NVIDIA Quantum-X800 InfiniBand Switch · Q3400 Series

NVIDIA Q3400-RA 800Gb/s InfiniBand Switch, 144 Ports, 72 OSFP, 4U Air-Cooled

The Q3400-RA is a 4U air-cooled 800Gb/s InfiniBand switch from the NVIDIA Quantum-X800 series. It provides 144 non-blocking 800Gb/s ports across 72 OSFP connectors, 115.2 Tb/s throughput, fourth-generation NVIDIA SHARP, telemetry-based congestion control and Quantum-X Photonics co-packaged optics, managed through NVIDIA NVOS and UFM.

B2B supply and project configuration support provided by Hong Kong Starsurge Group Co., Limited. Availability, lead time and pricing are confirmed per order.

Product Overview

NVIDIA Quantum-X800 InfiniBand switches deliver 800Gb/s throughput with ultra-low latency and advanced NVIDIA In-Network Computing, positioned for trillion-parameter-scale generative AI. The Q3400-RA is the air-cooled 4U system in this family, designed for standard 19-inch rack cabinets. It offers 144 ports at 800Gb/s distributed across 72 octal small form-factor pluggable (OSFP) cages, with 115.2 Tb/s throughput.

The platform incorporates remote direct-memory access (RDMA), fourth-generation NVIDIA Scalable Hierarchical Aggregation and Reduction Protocol (SHARP), adaptive routing, telemetry-based congestion control, self-healing and Quantum-X Photonics co-packaged optics (CPO) technologies. These features elevate overall application performance within HPC and AI data centers.

The Q3400-RA 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. All Quantum-X800 switches include a dedicated OSFP InfiniBand in-band management port for NVIDIA UFM, separated on the front panel from the data ports.

Key Features

• 144 × 800Gb/s non-blocking InfiniBand ports across 72 OSFP connectors, with 115.2 Tb/s throughput.
• Ultra-low latency and 200Gb/s per-lane SerDes technology for high-performance AI and HPC fabrics.
• Fourth-generation NVIDIA SHARP with FP8 precision and new collective operations such as ReduceScatter and ScatterGather.
• Enhanced adaptive routing and telemetry-based congestion control for nearly perfect effective bandwidth and performance isolation in multi-tenant and multi-job environments.
• Quantum-X Photonics CPO technology to reduce total power consumption, latency and TCO while enhancing resiliency and serviceability.
• Self-healing technology and communications framework offloads for resilient, high-throughput fabrics.
• NVIDIA NVOS software for comprehensive chassis management and system configuration, with CLI, REST APIs, SNMP and gNMI telemetry.
• Dedicated OSFP InfiniBand in-band management port for UFM, separated from data ports for simplified port allocation.
• Power capping and optional router capabilities for multi-site InfiniBand cluster expansion.

Technology

800Gb/s InfiniBand for Trillion-Parameter AI

NVIDIA Quantum-X800 InfiniBand switches deliver 800Gb/s throughput with ultra-low latency and advanced NVIDIA In-Network Computing, which is essential for handling trillion-parameter-scale generative AI. A two-tier Quantum-X800 fat-tree topology can support over 10,000 800Gb/s host connections, enabling 2x faster speeds and 5x higher scalability for AI compute fabrics compared with the previous generation.

Fourth-Generation NVIDIA SHARP

Quantum-X800 switches support SHARP for efficient offloading of compute operations to the network, boosting performance by up to 9x. The fourth generation of SHARP adds support for FP8 precision and new collective operations, such as ReduceScatter and ScatterGather, extending in-network computing to a broader set of AI workloads.

Adaptive Routing and Telemetry-Based Congestion Control

Quantum-X800 switches support enhanced adaptive routing and telemetry-based congestion control. This enables nearly perfect effective bandwidth, as well as performance isolation for multi-tenant and multi-job environments, helping operators maintain predictable application performance across shared fabrics.

Quantum-X Photonics Co-Packaged Optics

NVIDIA Quantum-X Photonics CPO technology reduces total power consumption, latency and total cost of ownership while enhancing resiliency and serviceability. This is achieved by eliminating the need for pluggable transceivers and minimizing the distance and number of connections between optics and electronics. The Q3400-RA is the pluggable-optics air-cooled model; the Q3450-LD is the liquid-cooled CPO MPO12 variant.

Management with NVOS and UFM

Quantum-X800 switches include NVIDIA Networking OS Software (NVOS) for comprehensive chassis management and system configuration. NVOS supports a command-line interface (CLI), REST APIs, Simple Network Management Protocol (SNMP) and gRPC Network Management Interface (gNMI) telemetry. A dedicated OSFP InfiniBand in-band management port for NVIDIA UFM is separated on the front panel from the data ports, allowing the full set of standard ports to be used for data network connectivity and streamlining topology design.

