NVIDIA Mellanox MQM8790-HS2F InfiniBand Switch Launched – Key Features & Technical Insights

As AI clusters and high-performance computing (HPC) environments continue to scale, the demand for ultra-low latency, high-throughput switching infrastructure has never been more critical. Addressing this need, NVIDIA has officially positioned the NVIDIA Mellanox MQM8790-HS2F as a key building block for next‑generation data centers. This 40‑port InfiniBand switch delivers 200Gb/s per port using HDR technology, bridging the gap between GPU‑driven workloads and network bottlenecks.

Why the MQM8790-HS2F Matters Today

Traditional Ethernet fabrics often struggle with the incast patterns and latency sensitivity of distributed AI training. The MQM8790-HS2F InfiniBand switch solves this by providing fully‑dedicated bandwidth, adaptive routing, and congestion control. For architects designing large‑scale supercomputers or enterprise AI clouds, the MQM8790-HS2F 200Gb/s HDR 40-port QSFP56 form factor enables a non‑blocking architecture—ensuring that every GPU node communicates at wire speed without packet drops.

Technical Highlights & Specifications

• Port density & throughput: 40 QSFP56 ports supporting up to 200Gb/s HDR per direction; aggregate switching capacity of 16 Tb/s non‑blocking.
• Low latency: Sub‑100ns switching delay, critical for collective operations like all‑reduce.
• Advanced features: SHARP (Scalable Hierarchical Aggregation and Reduction Protocol) for in‑network computing, adaptive routing, and enhanced congestion management.
• Compatibility: Full backwards compatibility with HDR100, EDR, and FDR links. Engineers can refer to the MQM8790-HS2F datasheet and MQM8790-HS2F specifications for detailed cable and transceiver support.

Key Deployment Scenarios

For large AI clusters using NVIDIA GPUs, the MQM8790-HS2F InfiniBand switch solution integrates seamlessly with the latest Mellanox network adapters. Typical use cases include:

• Training large language models (LLMs) requiring thousands of GPU nodes.
• HPGe simulations and earth‑science modeling with massive MPI workloads.
• Hyperconverged HPC storage backends leveraging NVMe over Fabrics (NVMe‑oF).

When evaluating procurement, IT managers often search for MQM8790-HS2F price and MQM8790-HS2F for sale options through authorized distributors. Meanwhile, engineers need to confirm MQM8790-HS2F compatible optics and cables — a wide ecosystem of QSFP56 modules is certified by NVIDIA. For those migrating from previous generations, the MQM8790-HS2F offers a smooth upgrade path to HDR while protecting existing cabling investments (e.g., using splitter cables for lower speeds).

Availability and Summary

With the rapid adoption of 200Gb/s HDR interconnects, the NVIDIA Mellanox MQM8790-HS2F stands out as a balanced choice between port density, power efficiency (~330W typical), and feature richness. Official MQM8790-HS2F datasheet and reference designs are now available through NVIDIA’s partner portal. For network architects and IT leaders, deploying the MQM8790-HS2F InfiniBand switch means future‑proofing AI infrastructure with deterministic performance and in‑network compute acceleration.