Mellanox (NVIDIA Mellanox) MQM8790-HS2F InfiniBand Switch | Product Launch and Key Feature Overview

Mellanox (NVIDIA Mellanox) MQM8790-HS2F InfiniBand Switch | Product Launch and Key Feature Overview

As HPC, AI training, and high-performance storage demand ever-increasing inter-node bandwidth, traditional Ethernet architectures are hitting hard limits on latency and congestion control. The Mellanox (NVIDIA Mellanox) MQM8790-HS2F, an InfiniBand switch built for the 200Gb/s HDR era, arrives precisely to address this generational shift. For network engineers and architects planning large-scale GPU clusters, the MQM8790-HS2F offers a smooth migration path from 100Gb/s EDR to 200Gb/s HDR.

Background: Why the MQM8790-HS2F Matters

In AI large-model training and scientific computing, AllReduce, parameter synchronization, and storage access are extremely sensitive to network latency. As clusters scale to hundreds or thousands of nodes, congestion and tail latency directly erode compute utilization. As a 40-port HDR InfiniBand switch with 200Gb/s per port, the NVIDIA Mellanox MQM8790-HS2F significantly reduces communication overhead. It tackles three typical problems in high-density environments—insufficient bandwidth, latency jitter, and limited scalability—making it a strong fit for teams with clear requirements for an MQM8790-HS2F InfiniBand switch.

Technical Highlights: MQM8790-HS2F Specifications and Key Features

The MQM8790-HS2F 200Gb/s HDR 40-port QSFP56 has a clear positioning: 40 QSFP56 ports in a 1U chassis with adaptive support for HDR, EDR, and FDR speeds. Its core features include:

• 40 × 200Gb/s HDR ports delivering up to 16Tb/s aggregate throughput, ideal for spine or aggregation layers in fat-tree or DragonFly+ topologies.
• SHARP (Scalable Hierarchical Aggregation and Reduction Protocol) in-network computing, offloading collective operations to the switch and reducing CPU/GPU involvement.
• Full compatibility with the NVIDIA Mellanox MQM8790-HS2F software ecosystem, including MLNX-OS, UFM management, and NetQ telemetry.
• For MQM8790-HS2F compatible deployments, it works with ConnectX-6 and ConnectX-7 adapters to form an end-to-end 200Gb/s solution.

Engineers reviewing the MQM8790-HS2F datasheet should focus on power, cooling, and port-split capabilities (such as 2×100Gb/s breakout) to plan rack power and cabling in advance. For MQM8790-HS2F price or MQM8790-HS2F for sale information, pricing typically depends on channel and project scale and is not covered here.

Deployment Scenarios and Selection Guidance

In practice, the NVIDIA Mellanox MQM8790-HS2F is commonly deployed at the aggregation or core layer of two-tier or three-tier fat-tree fabrics. For small-to-medium clusters, it can serve as the core switch connecting compute nodes directly; for very large clusters, multi-tier fabrics can interconnect tens of thousands of nodes. 

Conclusion

With 200Gb/s HDR, 40 QSFP56 ports, and SHARP in-network computing as its core strengths, the MQM8790-HS2F delivers a high-bandwidth, low-latency InfiniBand interconnect option for AI and HPC clusters. For network architects and IT managers planning next-generation compute fabrics, the Mellanox (NVIDIA Mellanox) MQM8790-HS2F deserves a spot on the shortlist, with MQM8790-HS2F specifications validated against actual topology requirements.