SU2
SU2 in the cloud: compressible CFD and adjoint shape optimisation
Upload your configuration file and mesh, choose the SU2 version and the number of cores, and launch. cloudHPC partitions the mesh, runs SU2 in parallel and merges the solution for you. You pay only while the machine runs.
An open-source suite for multiphysics analysis and design
SU2 is a C++ and Python suite for solving partial differential equations and PDE-constrained optimisation problems on unstructured meshes. It is developed by a worldwide community, promoted by the non-profit SU2 Foundation and released under the LGPL licence. It is used in aeronautics, automotive, naval and renewable energy applications.
Compressible flows
- Euler, Navier–Stokes and RANS from subsonic to supersonic
- Spalart–Allmaras and SST turbulence models
- Non-ideal compressible fluids for turbomachinery and ORC
Discrete adjoint
- Sensitivities of drag, lift or losses to every surface point, at the cost of about one extra solution
- Gradient-based shape optimisation with the Python tools
Incompressible and heat transfer
- Incompressible flows with energy equation
- Conjugate heat transfer and structural solvers in the same code
SU2 and OpenFOAM®, side by side on NASA test cases
We ran SU2 and OpenFOAM on three cases of the NASA Turbulence Modeling Resource with the Spalart–Allmaras model, on the same grids and on the same cloudHPC machines, up to grids of tens of millions of cells. The image at the top of this page shows the skin friction on the ONERA M6 wing, SU2 on the left and OpenFOAM on the right: both capture the lambda shock that crosses the wing. The results will be presented at the 7th SU2 Conference in Milan.
3D bump-in-channel
Grid convergence against the NASA reference codes, up to the finest NASA grid.
NACA 0012
Pressure and skin friction at 0°, 10° and 15° angle of attack, against experimental data.
ONERA M6
Transonic wing against wind tunnel data. On the newest AMD EPYC Turin machines, one iteration of the 8.65 million cell grid takes about one second on 192 cores.
How an SU2 run works
What you upload
- One
.cfgconfiguration file - The mesh it references, in SU2 native or CGNS format
- Restart files, if you want to continue from a previous solution
What cloudHPC does
- Partitions the mesh and runs SU2 with MPI on every physical core
- Merges the solution at the end
- Keeps
history.csvand the solver log for convergence checks
Meshes with cfMesh
cfMesh and CF-MESH+ generate tetrahedral meshes with boundary layers and export them in CGNS for SU2. Learn more
Meshes with Ennova
Import and repair CAD, mesh it and export to CGNS, on a remote desktop next to your runs. Learn more
Post-process in ParaView
Open the results on a remote desktop with the RAM they need, without downloading them. Learn more
What changes compared with your workstation
| Step | On a workstation or in-house cluster | On cloudHPC |
|---|---|---|
| Installation | Compile SU2 with MPI and keep it updated | Several SU2 versions pre-installed with MPI |
| Parallel run | Set up the MPI launch and the partitions for each machine | Partitioning and merging done automatically |
| Hardware | Limited to your cores and RAM | Up to 192 physical cores per run, with the RAM type you choose |
| Optimisation | Each design iteration waits for the previous one | Large meshes solved fast, and independent designs run at the same time |
| Cost | Hardware sized for the peak, idle the rest of the time | Pay per vCPU hour, only while the run lasts |
How many cores does my case need?
Rule of thumb
- SU2 runs one MPI process per physical core: choose highcore or hypercore, where vCPU = cores
- Start with about 50,000 cells per core
- hypercore uses the newest CPU generation and is the fastest option
Memory
- In our ONERA M6 runs SU2 needed roughly 10 GB of RAM per million cells
- For large grids pick a machine with enough memory, or spread the grid over more cores
- Not sure? Our team can help you size the run
CFD infrastructure, ready when you need it
Expert support
Our team runs SU2 and OpenFOAM every day and can help you set up, scale and debug your cases.
Pay per use
No subscription and no licence: you pay per vCPU hour, only while the machine runs. Check the cost with the price calculator.
Automate with the API
Drive optimisation loops and batches from your own scripts with the cloudHPC API. The platform is ISO 27001 certified.
Your first SU2 run in five steps
- Register on cloudHPC and activate your account.
- Prepare the
.cfgfile and the mesh, or mesh with cfMesh on cloudHPC. - Upload them to a storage folder.
- Choose the SU2 version and the cores on highcore or hypercore, and launch.
- Check convergence in
history.csvand post-process in ParaView.
Run your next SU2 case in the cloud
Create your account and get 300 free vCPU hours to try SU2 on your own case.