NVIDIA Q3400-RA 800Gb/s InfiniBand Switch 4U 144 XDR Ports over 72 OSFP Cages

Product Description

The NVIDIA Quantum-3 family of fixed-configuration switches revolutionizes the performance, scalability, and efficiency of high-performance computing and AI infrastructures, enabling faster and more effective AI processing and computation. These switches are available in both 4U and 2U systems.

The NVIDIA Quantum-X800 Q3400-RA/Q3401-RD 4U switches, the first to leverage 200Gb/s-per-lane serializer/deserializer (SerDes) technology, significantly enhance network performance and bandwidth. They feature 144 ports at 800Gb/s distributed across 72 octal small form-factor pluggable (OSFP) cages. The switches’ high radix supports a two-level, fat-tree topology capable of connecting up to 10,368 network interface cards (NICs) with minimal latency and optimal job locality, as well as other topologies providing connectivity to tens of thousands of GPUs.

The Q3400-RA is an air-cooled system designed for standard 19-inch rack cabinets, while the Q3401-RD offers the same air-cooled design with support for rack cabinets utilizing DC power. For smaller-scale platforms or integration with existing infrastructures, the NVIDIA Quantum-X800 Q3200 2U air-cooled configuration switch is ideal. This system houses two independent switches within a single enclosure, each providing 36 ports at 800Gb/s. The Q3200 fixed-configuration switches are well-suited for connecting new compute clusters to previous-generation Quantum and Quantum-2 InfiniBand storage infrastructure.

All NVIDIA Quantum-X800 switches include a dedicated OSFP InfiniBand in-band management port specifically for NVIDIA UFM® (Unified Fabric Manager), separated on the front panel from the other ports. This separation allows the full set of standard ports to be used for data network connectivity, simplifying port allocation and streamlining topology design. Additionally, NVIDIA Quantum-X800 switches feature optional router capabilities, facilitating the expansion of InfiniBand clusters to support a large scale of nodes located across multiple sites.