the EU and, once started, the FDS console, HRR and device charts, and a 3D preview of smoke and slices. Every job runs on its own server that is deleted afterwards; you pay for the actual time and download results during the run and for 30 days after it ends.

What to look at when choosing

• FDS version. Can you choose the same version the model was prepared and checked in? Results can differ slightly between versions.
• Checks before the start. The most expensive error is the one that shows up after two hours of computing. Check the file first — for free in the .fds file check, for instance.
• Time and cost forecast. Do you know before starting how long it will take and what you will pay? What drives the run time is explained in how long does an FDS simulation take.
• Billing. Per hour or per second, with or without a minimum charge, and what happens to the machine when the run ends.
• Where the data lives. Building models are often the client's confidential data; servers in the EU make GDPR compliance easier.
• Live view and partial results. Can you see the .out file, the time step and the HRR as they happen, and download results before the end?
• Resuming after a failure. Does the run write restart files (DT_RESTART), and can it be finished without starting from zero?
• Invoices. For a business: a proper VAT invoice, with reverse charge for EU companies.

Whatever you choose, how the model is split into meshes and processes decides how many cores make sense — see multiple meshes and MPI in FDS and the FDS mesh size calculator.

Sources: PyroSim documentation (Thunderhead Engineering), “Cloud”; AWS HPC Blog, FDS fire simulations on AWS ParallelCluster; FDS User Guide (NIST SP 1019), chapters 3 and 6.3. FDS and Smokeview are software of the National Institute of Standards and Technology; NIST is not affiliated with fdsrun.