The Mellanox MQM8790-HS2F is a high-performance 1U InfiniBand switch engineered for next-generation data center fabrics. It provides 40 non-blocking 200Gb/s QSFP56 ports with 16Tbps bidirectional switching capacity and ultra-low sub-130ns port-to-port latency. The switch integrates NVIDIA SHARP technology for in-network computing acceleration, enabling optimized collective operations for AI training and HPC simulations. Each port supports flexible configuration via breakout cables (2x100G per 200G port) and backward compatibility with EDR/FDR protocols. Dual hot-swappable AC power supplies with 80 PLUS Gold certification ensure high availability. The switch supports advanced routing topologies including Fat Tree, DragonFly+, and 6-Dimensional Torus, with intelligent adaptive routing and congestion control mechanisms. Management capabilities include SNMP, CLI, web interface, and RESTful APIs via NVIDIA UFM platform. Designed for hyperscale cloud deployments, financial services, research institutions, and enterprise HPC environments requiring maximum bandwidth density and minimal latency.
40x 200Gb/s HDR InfiniBand ports in 1U form factor.
Dual hot-swappable 80 PLUS Gold power supplies for high availability.
Supports flexible port splitting to 2x100G via breakout cables.
Expedited and overnight shipping options are available at checkout, priced by destination and service.
Purchase orders accepted from qualified institutions.
Refunds within 14 days and replacements within 30 days of delivery with an RMA (25% restocking fee on customer-initiated returns), and every part carries our Standard 30-Day Warranty - see our return policy or contact US IT Hardware.
Featuring
Manufacturer part #
MQM8790-HS2F
Brand
Mellanox
Condition
New
Model
QM8790 HDR InfiniBand Switch
Telemetry
In-band network telemetry for real-time performance monitoring
QoS Levels
16 hardware-based priority levels with traffic prioritization
Product Line
NVIDIA Quantum QM8700 Series
Product Type
1U Data Center InfiniBand Switch
Buffer Memory
Deep buffer architecture for burst traffic and congestion handling
Layer 2 Features
VLAN (802.1Q), LACP link aggregation, jumbo frames (9216 bytes)
Layer 3 Features
Compatible systems
The MQM8790-HS2F integrates seamlessly with enterprise and hyperscale data center infrastructure. It requires HDR-capable InfiniBand HCAs on compute nodes and QSFP56 optical transceivers for connectivity. Backward compatibility with EDR/FDR enables gradual fabric upgrades. Management operates via native MLNX-OS or NVIDIA UFM platform. Supports Linux, Windows Server, and native HPC operating systems on connected systems.
NVIDIA DGX SuperPOD
Model: DGX A100/H100
Form factor: Data Center Fabric
Generation: Current
Primary fabric for GPU cluster interconnect in NVIDIA's reference architectures.
HPE Apollo
Model: Apollo 6500 Gen10 Plus
Form factor: HPC Cluster
Generation: Gen10+
Compatible with HPE's high-performance computing systems via InfiniBand adapters.
Dell PowerEdge
Model: XE8640, XE9680
Form factor: 2U/4U Servers
Generation: 3rd Gen
Works with Dell PowerEdge servers equipped with Mellanox InfiniBand HCAs.
Lenovo ThinkSystem
Model: SR670 V3, SR860 V3
Form factor: 1U/2U Servers
Generation: V3
Compatible with Lenovo enterprise servers supporting InfiniBand connectivity.
Supermicro SYS-2U
Model: 2124US-TNRT, 2224BT-DNRT
Form factor: 2U/4U Systems
Generation: Current
Supports Supermicro HPC and AI systems with InfiniBand fabric integration.
Details
Shipping, warranty & returns
Customer reviews
No reviews yet — be the first to review this part.
Free ground shipping over $100
Secure SSL checkout. Visa, Mastercard, Amex and Discover accepted. ACH / wire available for businesses.
Buy with confidence
New, certified-refreshed and OEM stock with guaranteed compatibility, backed by warranty on every part.
Dual 740W AC power supplies, hot-swappable, 80 PLUS Gold certified
Power Consumption
224W typical operating power
Physical Characteristics
Weight
Approximately 26 pounds (11.8 kg)
Dimensions
1.72 inches height x 16.85 inches width x 27.63 inches depth (4.37cm x 42.8cm x 70.2cm)
Form Factor
1U rack-mountable, standard depth
Airflow Direction
Front-to-back (P2C) airflow
Included Accessories
Rail kit for standard 19-inch rack mounting
Environmental
Cooling
Hot-swappable redundant fan modules, front-to-back airflow
Operating Humidity
5% to 85% non-condensing
Operating Temperature
0°C to 45°C (32°F to 113°F)
Interface/Ports
Port Splitting
Each 200G port splits to 2x100G via breakout cables
Port Configuration
40x 200Gb/s QSFP56 HDR InfiniBand ports
Management Interfaces
1x RJ-45 Ethernet, 1x USB, serial console port
Optical Transceiver Support
200GBASE-SR4, CSR4, LX4, LR4, PLR4, PIR4, ER4
Best for
Artificial Intelligence & Machine Learning
Large-scale distributed AI training clusters requiring ultra-low latency collective operations, gradient aggregation, and parameter synchronization across hundreds of GPU nodes.
