11(1)
home Products Infiniband

MSB7800-ES2F

share to:

The MSB7800-ES2F is a high-performance EDR 100Gb/s InfiniBand 1U switch developed by NVIDIA (formerly Mellanox). Designed for demanding workloads in hyperscale computing and enterprise data centers, it delivers ultra-low latency and scalable bandwidth for AI/ML, HPC, and cloud-native environments.

Introduction

1、Network Performance

  • Port Configuration: 36 × QSFP28 ports supporting EDR 100Gb/s InfiniBand (backward compatible with FDR 56Gb/s and QDR 40Gb/s).
  • Throughput: Non-blocking 7.2Tb/s aggregate switching capacity.
  • Latency: Industry-leading 90ns port-to-port switching delay.

2、Hardware Design

  • Chassis: 1U rack-mountable form factor (43.6mm H × 428mm W × 685.8mm D).
  • Redundancy:
    • Power: 1+1 redundant hot-swappable power supplies (80+ Gold/Energy Star certified).
    • Cooling: N+1 redundant hot-swappable fans.
  • Power Efficiency: Typical consumption of 122W, optimized for energy-sensitive deployments.

3、Advanced Technologies

  • SHARP (Scalable Hierarchical Aggregation Protocol): Offloads collective communications (e.g., MPI operations) to the switch, reducing CPU overhead by up to 30%.
  • Adaptive Routing & Congestion Control: Dynamically optimizes traffic paths to prevent bottlenecks.
  • VL2VL Mapping: Enables quality-of-service (QoS) prioritization for critical workloads.

4、Management & OS

    • Management Interface: Embedded Intel Celeron Dual Core 1047UE processor with 4GB DDR3 RAM and 16GB SSD.
    • Operating System: Runs MLNX-OS for simplified configuration and monitoring.
    • Firmware: Supports in-service upgrades and compatibility with NVIDIA UFM platform.

5、Target Applications

  • High-Performance Computing (HPC): Accelerates distributed simulations and genomics research.
  • AI/ML Clusters: Optimizes GPU-to-GPU communication in large-scale training environments.
  • Enterprise Data Centers: Supports storage disaggregation and NVMe-oF (NVMe over Fabrics).
  • Cloud Infrastructure: Enables high-density virtualization and Kubernetes-native networking.