 
|
NVIDIA MSB7800-ES2F Quantum-2 InfiniBand Switch 64-Port 400G NDR Managed Switch
Product Details:
| Brand Name: | Mellanox |
| Model Number: | MSB7800-ES2F |
| Document: | MSB7800-ES2F.pdf |
Payment & Shipping Terms:
| Minimum Order Quantity: | 1pcs |
|---|---|
| Price: | Negotiate |
| Packaging Details: | outer box |
| Delivery Time: | Based on inventory |
| Payment Terms: | T/T |
| Supply Ability: | Supply by project/batch |
|
Detail Information |
|||
| Application: | Server | Condition: | New And Original |
|---|---|---|---|
| Switching Capacity: | 2.6 Tbps | Uplink Connectivity: | 200 Gbps |
| Name: | MSB7800-ES2F 36 Port Mellanox Ethernet Switches | Keyword: | Mellanox Ethernet Switches |
| Highlight: | NVIDIA Quantum-2 InfiniBand Switch,400G NDR Managed Switch,64-port InfiniBand network switch |
||
Product Description
NVIDIA MSB7800-ES2F Quantum-2 InfiniBand Switch
64-Port 400G NDR Managed Switch with 51.2Tb/s Aggregate Bandwidth
Product Overview
The NVIDIA MSB7800-ES2F is a 64-port 400G NDR InfiniBand managed switch based on the NVIDIA Quantum-2 platform. It delivers 51.2Tb/s of aggregate bandwidth with sub-100 nanosecond latency, making it ideal for the most demanding HPC, AI, and enterprise workloads. The switch supports advanced features including SHARPv3 for in-network computing, NVIDIA's unique congestion management, and comprehensive telemetry.
Key Features
- Ports: 64 ports of 400G NDR (32 OSFP cages with twin-port)
- Aggregate bandwidth: 51.2Tb/s
- Latency: Sub-100ns
- Form factor: 2U rack-mount
- Power efficiency: Optimized for data center deployment
- In-network computing: SHARPv3 collective acceleration
- Congestion management: Advanced flow control and telemetry
- Management: Unified Fabric Manager (UFM) support
- Cooling: Air-cooled with finned-top transceivers
- Compliance: RoHS, CE, FCC, and more
Technology & Design
The Quantum-2 switch ASIC integrates 64 SerDes lanes running at 100G PAM4, enabling 64 ports of 400G NDR connectivity. The switch supports twin-port OSFP transceivers and DACs, allowing each OSFP cage to deliver 2*400G. NVIDIA's SHARPv3 technology offloads collective communication operations from the CPU to the switch, dramatically accelerating MPI workloads.
Typical Deployments
- AI training clusters: High-bandwidth, low-latency interconnect for GPU-based AI training
- HPC supercomputing: Large-scale MPI clusters requiring extreme performance and scalability
- Data center spine-leaf: Fat-tree and dragonfly topologies for enterprise data centers
- DGX SuperPOD: Integrated with NVIDIA DGX systems for turnkey AI infrastructure
- Cloud and enterprise: High-performance storage and compute interconnect
Specifications
| Product Type | InfiniBand Managed Switch |
| Platform | NVIDIA Quantum-2 |
| Ports | 64 ports of 400G NDR (32 OSFP cages, twin-port) |
| Aggregate Bandwidth | 51.2 Tb/s |
| Latency | Sub-100ns |
| Form Factor | 2U rack-mount |
| Power Supply | Dual redundant, hot-swappable |
| Cooling | Air-cooled (front-to-back or back-to-front) |
| In-Network Computing | SHARPv3 (Scalable Hierarchical Aggregation Protocol) |
| Model Number | MSB7800-ES2F |
Specifications are based on NVIDIA Quantum-2 platform. For complete details, please contact our sales team.
Advantages
- Unmatched performance: 51.2Tb/s aggregate bandwidth with sub-100ns latency
- In-network acceleration: SHARPv3 reduces MPI collective communication time by offloading to the switch
- High density: 64 ports of 400G in a 2U form factor – ideal for space-constrained data centers
- Intelligent management: UFM provides comprehensive visibility, automation, and telemetry
- Scalable: Supports fat-tree, dragonfly+, and custom topologies for any cluster size
- Reliable: Redundant power supplies and hot-swappable fans for high availability
Compatibility
| Transceivers | MMA4Z00-NS (800G twin-port OSFP SR8), MCP4Y10-Nxxx (800G twin-port OSFP DAC) |
| Adapters | NVIDIA ConnectX-7 and ConnectX-6 InfiniBand adapters |
| DPUs | NVIDIA BlueField-3 InfiniBand DPUs |
| Management | NVIDIA Unified Fabric Manager (UFM) and UFM Telemetry Platform |
| Topology | Supports fat-tree, dragonfly+, and custom topologies |
Frequently Asked Questions
What is the maximum number of switches supported in a fat-tree topology?
The Quantum-2 switch supports up to thousands of nodes in a fat-tree or dragonfly+ topology, scaling to 100,000+ nodes.
What is SHARPv3?
SHARPv3 (Scalable Hierarchical Aggregation Protocol) is NVIDIA's in-network computing technology that offloads collective operations from the CPU to the switch, dramatically accelerating MPI performance.
Can this switch be used with Ethernet?
This is an InfiniBand switch. NVIDIA Spectrum-4 switches are available for Ethernet deployments.
Handling Precautions
- Ensure proper airflow and cooling: front-to-back or back-to-front depending on model
- Install only in a rack with adequate ventilation and cooling
- Use only NVIDIA-qualified transceivers and cables
- Observe ESD precautions during installation and handling
- Ensure power supplies are connected to separate circuits for redundancy
About Hong Kong Starsurge Group
Hong Kong Starsurge Group Co., Limited is a technology-driven provider of network hardware, IT services, and system integration solutions. Since 2008, we have served customers worldwide with networking equipment including switches, NICs, wireless access points, controllers, cables, and optical modules. Our experienced teams support diverse industries such as government, healthcare, manufacturing, education, finance, and enterprise.
Buyer Checklist
- Confirm your cluster size and required number of switch ports
- Select appropriate transceivers or cables (optical vs. copper) based on reach
- Determine cooling direction: front-to-back or back-to-front
- Plan power requirements (dual redundant supplies recommended)
- Evaluate UFM licensing requirements for advanced management features
- Consider SHARPv3 benefits for MPI-intensive workloads
Want to Know more details about this product