Meet Computational Fluid Dynamics FPGA Accelerator!
CFD SEMI One
Thanks to the high-order Discontinuous Galerkin method we can achieve 3rd-order of accuracy
8 x Xilinx Zynq – XCZU15EG with 4GB DDR4@2400MHz, 44 Computing Units clocked @480MHz in each FPGA
27 DOF in each element gives the effect of automatic mesh refinement three times in each direction
The DG method allows making high-order computations on unstructured hexahedral meshes.
The solution proposed by CFD SEMI is a new-generation hardware accelerator based on FPGAs and high-order discontinuous Galerkin method. Our technology is based on three pillars.
First, the Field-Programmable Gate Array which was configured for one task-quick solving N-S equations in the framework of the DG method. Each of its part was optimized to effectively perform the computations of given algorithm. We named it DG Solver-on-FPGA (DGSoF) because the solver is embedded inside that chip.
Second, the algorithm which solves Navier-Stokes equations in the framework of modal high-order Discontinuous Galerkin method. Each part of the algorithm was optimized to effectively perform computations in the DGSoF.
Third, hyper scalability, which is the ability to create systems with many DGSoFs with an almost linear speedup factor.
– 8x Xilinx Zynq – XCZU15EG
– 352 Specialized Computing Units (44 in each Zynq) @480MHz
– 32GB DDR4@2400MHz total FPGA RAM memory
– hexahedral unstructured meshes
– 800.000 hexahedral finite elements total capacity
– 800 000 hexahedral finite elements total capacity
– 21 600 000 DOF total capacity
– 600 000 elements (16 200 000 DOF) per second