DBNS
Density-Based Navier–Stokes
Flow regime
Compressible, transonic–hypersonic
Primary use
External aerodynamics, turbomachinery, high-speed flow
AI Agentic CFD on proven numerics
Describe the case to an agent. Leave the solving to the GPU.
Wake CFD brings GPU-native density-based CFD to external aerodynamics. Flux Reconstruction (FR) and pressure-based (PBNS) solvers are in development. The Community Edition is free.
5–33×
DBNS GPU speedup — one CPU chip vs. one GPU
M 0.05–3
Validated Mach range
3
Solver families — DBNS; FR and PBNS in development
Free
Community Edition — single GPU
What sets it apart
Capabilities that carry real analysis work
An AI agent sets up the case, the GPU solves it, and the result can be checked bit-for-bit on CPU.
AI agentic control plane
Describe a case in plain language and an AI agent writes the setup and runs it. The GUI, the agent, and the backend all work on the same case.
Native desktop GUI
Boundary conditions, run control, and post-processing in a native client. Large meshes stay responsive.
CPU–GPU bit consistency
Bit-level consistency between CPU and GPU has been a design constraint from the start. You can reproduce any case on CPU and compare.
GPU-native, CPU-capable
Every stage of the solve runs directly on the GPU. The same case also runs on a CPU cluster.
Precision you choose
Weigh accuracy against throughput and pick what the case needs — double precision (FP64), mixed precision, or single precision (FP32).
Production RANS turbulence
Spalart-Allmaras and SST, adapting automatically to industrial wall-function meshes and wall-resolved meshes alike.
Rotating domains
Propellers, rotors, and fans solved with a single rotating frame (SRF), frozen rotor, or sliding mesh.
MPI + multi-GPU scaling
Meshes are partitioned automatically and spread across GPUs. Verified on production-size cases.
Solver paths
DBNS today, FR and PBNS in development
DBNS is the primary path with published external-flow results. FR and PBNS are available for development and case-specific evaluation.
In production
FR
Flux Reconstruction
Flow regime
High-order, unsteady — compressible and incompressible (AC)
Primary use
High-order unsteady-flow studies; LES and aeroacoustics are development targets
PBNS
Pressure-Based Coupled Solver
Flow regime
Low-Mach and pressure-based flow, 3D only
Primary use
Internal-flow and low-speed aerodynamics studies
Published FR and PBNS examples establish only the conditions shown in those reports. Their development status applies to both desktop and command-line use.
From setup to results
A GPU CFD workflow with nothing in the way
01
Set up the case
Write it in JSONC by hand, build it in the desktop GUI, or describe it to an AI agent connected over MCP. Either way it is recorded in the same project model.
02
Run the solve
Run steady and unsteady analyses on GPU. Use several GPUs or a CPU cluster, and choose the precision that suits the case.
03
Review the results
Watch residuals and force monitors live, then post-process in the desktop client or in ParaView.
Applications
Primary application areas
Aerospace & defense
External aerodynamics of wings, airfoils, and vehicles from subsonic to supersonic. High-speed flow with shocks and separation.
Turbomachinery & rotating machinery
Compressor and turbine stages, propellers, fans, and rotors — solved with rotating frames and rotor–stator interfaces.
Unsteady flow & vortices
Vortex shedding, oscillatory loads, and wake interaction. Time histories come out of the dual-time unsteady framework.
Scale-resolving & aeroacoustics
FR is a development path for high-order unsteady flow, with LES and noise prediction as longer-term targets.
What changes in practice
This is how the waiting changes
Sooner, and for less. Hardware and power come down with it.
Dozens a day
Steady-state design iterations
Change the geometry and run it again — dozens of times over in a single day.
A month → a day
Coefficient tables · unsteady analyses
AoA and Mach sweep polars, and dual-time unsteady runs, land within the day.
50%
Lower cost per analysis
One GPU replaces several CPU nodes, cutting hardware, power, and floor space together.
5–33×
DBNS GPU speedup
One CPU chip against one GPU (chip-to-chip). The figure moves with problem size.
Verified numerics
See the results in the published reports
Comparisons against analytical solutions, reference codes, and published experimental data — conditions, numbers, and errors included as they are. Every report states the solver version and the environment it ran in.
Supersonic projectile aerodynamics
Mach 3 secant-ogive-cylinder, static & Magnus
Against BRL supersonic wind-tunnel data
Turbulent flat-plate boundary layer
3D SA-RANS, zero pressure gradient
Against NASA TMR CFL3D/FUN3D
RAE2822 airfoil
Transonic shock-induced separation
AGARD AR-138 experimental Cp
Cylinder vortex shedding (FR development)
Re = 100 von Kármán, unsteady FR case
Strouhal number vs. Williamson experiment
Supersonic wedge
M = 2.5, 15° oblique shock
Against the θ–β–M analytical solution
Forward-facing step
M = 3 supersonic shock system
Against Woodward–Colella
Get Wake CFD Community
Wake CFD is currently available as a technical preview. Request access to try the latest Community build and share your feedback.
Windows
The Community Edition supports Windows with 1 active GPU and meshes up to 5.12 million cells, and requires an NVIDIA CUDA GPU (Ampere / Ada). For Linux, multi-GPU, larger simulations, or team deployment, tell us about your setup and we will discuss the available options.
Technical specifications
Platform
- Linux · Windows
- Solver core and desktop GUI alike
GPU
- NVIDIA CUDA only
- Ampere · Ada architectures
Precision
- Double precision (FP64)
- Mixed precision
- Single precision (FP32)
Parallelism
- MPI domain partitioning (ParHIP / KaHIP)
- Multi-GPU execution
Solvers
- DBNS — density-based (in production)
- FR — flux reconstruction (in development)
- PBNS — pressure-based coupled (in development)
Turbulence models
- Spalart-Allmaras (SA)
- Menter SST
- Automatic wall-function blending
Frequently asked questions
Is Wake CFD open source?
No. The Community Edition is free, but it is not released under an OSI-approved open-source license — please check the licensing terms included with each build. There is no restriction on how you use the Community Edition, including commercially.
Which platforms are supported?
Both the solver core and the native desktop GUI run on Linux and Windows. A NVIDIA CUDA GPU (Ampere or Ada architecture) is required. The Community Edition technical preview is available for Windows; contact us to discuss Linux availability for your setup.
Do I need a GPU?
Wake CFD is designed GPU-native, but it also solves on CPU. The same case runs on a CPU cluster over MPI, and the results are comparable bit-for-bit with the GPU. Speed, however, is decisively in the GPU’s favour.
Can I drive it from an AI agent or an automation pipeline?
Yes. We ship an MCP (Model Context Protocol) server, so an AI agent, the desktop GUI, and the backend all share one project model. Wherever a case is set up, it is recorded in the same place and runs the same way.
Which turbulence models and interfaces are available?
Spalart-Allmaras and Menter SST RANS models with automatic wall-function blending. Rotating domains are handled with a single rotating frame (SRF), frozen rotor, or sliding-mesh interface.
What is currently available to try?
You can request access to the Community Edition technical preview for Windows. It supports 1 active GPU, 1 concurrent run, and meshes up to 5.12 million cells. If you need Linux, multi-GPU, larger simulations, or deployment within your organization, tell us about your requirements and we will discuss the available options.
Ready to put GPUs to work on your CFD?
The Community Edition is free, with no restriction on use. Take it and run it yourself.
Get in touch
Questions about Wake CFD? We are glad to hear from you.
Contact us