Skip to main content
Advances in Aerodynamics
Fig. 17 | Advances in Aerodynamics

Fig. 17

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

Fig. 17

Performance of HiResX running on different GPU topology. Cases in which all GPUs are in the same PCIe switch get best communication efficiency, see case I and II. For the case in which GPUs are located in different PCIe switches, if GPU devices are optimally assigned according to communication load, communication efficiency can also be improved, see case IV. Without GPU to GPU optimization, the communication efficiency is lower, see case III and V

Back to article page