Sign up now! New useSign up now! New users get $20 in free creditsDeepSeek V3.1

InfiniBand Networking

The Backbone of Next-Generation AI Supercomputing

InfiniBand Networking

InfiniBand Networking is used to support next-generation performance and scalability. We built our NVIDIA Quantum InfiniBand Cluster Networking with QM9700 and QM9790

Designed for high performance work

2x
Data Throughput
400Gb/s
4x
MPI Performance
New MPI All-to-All IN-Network Computing acceleration engine
3.2x
Switch System Ability
Aggregated bidirectional tHroughput of 51.2 Tbit/s with 64 400Gb/s ports
6.5x
Higher Scalability
Connecting>1M nodes with 3 hops
32x
More AI Acceleration
NVIDIA SHARP In-Networking Computing technology
Multi-GPU Instances

Accelerated In-Network Processing

InfiniBand leverages technologies like NVIDIA SHARP to offload collective communication tasks directly to the network switches. This reduces the volume of data moving across the network, shortens the duration of MPI operations, and boosts data center efficiency.

Multi-GPU Instances

Enhanced Service Quality

InfiniBand delivers superior quality of service through advanced congestion management and adaptive routing. These capabilities guarantee efficient, reliable data transmission and maintain optimal network performance, even under heavy workloads.

Private Cloud

Flexible Topology Support

InfiniBand provides centralized management and supports a wide variety of network topologies. Its flexible routing algorithms allow organizations to optimize network performance for specific application communication patterns.

The NVIDIA Quantum InfiniBand Platform

InfiniBand Adapters

The NVIDIA ConnectX-7 InfiniBand adapter supports PCIe Gen5, offers multiple form factors, and delivers single or dual network ports with speeds up to 400Gb/s

InfiniBand Adapter

Optical Transceivers and Optic Cables

Use OSFP transceivers and OSFP cables deliver high bandwidth, low latency, and reliable connectivity between InfiniBand components

Optical Transceiver
Optical Transceiver

MQM9700-NS2F

Low Latency

High Scalability

Features 32 OSFP ports that flexibly break out into 64x400GbE configurations, supporting large-scale clusters of up to 2,000 nodes

Computing

Ultra-Low Latency

Achieves a massive 25.6 Tb/s total throughput, accelerated by RDMA and NVIDIA SHARP In-Network Computing for ultra-low latency and peak application performance

QoS

Energy Efficiency

Advanced power-saving features to minimize energy usage and lower operating expenses

Tolerance

Robust Hardware Reliability

Ensures maximum uptime with a compact 1U design, redundant hot-swappable power and fans, and a self-healing network for up to 5000x faster failure recovery

Scalability

Difference between 9700 and 9790

9700 is managed: built-in NOS, subnet manager, management port. 9790 is unmanaged: no CPU/NOS, configured via mlxconfig

Actual use scenarios

High Performance

High-Performance Computing

Climate simulation,and Protein structure analysis and so on

HPC Climate Simulation
Workloads

AI Training and Interference

Support multi-node distributed training like GPT-4 and LLaMA

HPC Climate Simulation
Scalability

High-Frequency Trading

Low latency and DetNet meet financial high-frequency trading needs

HPC Climate Simulation
Advanced Congestion Control

Storage Network

Replace traditional Fibre Channel to achieve high-speed storage access

HPC Climate Simulation

Why NVIDIA Quantum InfiniBand on
Canopy Wave Networking Service?

NVIDIA GPUs

Robust Supply Chain & Equipment Sourcing

The integrated supply-chain governance and strategic vendor alliances deliver scalable, expedited procurement of critical network hardware-adapters, switches, transceivers, and cables-eliminating delays and complexity

Multi-GPU Instances

24/7 Support

Dedicated 24/7 expert coverage delivers instantly demand acknowledgement via real-time interactive channels, ensuring urgent issues are resolved within the same operational window without overnight delay

Private Cloud

Cost-effective

Delivers high performance at a cost-effective price, helping businesses reduce operational costs while maintaining optimal efficiency. Optimize hardware and network architecture to ensure a faster return on investment

Incredibly Fast Networking
Services For GenAI

Don’t let a shoddy network slow you down. Get the absolute most out of your GPUs

Contact us