NVIDIA Mellanox MSB7890-ES2F 100G InfiniBand Switch 36-Port 7.2Tb/s Unmanaged

The NVIDIA Mellanox MSB7890‑ES2F is an unmanaged 100Gb/s InfiniBand smart switch designed for large‑scale data center deployments requiring centralized fabric management. Part of the NVIDIA SB7800 series, this switch delivers 36 QSFP28 ports operating at 100Gb/s per port, with an aggregate non‑blocking throughput of 7.2 Tb/s and ultra‑low port latency. Unlike its managed counterpart (MSB7800), the MSB7890‑ES2F is optimized for external management via NVIDIA Unified Fabric Manager (UFM™), enabling data center operators to efficiently provision, monitor, troubleshoot, and maintain modern data center fabrics at scale.

The MSB7890‑ES2F features the same high‑performance hardware as the managed SB7800 series, including an onboard dual‑core x86 CPU for switch hardware management (firmware, power supplies, fans) while relying on external Subnet Managers for fabric orchestration. With support for NVIDIA SHARP™ in‑network computing technology, the switch offloads collective operations from CPUs to the network fabric, dramatically accelerating MPI and deep learning workloads. The MSB7890‑ES2F variant features P2C (port‑to‑power) airflow, making it suitable for data center cooling architectures where cold air is supplied from the port side.

• 100Gb/s InfiniBand per port – 36 QSFP28 ports delivering full bidirectional bandwidth with non‑blocking architecture.
• 7.2 Tb/s Aggregate Throughput – High‑density switching capacity in a compact 1U form factor.
• Unmanaged SKU for External Control – Optimized for centralized management via NVIDIA UFM™ platform; ideal for large‑scale fabrics requiring unified orchestration.
• In‑Network Computing Acceleration – NVIDIA SHARP™ technology offloads collective operations from CPUs to the switch fabric, reducing MPI operation time and freeing valuable CPU resources.
• UFM Ready – Seamless integration with NVIDIA Unified Fabric Manager for advanced telemetry, predictive analytics, and automated fabric orchestration across hundreds or thousands of switches.
• Redundant & Hot‑Swappable Components – 1+1 redundant power supplies with 80 Plus Gold certification, plus hot‑swappable fan modules for maximum availability.
• Energy Efficient Design – ATIS weighted power consumption as low as 122W for a fully populated system; dynamic power scaling reduces consumption when ports are not fully utilized.
• Flexible Cooling Options – P2C (port‑to‑power) airflow configuration (MSB7890-ES2F) with front‑to‑rear cooling.
• Backward Compatible – Seamless interoperability with previous InfiniBand speeds (EDR, FDR, QDR).

High‑performance computing and AI workloads demand intelligent interconnect solutions to achieve maximum application performance. NVIDIA SHARP™ (Scalable Hierarchical Aggregation and Reduction Protocol) technology is embedded directly in the switch silicon, enabling in‑network computing that offloads collective operations from the CPU to the switch network. This innovative approach eliminates the need to send data multiple times between endpoints, dramatically reducing MPI operations time and decreasing the amount of data traversing the network.

By accelerating collective communications such as all‑reduce, barrier, and broadcast, SHARP™ technology delivers order‑of‑magnitude performance improvements for deep learning frameworks (NCCL, TensorFlow, PyTorch) and traditional HPC applications. Additional benefits include freeing up valuable CPU resources for computation rather than consuming them for communication processing, resulting in higher overall application throughput — all without requiring on‑board switch management.

• Large‑Scale AI & ML Clusters – GPU‑based systems requiring centralized fabric management and telemetry across hundreds or thousands of nodes with UFM orchestration.
• Hyperscale Data Centers – Fat‑tree, DragonFly+, and multi‑dimensional torus topologies managed through UFM for maximum efficiency and unified control.
• High‑Performance Computing (HPC) – Research labs, national labs, and universities where external fabric managers provide enhanced monitoring and automation capabilities.
• Enterprise & Cloud Service Providers – Environments requiring unified control plane across multiple switch fabrics with advanced telemetry and predictive analytics.
• Top‑of‑Rack (ToR) with Centralized Management – High‑density 100Gb/s leaf connectivity with fabric‑wide visibility and automated provisioning.

