Infrastructure · Infrastructure resource
GPU Switch Fabric
Infrastructure resourceInfrastructureInfrastructurearc:GPUSwitchFabric
A hardware switch fabric giving every GPU in a node dedicated all-to-all paths at full link bandwidth without multi-hop contention.
Responsibility. Provides contention-free all-to-all GPU connectivity.
Also known as: NVSwitch
Variant of GPU Interconnect abstract
When to choose. Choose for 8+ GPU nodes and high-batch tensor-parallel inference where all-reduce bandwidth dominates.
Relationships
alternative to variability
Quantitative guidance
As stated by the sources; verify before use.
- DGX H100 uses 4 NVSwitch chips for 8-way all-to-all at 900 GB/s; 1.4-1.5x over P2P NVLink and 2.1x over PCIe; negligible 1.05x at batch 1 (Ch7.1A).
Classification
- Technologies
- NVIDIA NVSwitchDGX H100
Sources
- Ch7.1A: T. Nguyen, "Advanced Implementation with Nvidia NEMO Framework and Nvlink," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 7.1A. ISBN: 9798244538229.