NVIDIA Mellanox MQM8790-HS2F InfiniBand Switch | Product Launch & Key Feature Highlights

NVIDIA Mellanox MQM8790-HS2F InfiniBand Switch | Product Launch & Key Feature Highlights

As artificial intelligence (AI) and high-performance computing (HPC) workloads continue to scale, the demand for high-bandwidth, low-latency interconnect fabrics has never been greater. The introduction of the NVIDIA Mellanox MQM8790-HS2F marks a significant advancement in InfiniBand switching technology, delivering a 40-port HDR InfiniBand switch that provides 200Gb/s per port in a compact, energy-efficient 1U form factor. This MQM8790-HS2F InfiniBand switch is designed to power the most demanding AI, HPC, and enterprise data center environments, offering exceptional performance, scalability, and operational simplicity.

Why the MQM8790-HS2F Matters Now

Because AI training clusters and HPC systems are increasingly deploying thousands of GPUs that require massive parallel communication, the underlying InfiniBand fabric must deliver both high bandwidth and ultra-low latency at scale. The NVIDIA Mellanox MQM8790-HS2F addresses this need by providing 40 ports of 200Gb/s HDR InfiniBand in a 1U chassis, delivering a total switching capacity of 8Tb/s in a compact form factor. This density enables network architects to build large-scale, non-blocking fabrics with fewer switches, reducing both capital expenditure and operational complexity.

Because the MQM8790-HS2F 200Gb/s HDR 40-port QSFP56 switch supports the latest HDR InfiniBand technology, it delivers industry-leading performance with sub-100 nanosecond latency and supports advanced features such as congestion control, adaptive routing, and SHARP (Scalable Hierarchical Aggregation and Reduction Protocol) for in-network compute acceleration. These capabilities are essential for modern AI workloads, where all-to-all communication patterns and collective operations can significantly impact overall training time.

Because the switch is part of the NVIDIA Mellanox Quantum family of InfiniBand switches, the NVIDIA Mellanox MQM8790-HS2F is fully MQM8790-HS2F compatible with the broader NVIDIA InfiniBand ecosystem, including ConnectX and BlueField adapters, and can be seamlessly integrated into existing Quantum-based fabrics. This compatibility ensures that organizations can scale their InfiniBand networks incrementally, adding the MQM8790-HS2F as leaf switches in large-spine topologies or as standalone switches in smaller HPC clusters.

Key Technical Specifications

Because performance, density, and energy efficiency are critical factors in InfiniBand switch selection, the MQM8790-HS2F delivers a compelling combination of specifications that network planners should reference during architecture design:

These parameters are comprehensively documented in the MQM8790-HS2F datasheet, which includes detailed performance graphs, thermal specifications, and mechanical drawings for integration into rack layout tools.

Deployment Flexibility and Ecosystem Integration

Because the MQM8790-HS2F InfiniBand switch solution is designed for maximum deployment flexibility, it can be used in a variety of topologies. In small to medium-sized HPC clusters, the switch can serve as a standalone fabric core, providing 40 ports of 200Gb/s connectivity to compute and storage nodes. In larger deployments, the MQM8790-HS2F functions as a leaf switch in a spine‑leaf architecture, connected to higher‑density spine switches via its 40 QSFP56 ports to build fabrics with hundreds or thousands of nodes.

The switch's 40‑port configuration is particularly well‑suited for GPU‑accelerated environments. Because each 200Gb/s port can be broken out into two 100Gb/s connections (HDR100) using standard QSFP56 to 2×QSFP56 breakout cables, the MQM8790-HS2F can support up to 80 100Gb/s endpoints. This MQM8790-HS2F 200Gb/s HDR 40-port QSFP56 configuration provides exceptional flexibility for connecting GPU servers that may have multiple 100Gb/s or 200Gb/s interfaces.

Because the switch supports advanced InfiniBand features including adaptive routing, congestion control, and SHARP in‑network compute, network architects can optimize fabric performance for specific workload patterns. Adaptive routing dynamically distributes traffic across available paths to avoid congestion, while SHARP offloads collective communication operations (such as all‑reduce and broadcast) from the CPU/GPU to the switch, significantly accelerating AI training and HPC applications.

Availability and Ordering Information

Because demand for high‑performance InfiniBand switches continues to grow alongside AI workload expansion, the MQM8790-HS2F for sale is now available through authorized NVIDIA Mellanox distributors globally. Volume shipments commenced in mid‑2026, with standard lead times of 2–4 weeks for bulk orders and expedited options for critical infrastructure projects. For compatibility validation and technical specification details, the MQM8790-HS2F specifications are published on the official NVIDIA product portal, alongside application notes covering fabric design, cabling recommendations, and subnet management best practices.

Organizations considering the MQM8790-HS2F price relative to alternative InfiniBand switching solutions should factor in the switch's industry‑leading port density, energy efficiency, and advanced feature set — all of which contribute to a lower total cost of ownership over the deployment lifecycle. For detailed cost modeling assistance and volume pricing, contact your regional NVIDIA Mellanox sales representative or authorized distribution partner.

Summary

Because the NVIDIA Mellanox MQM8790-HS2F combines 40 ports of 200Gb/s HDR InfiniBand, sub‑100 nanosecond latency, and advanced in‑network compute capabilities in a compact 1U form factor, it directly addresses the performance, density, and scalability challenges that confront modern AI and HPC data centers. Network engineers gain the ability to build high‑performance, non‑blocking fabrics with fewer switches. Architects benefit from the switch's deployment flexibility, supporting both standalone and spine‑leaf topologies. And IT managers appreciate the predictable operational model, because the unified SKU reduces spare‑part complexity and simplifies fabric management through industry‑standard subnet management interfaces.

For more detailed performance graphs, thermal simulation data, and environmental qualification reports, refer to the official product documentation.