Comprehensive Product Overview

The Mellanox MQM8790-HS2F is a flagship Quantum-series HDR InfiniBand switch engineered by NVIDIA Mellanox to address the exponential bandwidth and latency demands of modern HPC, AI, and big data environments. As a 1U form-factor solution, it packs 40 non-blocking QSFP56 ports supporting 200Gb/s HDR InfiniBand, with the flexibility to split into 80 x 100Gb/s HDR100 ports via breakout cables.

Built on the Mellanox Quantum switch ASIC and driven by an x86 ComEx Broadwell dual-core CPU, the MQM8790-HS2F is an unmanaged (external management) model, ideal for environments leveraging centralized fabric management. It features port-to-cable (P2C/front-to-rear) airflow, redundant hot-swap 1100W AC power supplies (1+1), and hot-swap fans (N+1) for maximum uptime in mission-critical deployments. With a total switching capacity of 16Tb/s and forwarding rate of 7,200 Mpps, it delivers wire-speed performance across all ports while maintaining ultra-low latency (<130ns) for latency-sensitive applications.

Ideal for building scalable, high-density InfiniBand fabrics, the MQM8790-HS2F supports the SHARP v2 scalable hierarchical aggregation and reduction protocol, adaptive routing, and advanced congestion management—making it the backbone of large-scale HPC clusters, AI training infrastructure, and enterprise storage networks.

Key Features & Business Benefits

Key Features

• 40 x QSFP56 Ports: 200Gb/s HDR InfiniBand per port; splits to 80 x 100Gb/s HDR100 ports。
• 16Tb/s Non-Blocking Throughput: Wire-speed performance for all ports。
• Ultra-Low Latency: <130ns port-to-port latency (InfiniBand)。
• 1U Rack-Mount Form Factor: Standard depth (23.2”) for easy integration。
• P2C (Front-to-Rear) Airflow: Optimized for standard data center cooling layouts。
• Redundant Hot-Swap Components: 1+1 1100W AC power supplies; N+1 hot-swap fans。
• x86 Dual-Core CPU: ComEx Broadwell D-1508 (2.20GHz) with 8GB system memory。
• External Management: Unmanaged design for centralized fabric control。
• SHARP v2 Support: Accelerates collective operations for HPC/AI workloads。
• Adaptive Routing & Congestion Control: Optimizes fabric efficiency and reliability。

Business Benefits

• Accelerate HPC/AI Workloads: 200Gb/s bandwidth and <130ns latency slash job completion times for AI training, simulation, and modeling。
• Maximize Data Center Density: 40 ports in 1U reduces rack space and cabling complexity, lowering TCO。
• Ensure 99.999% Uptime: Redundant hot-swap power/fans minimize downtime。
• Simplify Fabric Scalability: Supports up to 2K nodes for seamless cluster expansion。
• Reduce Operational Costs: Energy-efficient design (80+ Platinum-certified power supplies) cuts power consumption。
• Future-Proof Investment: Backward compatible with EDR/FDR/QDR InfiniBand (100G/56G/40G) for legacy integration。

Technical Specifications

Ideal Application Scenarios

• High-Performance Computing (HPC) Clusters: Supercomputing, scientific simulation, computational fluid dynamics (CFD), and finite element analysis (FEA)。
• AI & Machine Learning (AI/ML) Infrastructure: Large-scale model training, deep learning, and GPU cluster networking (e.g., NVIDIA DGX)。
• Enterprise Data Centers (EDC): High-density virtualization, big data analytics, and distributed storage (NVMe-oF)。
• Cloud & Hyperscale Environments: Scalable InfiniBand fabrics for cloud HPC/AI services。
• Research & Academic Labs: University supercomputing centers and research clusters。

Supported Accessories & Modules

• QSFP56 Optics: 200Gb/s HDR InfiniBand transceivers (SR4, LR4, ER4)。
• Breakout Cables: 200Gb/s-to-2x100Gb/s HDR100 passive/active copper cables。
• Power Supplies: MTEF-PSF-AC-C (1100W AC, P2C airflow)。
• Fan Modules: MTEF-FANF-C (P2C airflow, hot-swap)。
• Rack Accessories: Rail kit (included), cable management arms, and 1U blanking panels。
• Management Tools: NVIDIA Mellanox OFED, centralized SM (Subnet Manager) software。

Frequently Asked Questions (FAQ)

Q1: What is the main difference between MQM8790-HS2F and MQM8700-HS2F?

A: The MQM8790-HS2F is an external management (unmanaged) model, while the MQM8700-HS2F includes on-board management (managed) with integrated SM。

Q2: Can the MQM8790-HS2F work with existing EDR InfiniBand hardware?

A: Yes, it is backward compatible with EDR (100G), FDR (56G), and QDR (40G) InfiniBand, supporting seamless integration with legacy fabrics。

Q3: What is the maximum cluster size supported with external SM?

A: When paired with a centralized Subnet Manager, it supports fabrics of up to 2048 nodes for large-scale HPC/AI deployments。

Q4: Does the switch support remote monitoring?

A: Yes, it supports SNMP for remote monitoring when managed via a centralized controller。

Q5: What is the typical power consumption of the MQM8790-HS2F?

A: Typical power consumption is ~540W, with a maximum of ~800W under full load.