Skip to main content
Fig. 16 | Advances in Aerodynamics

Fig. 16

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

Fig. 16

Cases configuration. In figure (a), there are 100 points in the spanwise direction. Data exchange in vertical direction (red) is larger than data exchange in horizontal direction (blue). In figure (b), five cases are set to test the communication performance of different strategies of computations assigned to different GPUs. The number in block presents the GPU ID in which computation is performed. Blocks with the same color indicate that they are in the same PCIe switch. In cases I and II, computations are performed on GPUs that are in two different PCIe switches respectively. In cases III to IV, computations are performed on GPUs belonging to both two PCIe switches. For case III, the maximum data transfers are performed across the PCIe switch. In case IV, the maximum data transfers are performed within the same PCIe switch. In case V, all data transfers are performed across the PCIe switch

Back to article page