Skip to main content
Advances in Aerodynamics
Fig. 6 | Advances in Aerodynamics

Fig. 6

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

Fig. 6

Several typical PCIe root architectures of GPU server. For type (a), each NUMA node has a PCIe switch attached on them, and there are two GPUs mounted on each PCIe switch. For type (b), only one PCIe switch is mounted on each NUMA node, and there are 4 GPUs attached on one PCIe switch. For type (c), all GPUs are mounted on one NUMA node by two PCIe switches. For type (d), all GPUs are mounted on one NUMA node

Back to article page