True convex polyhedra with volume based contact, SPH fluids and melts, bonded Elements that crack, and CFD coupling, all native on the GPU. Models train on the live simulation through our CUDA bridge, and the solver physics self heals their predictions.
Crystal properties computed with machine learned potentials.
Grains, powders and ore as true polyhedra.
Bonded Elements that deform and crack.
SPH and CFD.
Single, mixed and double precision builds for Linux and Windows, on GPU architectures from Turing (RTX 20 series) to the newest.
BlazeViewer: DEM, SPH, CFD (OpenFOAM) and bonded Element results in one scene, on a desktop or headless on a server.
Native GPU multiphysics and Physical AI compared with commercial DEM and open source or CPU codes.
| Capability | BlazeMultiPhysics | Commercial Enterprise DEM | Open-Source / CPU DEM |
|---|---|---|---|
| 1. Hardware Efficiency GPU memory execution |
✔ High Single-GPU Efficiency Runs entirely on the GPU with direct in-memory execution. |
◐ Multi-GPU Available Multi-GPU solvers stall on CPU/PCIe transfers during multiphysics. |
✘ CPU Cluster Bound Severe memory bandwidth bottlenecks across distributed CPU nodes. |
| 2. Particle Shapes Geometry & breakage |
✔ True Polyhedra & Elements True convex polyhedra & deformable Element meshes. |
◐ Clumps / Polyhedra Multi-sphere clumps or older volume formulations; limited native crack breakage. |
◐ Spheres & Bonds Primarily spherical particles and bonded sphere clumps used to approximate irregular shapes. |
| 3. Fluids & Melt Hydrodynamics & thermal |
✔ Native GPU SPH & Melt Native GPU SPH fluids, slurry, viscoelastic melt, and thermal phase change. |
◐ External CFD Add-Ons Requires paid external CFD add-ons and expensive co-simulation licences. |
◐ Custom User Routines Requires manual patching of separate CFD codes or custom user routines. |
| 4. CFD Co-Simulation Mesh mapping & drag |
✔ Direct GPU Mesh Mapping Native GPU unstructured mesh mapping and compiled kernels (OpenFOAM); two-way CFD-DEM coupling with Kratos through Python. |
✔ Turnkey CFD Suites Established commercial CFD suites, but couplings stall on CPU cell search. |
✔ CFDEMcoupling Established open-source CFD-DEM coupling on CPU; memory-bound scaling. |
| 5. AI Integration Live tensors & GNN datasets |
✔ Direct VRAM Tensors CUDA bridge to the simulation tensors and pointers ( MLLive), solver computed WL3 hypergraph metrics & HDF5 graphs (MLTrainGNN). |
◐ Python / Sockets Python scripting and socket interfaces; cannot share live GPU arrays mid-step. |
◐ File Dumps & Hooks File-based I/O or custom C++ memory hooks; high latency for live RL loops. |
| 6. Ecosystem & Support Ecosystem maturity |
◐ Specialised GPU Architecture Agile specialist solvers and Python bridge; focused multiphysics workflow. |
✔ Broad Global Ecosystem Extensive CAD interfaces, mature multi-physics suites, and global support. |
✔ Fully Inspectable Open Source Zero licence cost, community-driven code, and fully open academic research base. |
Over ten years of peer-reviewed GPU DEM development: BLAZE-DEM, to our knowledge the first published GPU DEM framework for convex polyhedral particles (2014); volume based polyhedral contact (2015); the open source Blaze-DEMGPU release (2016).
