NVIDIA Mellanox® MQM8790-HS2F Quantum™ HDR InfiniBand Switch: This is a cost-optimized, unmanaged 40-port HDR (High Data Rate) switch designed for high-performance scale-out fabrics. Built on the Quantum™ ASIC, it delivers the same 16Tb/s aggregate throughput and ultra-low latency as its managed counterparts but omits the onboard x86 management CPU. It functions as a high-speed "Leaf" or edge switch, relying on an external Subnet Manager (running on a Host or a Managed Switch) to orchestrate traffic. It supports full 200Gb/s bandwidth per port via QSFP56 interfaces.
🔧 Technical Specifications
OEM: NVIDIA / Mellanox
Model: MQM8790-HS2F (Unmanaged)
P/N: 920-9B110-00FH-0D0
Throughput: 200Gb/s per port (HDR)
Ports: 40x QSFP56
Airflow: P2C (Power-to-Connector / Blue Latch / Rear Intake)
Power Supply: Dual Redundant Hot-Swap AC
🌐 Engineering Context: The MQM8790-HS2F is the "Unmanaged" variant of the Quantum HDR series, making it the ideal choice for "Leaf" switch deployments where cost-efficiency and reduced power consumption are priorities. It requires an external Subnet Manager (OpenSM) to function. The P2C airflow design routes cooling air from the power supply side (Rear) to the connector side (Front), specifically engineered for rear-cooled racks or hot-aisle containment systems often found in dense GPU clusters.
🧠 Typical Use Cases
Leaf Switch in Fat-Tree topologies (connected to Managed Spine switches).
Expansion of existing HDR InfiniBand clusters managed by a central node.
Cost-sensitive AI Training Clusters (using Host-based OpenSM).
NVMe-oF Storage expansion racks requiring 200Gb/s throughput.
 
✅ Compatible With: NVIDIA ConnectX-6 and ConnectX-7 VPI/HDR Adapter Cards. Compatible with any HDR/HDR100 cabling (DAC/AOC). Requires an active Subnet Manager running elsewhere in the fabric (e.g., on a Managed Switch or a Server Node).
❌ NOT Compatible With: Standalone deployments without a Subnet Manager (the switch will not pass traffic). Standard Ethernet environments (unless configured as a VPI endpoint, though this is rare for unmanaged IB switches). Environments requiring C2P (Front-to-Back) airflow.
 
⚠️ Unmanaged Device: This unit does not have a management port (RJ45/Console) for OS configuration in the traditional sense. All fabric management is handled externally. It acts purely as a high-speed forwarding plane.
⚠️ Airflow Direction (P2C): This unit is "Power-to-Connector" (Rear Intake). Ensure your rack cooling design pushes cold air into the rear of the switch (PSU side) and accepts hot exhaust at the front (Port side). Mixing airflow directions in a single rack can cause thermal shutdown.
 
🤝 Why Choose T.E.S IT-SOLUTIONS? We specialize in enterprise-grade NVIDIA Mellanox networking hardware. Every MQM8790 unit is strictly validated for port throughput and ASIC health in our engineering lab. We offer expert consultation (including Subnet Manager configuration advice), fast worldwide shipping, and a comprehensive warranty to ensure your data center infrastructure runs without compromise.