Skip to main content
Advances in Aerodynamics
Fig. 4 | Advances in Aerodynamics

Fig. 4

From: Accelerating CFD simulation with high order finite difference method on curvilinear coordinates for modern GPU clusters

Fig. 4

Framework of communication of HiResX without GPU peer to peer communication technology support. For communication between processes that are all running on CPU, if processes are located in the same node, they exchange data within RAM with MPI-3 shared memory technique, see red path. If processes are located in different nodes, the standard MPI communication is utilized, see yellow path that connects triangle pair. For processes with GPU acceleration, data on GPU should be downloaded back to CPU, and then processes communicate in the same way as processes without acceleration

Back to article page