The MSB7890‑ES2F works seamlessly with NVIDIA ConnectX‑5, ConnectX‑6, and ConnectX‑7 InfiniBand adapters, as well as BlueField DPU families. It is backward compatible with previous InfiniBand speeds (EDR, FDR, QDR), ensuring investment protection and smooth migration paths. Fully interoperable with other NVIDIA SB7800 and Quantum series switches. For management, the switch is designed to be controlled via NVIDIA Unified Fabric Manager (UFM) platform, which provides comprehensive fabric provisioning, monitoring, and predictive troubleshooting, or can be managed by any standards‑compliant InfiniBand Subnet Manager (OpenSM, etc.). Operating system support includes all major Linux distributions (RHEL, Ubuntu, Rocky Linux, SuSE) and Windows Server with InfiniBand stack.

For large‑scale deployments requiring centralized fabric management via UFM, MSB7890‑ES2F (P2C airflow) is the ideal choice. For rear‑to‑front cooling designs, select MSB7890‑ES2R. Managed variants (MSB7800) are suitable for smaller clusters or standalone deployments requiring on‑board Subnet Manager.

• Centralized Fabric Management – Leverage NVIDIA UFM for unified visibility, automation, and predictive analytics across the entire fabric — ideal for large‑scale deployments.
• Superior ROI – High port density in 1U reduces switch count and cabling complexity, lowering capital expenditure.
• Energy Efficient – Industry‑leading power efficiency (122W typical) with ATIS weighted measurement; dynamic power scaling reduces operational costs.
• SHARP™ Acceleration – Up to 10x faster collective communications without consuming host CPU cycles, maximizing compute utilization.
• Reduced IT Complexity at Scale – Unified management plane across hundreds or thousands of switches simplifies operations and reduces administrative overhead.
• Enhanced Telemetry & Analytics – UFM integration provides real‑time network telemetry, predictive troubleshooting, and AI‑powered cyber intelligence.
• Investment Protection – Backward compatible with previous InfiniBand generations, protecting existing infrastructure investments.

Starsurge Group provides end‑to‑end lifecycle services for NVIDIA Mellanox switches, including pre‑sales architecture consulting, proof‑of‑concept testing, and global logistics. Our experienced technical team offers remote troubleshooting, firmware upgrades, and RMA coordination. For UFM deployments, we offer professional services for platform configuration, integration, and training. Warranty extension options and 24x7 priority support available upon request. Multilingual support for EMEA, Americas, and APAC regions ensures rapid response for mission‑critical deployments.

Hong Kong Starsurge Group Co., Limited is a technology‑driven provider of network hardware, IT services, and system integration solutions. Founded in 2008, the company serves customers worldwide with products including network switches, NICs, wireless access points, controllers, cabling, and infrastructure equipment. Backed by an experienced sales and technical team, Starsurge supports industries such as government, healthcare, manufacturing, education, finance, and enterprise.

With a customer‑first approach, Starsurge focuses on reliable quality, responsive service, and tailored solutions. As an authorized partner for leading networking brands, we deliver global logistics, custom software development, and multilingual support — helping clients build efficient, scalable, and dependable network infrastructure.

• NVIDIA Unified Fabric Manager (UFM) Platform Licenses
• NVIDIA Quantum QM8700 Series (200G InfiniBand Switches)
• NVIDIA ConnectX‑6 VPI Adapter Cards (100Gb/s Dual‑port)
• NVIDIA ConnectX‑7 Adapter Cards (400Gb/s InfiniBand)
• NVIDIA BlueField‑3 DPU for infrastructure acceleration
• Starsurge Custom Rack Integration Kits & QSFP28 Cables