NVIDIA Quantum-3 InfiniBand

Brand: NVIDIA | Category: Network Switches

SKU: MQM9700-NS2F | Part #: MQM9700-NS2F | MPN: MQM9700-NS2F

Contact for Pricing — Request a Quote

NVIDIA's Quantum-3 InfiniBand fabric represents a deliberate engineering choice for data centre operators who need to move extremely large data sets at predictable, low-latency rates. The MQM9700-NS2F operates as a Layer 2–only switching platform, which means it makes forwarding decisions based purely on MAC addresses and does not perform IP routing internally. This architectural constraint is actually an advantage in purpose-built HPC and AI clusters where east–west traffic dominates and north–south routing to external networks is minimal or handled by dedicated border leaf switches. The absence of Layer 3 forwarding removes complexity from the fabric itself, allowing NVIDIA engineering to optimise every clock cycle for throughput and latency. Operators who plan to build large-scale model training environments, distributed storage arrays, or GPU clusters will find this design philosophy aligned with their infrastructure requirements.

At the heart of the MQM9700-NS2F lies 64 ports of NVIDIA's NDR (Next-Data-Rate) InfiniBand, each capable of 400 Gbps point-to-point, delivering a non-blocking fabric throughput of 3.2 terabits per second. In practical terms, a network engineer deploying this switch will see consistent, wire-rate performance across all ports simultaneously—no oversubscription, no hidden backplane limits. The 2U rack form factor allows dense deployment; a single MQM9700-NS2F can serve as a spine or leaf in fabric topologies without consuming excessive rack space. For data centre teams tasked with connecting dozens of GPU nodes or expanding a Lustre file system, the 64-port density makes this switch an economical building block. NVIDIA's decision to move to NDR InfiniBand reflects the bandwidth appetite of modern AI workloads; the jump from earlier HDR (200 Gbps) to NDR (400 Gbps) per port matters significantly when every GPU node is waiting for its training data.

Redundancy and operational reliability are built into the platform rather than bolted on. The MQM9700-NS2F ships with hot-swap redundant power supplies and redundant hot-swap cooling fans, meaning maintenance can happen without draining power from the entire switch or invoking a maintenance window. A network engineer or infrastructure operator can replace a fan or PSU during business hours without risk to running jobs. BMC (Baseboard Management Controller) with remote management capabilities allows out-of-band monitoring and alerting, so your team can track port health, temperature, and power consumption from a central management console. NVIDIA backs the hardware itself with a lifetime warranty, provided vendor-specific terms are met, which signals confidence in component quality and keeps replacement costs predictable over the switch's operational lifespan.

The control plane of the MQM9700-NS2F does not include native BGP or OSPF routing; those protocols are Layer 3 functions and sit outside the switch's scope. This design is intentional: the switch excels at what it does—moving packets through the fabric at the lowest possible latency—and delegates routing decisions to adjacent systems. If your deployment requires VXLAN or EVPN overlays (common in multi-tenant or hybrid-cloud scenarios), those would be implemented at the compute nodes or at dedicated border switches, not within the MQM9700-NS2F itself. This separation of concerns means the switch does not consume CPU cycles on routing lookups, keeping forwarding latency constant regardless of the size of your routing table. For storage architects and AI infrastructure teams building single-tenant, tightly coupled clusters, this constraint is not a limitation—it is a deliberate optimisation.

Omnixon Global is the established NVIDIA InfiniBand distributor across the UAE, GCC region, and into Asia. We maintain stock of the MQM9700-NS2F and the broader Quantum-3 family, backed by local technical support and regional logistics. If you are evaluating fabric architecture for a new data centre, HPC refresh, or AI lab build-out, we encourage you to request a formal quotation and technical consultation. Our team can assist with fabric design reviews, compatibility verification with your existing infrastructure, and procurement timelines aligned with your project schedule.

NDR InfiniBand Fabric Specifications

• 64 × 400 Gbps NDR InfiniBand ports with 3.2 Tbps aggregate throughput and non-blocking fabric architecture
• Layer 2 switching only; no native BGP, OSPF, or IP routing—optimised for low-latency east–west traffic in tightly coupled clusters
• 2U form factor with hot-swap redundant power supplies and redundant hot-swap cooling fans for maintenance without downtime
• Baseboard Management Controller with remote out-of-band monitoring, alerting, and port health telemetry
• Operating temperature range of 0–40°C with thermal and power monitoring via standard management interfaces
• Part number MQM9700-NS2F backed by NVIDIA lifetime hardware warranty under vendor terms