Engineering Architecture Brief
NVIDIA Quantum InfiniBand Switches deliver industry-leading networking performance for AI, HPC, and cloud-scale infrastructures. Featuring ultra-low latency, adaptive routing, congestion control, SHARP in-network computing acceleration, and intelligent fabric management, the Quantum family enables massive GPU clusters to communicate with exceptional efficiency. Available in 400Gb/s and 800Gb/s generations, these switches are designed to scale AI factories and supercomputing environments while maximizing throughput and minimizing infrastructure complexity.
NV-QM-IB-800
NVIDIA Quantum InfiniBand Switches
Ultra-low latency InfiniBand switch systems engineered for AI factories, supercomputing, and hyperscale high-performance computing infrastructures.
System Metrics Table
Adaptive Routing
Hardware Accelerated Adaptive Routing
Architecture
NVIDIA Quantum InfiniBand
Congestion Control
Advanced Congestion Control and QoS
Deployment
AI Factories, HPC Clusters, Supercomputers, Cloud Infrastructure
Fabric Management
NVIDIA Unified Fabric Manager (UFM)
Form Factor
High-Density 1U Data Center Switch Platform
In-Network Computing
NVIDIA SHARP™ Acceleration
Latency
Ultra-Low Deterministic Latency
Maximum Port Speed
Up to 800Gb/s
Network Intelligence
Real-Time Telemetry and Performance Analytics
Network Technology
InfiniBand
Operating System
NVIDIA NV-OS
Platform Type
InfiniBand Switch System
Port Density
Up to 144 x 800Gb/s Ports
Power Efficiency
Optimized High-Density Low-Power Architecture
Primary Workloads
Large Language Model Training, Generative AI, HPC, Scientific Computing
RDMA Support
Native Remote Direct Memory Access (RDMA)
Scalability
Designed to Scale to Millions of GPUs
Self-Healing Network
Automatic Fault Recovery
Switch Throughput
Up to 115.2 Tb/s