PortLBM delivers cross-GPU fluid simulations with SYCL portability
New tool optimizes fluid flow simulations across AMD, NVIDIA, and Intel GPUs
PortLBM is a portable lattice Boltzmann framework built on SYCL that supports cross-platform GPU computing and real-time visualization. It handles simulations such as Kármán vortex streets, wing flows, and porous media. A performance study on NVIDIA, AMD, and Intel GPUs found no single configuration works best across all vendors: the stream layout maximizes throughput on the tested NVIDIA and Intel GPUs, while the bundle layout excels on the AMD GPU. Two-lattice schemes achieve higher throughput, whereas one-lattice schemes are preferable under memory constraints.
- PortLBM uses SYCL to run on AMD, NVIDIA, and Intel GPUs without vendor-specific code changes.
- Stream data layout is optimal for NVIDIA/Intel GPUs; bundle layout outperforms on AMD via better cache usage.
- Two-lattice schemes maximize throughput; one-lattice schemes are preferred when GPU memory is limited.
Why It Matters
Enables portable, high-performance fluid simulations across heterogeneous GPU hardware, simplifying deployment for HPC and engineering workloads.