Illustration of a B300 AI cluster connected through an InfiniBand XDR leaf-spine fabric

InfiniBand XDR Fabric for B300 AI Clusters

Build a high-bandwidth, low-latency, and scalable network for B300 AI clusters with InfiniBand XDR.

Start Your B300 Fabric Design

Share your B300 cluster size, target topology, and deployment requirements.

XDR FABRIC OVERVIEW

InfiniBand XDR for B300 AI Clusters

B300 GPU compute nodes, XDR switching, and matched 800G interconnects form an InfiniBand fabric for distributed AI workloads.

AI data center racks supporting high-speed InfiniBand XDR networking
01

800G XDR Connectivity

High-throughput interconnects for intensive GPU-to-GPU traffic.

02

Adaptive Routing & Congestion Control

Balances traffic across available fabric paths and helps contain congestion hotspots.

03

SHARP In-Network Computing

Accelerates supported collective operations within the network, reducing communication overhead between GPUs.

B300 InfiniBand XDR Solution Architecture

A compute network architecture for B300 GPU clusters.

Network topology for a B300 GPU cluster using an InfiniBand XDR leaf-spine fabric

Cluster and Fabric Scale

B300 GPU Servers
1,296servers
800Gb/s NICs
10,368800Gb/s NICs
Leaf Switches
144switches
Spine Switches
72switches

Optical Interconnect

800G Optical Transceivers
10,368800G modules
1.6T Optical Transceivers
15,5521.6T modules
Optical Fiber Patch Cords
20,736patch cords

Related Products

Components for the B300 InfiniBand XDR fabric.

Active copper cable with 1.6T OSFP224 finned-top connectors on both ends for InfiniBand XDR

InfiniBand XDR ACC Cables

  • 1.6T OSFP224 Active Copper
  • 1–2.5m Custom Lengths
  • Power Consumption ≤ 0.1W
View XDR ACC Cables

WHY AICPLIGHT

AICPLIGHT Solution Support

Solution design, validated products, and dedicated project support.

Network architecture and rack layout planning for a customer deployment
01

Free Solution Design

A tailored solution architecture and matched product plan based on your requirements, site conditions, and deployment goals.

Optical transceivers and interconnects undergoing product validation testing
02

Tested Products for Deployment

Demo testing can be arranged for project validation when required. All optical transceivers and fiber patch cords undergo 100% inspection before shipment.

Technical project support for solution review, delivery, and ongoing maintenance
03

Single Point of Contact

A dedicated AICPLIGHT contact supports the project from requirement review through delivery and ongoing maintenance.

FAQ

01

What information is needed to design an XDR fabric?

To design the fabric, we need the number of servers, the installed InfiniBand adapter configuration, and the target oversubscription ratio. AICPLIGHT uses this information to determine Leaf and Spine port capacity, interconnect media, and cabling.

02

How are the Server–Leaf and Leaf–Spine links specified?

The media for Server–Leaf and Leaf–Spine links is selected according to the port speed, port form factor, and required reach at both ends of the connection. These requirements determine the appropriate optical transceivers and fiber patch cords for each link.

03

When should I use ACC, and when should I use optical transceivers with MTP/MPO patch cords?

ACC is suited to low-latency, short-reach connections within a rack or between adjacent racks. For longer reaches or fiber-based cabling, optical transceivers can be paired with MTP/MPO patch cords.

04

What does AICPLIGHT provide for an XDR fabric project?

After confirming the server configuration and network requirements, AICPLIGHT can provide a network topology, device and interconnect recommendations, link-media and cabling planning, and a project product list.

05

How are components tested and shipped before delivery?

Demo testing can be arranged for project validation when required. All optical transceivers and fiber patch cords undergo 100% inspection before shipment. NVIDIA InfiniBand switches and NICs are shipped in their original factory packaging with the anti-counterfeit seals intact.

CUSTOMER SUCCESS

Related AI Fabric Case Study

RELATED AI FABRIC CASE STUDY

Turnkey 128× B200 AI Platform for Fin-LLM Training

AICPLIGHT designed and delivered a 128-GPU NVIDIA B200 platform for a global financial services leader, including a Quantum-2 InfiniBand NDR computing fabric, optimized storage network, UFM management, and validated 800G/400G optical interconnects.

Read Case Study

START YOUR PROJECT

Start Your Network Solution Design

Share your application, existing environment, and deployment requirements. AICPLIGHT will help assess the architecture and recommend suitable products.

We typically respond within 4 business hours.