High-Performance Computing
Scientific research, molecular dynamics simulations, weather modeling, and computational fluid dynamics requiring massive bandwidth and minimal latency for inter-node communication.
Cloud Data Center Interconnect
Hyperscale cloud provider spine/core switching for east-west traffic, multi-rack aggregation, and inter-datacenter connectivity with non-blocking fabric architecture.
Storage Area Networks
High-speed SAN deployments, NVMe-oF fabrics, and data-intensive applications requiring consistent low-latency access to distributed storage systems.
Financial Services
High-frequency trading systems and quantitative analysis platforms demanding ultra-low latency and deterministic performance for real-time market data distribution.
Research & Academic Supercomputing
University computing centers and national labs requiring petascale interconnect fabrics for collaborative research, climate modeling, and physics simulations.
Frequently asked questions
What is the total switching capacity and port configuration of the MQM8790-HS2F?
The MQM8790-HS2F delivers 16Tbps non-blocking bidirectional switching capacity with 40x 200Gb/s QSFP56 ports. Each port supports full-duplex 200Gb/s throughput simultaneously, enabling 7.2 billion packets per second forwarding rate. Ports can be split into 2x100Gb/s configurations via breakout cables, providing up to 80 logical 100G ports for mixed-speed deployments.
How does NVIDIA SHARP technology improve performance in AI and HPC workloads?
NVIDIA SHARP enables in-network computing by offloading collective operations (AllReduce, AllGather, Broadcast) directly to the switch hardware. This reduces CPU overhead, minimizes latency, and accelerates gradient aggregation in distributed AI training. For HPC, SHARP optimizes MPI collective communication, reducing synchronization bottlenecks in large-scale simulations across thousands of compute nodes.
What is the port-to-port latency and how does it compare to other data center switches?
The MQM8790-HS2F achieves sub-130ns port-to-port latency, among the lowest in the industry. This deterministic, ultra-low latency is critical for AI training synchronization, HPC inter-node communication, and financial trading systems where microsecond delays impact performance. The non-blocking architecture ensures consistent latency across all ports regardless of traffic patterns.
What routing topologies and congestion control mechanisms does the switch support?
The MQM8790-HS2F supports advanced topologies including Fat Tree, DragonFly+, 6-Dimensional Torus, and SlimFly. It features adaptive routing to dynamically avoid congestion, intelligent flow control, and 16 hardware-based QoS priority levels. These mechanisms eliminate hot spots and maximize effective fabric bandwidth in large-scale deployments with thousands of endpoints.
Is the MQM8790-HS2F backward compatible with older InfiniBand generations?
Yes, the MQM8790-HS2F is fully backward compatible with EDR (100Gb/s), FDR (56Gb/s), and lower-speed InfiniBand protocols. This allows gradual fabric upgrades without replacing existing infrastructure. The switch also supports 200GBASE-SR4, CSR4, LX4, LR4, PLR4, PIR4, and ER4 optical transceivers for flexible connectivity options.
What management and monitoring capabilities are included?
Management is provided via SNMP, CLI, web-based interface, and RESTful APIs compatible with NVIDIA UFM (Unified Fabric Manager) platform. The switch includes in-band network telemetry for real-time performance monitoring, advanced diagnostics, and proactive health tracking. Serial console and USB ports enable local management access. SSH and SNMPv3 provide secure remote administration.
What are the power consumption and cooling requirements for the MQM8790-HS2F?
The switch consumes 224W typical operating power with dual 740W AC power supplies providing N+1 redundancy. Hot-swappable fan modules deliver front-to-back airflow for efficient heat dissipation. Operating temperature range is 0°C to 45°C with 5-85% non-condensing humidity tolerance. The 1U form factor (1.72 inches height, 26 pounds) fits standard 19-inch racks.
Cisco MerakiMS125-48LP-HW
Allied TelesisAT-8624T/2M
Allied TelesisAT-9000/24
Dell469-3414
Allied TelesisAT-GS910/24-50
Allied TelesisAT-FS717FC/SC-10
Allied TelesisAT-9424T/SP