Typical Deployments

• Trillion-parameter generative AI clusters: fabrics that require 800Gb/s per port and in-network computing for large-scale training.
• Two-level fat-tree AI compute fabrics: topologies connecting up to 10,368 NICs with minimal latency and optimal job locality.
• Extreme-scale HPC: data centers that need high radix, ultra-low latency and self-healing InfiniBand networking.
• Multi-tenant and multi-job environments: fabrics that require performance isolation and nearly perfect effective bandwidth.
• Standard 19-inch air-cooled racks: the Q3400-RA is designed for air-cooled data centers using pluggable optics.
• Government, healthcare, manufacturing, education, finance and enterprise: organizations building large-scale AI and HPC infrastructure.

Compatibility

The Q3400-RA is an NVIDIA Quantum-X800-series 800Gb/s InfiniBand switch with 144 non-blocking 800Gb/s ports across 72 OSFP connectors. It is an air-cooled system designed for standard 19-inch rack cabinets, managed through NVIDIA NVOS and UFM. Compatibility depends on the specific fabric design, cable plant, optics and host channel adapters in your environment.

• 72 OSFP connectors; pluggable optics connectivity. Exact optics and cable compatibility should be confirmed against the NVIDIA compatibility list for your deployment.
• Dedicated OSFP 400Gb/s InfiniBand in-band management port for UFM, separated from data ports.
• NVOS software with CLI, REST APIs, SNMP and gNMI telemetry.
• Power feed: 200–240 VAC; confirm site power and rack clearance before ordering.
• Optional router capabilities support expansion of InfiniBand clusters across multiple sites.

For fabric-level compatibility, optics qualification and firmware matching, please confirm with our technical team before ordering.

Specification Table

Source: NVIDIA Quantum-X800 Q3400-RA datasheet provided by the customer. Parameters not publicly specified in the source are marked accordingly.

Selection Guide

The Quantum-X800 family includes air-cooled and liquid-cooled systems with different form factors, connectivity and power options. Use the table below to match rack type, cooling and connectivity to your data center design.

Selection note: the Q3400-RA is the air-cooled, AC-powered 4U model for standard 19-inch racks with pluggable optics. For DC power in air-cooled racks, consider Q3401-RD; for liquid-cooled CPO deployments, consider Q3450-LD or Q3300-LD.

Advantages

• Highest scale for AI: 2x faster speeds and 5x higher scalability for AI compute fabrics; a two-tier fat-tree can support over 10,000 800Gb/s host connections.
• Next-generation in-network computing: SHARP boosts performance by up to 9x, with FP8 precision and new collective operations.
• Higher effective bandwidth: enhanced adaptive routing and telemetry-based congestion control deliver nearly perfect effective bandwidth and performance isolation.
• Enhanced power efficiency and resiliency: Quantum-X Photonics CPO reduces total power consumption, latency and TCO while improving resiliency and serviceability.
• Enhanced software operations: NVOS supports CLI, REST APIs, SNMP and gNMI telemetry for comprehensive chassis management.
• Simplified topology design: a dedicated OSFP in-band management port for UFM is separated from data ports, so the full set of standard ports is available for data connectivity.

Service & Support

Hong Kong Starsurge Group Co., Limited supports this product with an experienced sales and technical team. Support scope can include pre-sales configuration review, rack and power confirmation, compatibility checks, and order follow-up.

• Pre-sales selection support for form factor, cooling, connectivity and power feed.
• Compatibility confirmation for OSFP optics, cables and host adapters before ordering.
• Order configuration and lead-time coordination.
• Multilingual support and global delivery available through Starsurge.

FAQ

What is the Q3400-RA?

It is a 4U air-cooled 800Gb/s InfiniBand switch in the NVIDIA Quantum-X800 series, with 144 non-blocking 800Gb/s ports across 72 OSFP connectors, 115.2 Tb/s throughput, NVOS management and UFM in-band management port.

How many ports and what throughput does it provide?

144 × 800Gb/s non-blocking ports across 72 OSFP connectors, with 115.2 Tb/s throughput.

What is the difference between Q3400-RA and Q3401-RD?

Both are 4U air-cooled 800Gb/s systems. The Q3400-RA uses a 200–240 VAC power feed, while the Q3401-RD uses a 48–54 VDC power feed for rack cabinets utilizing DC power.

Does the Q3400-RA use co-packaged optics?

The Q3400-RA is the pluggable-optics air-cooled model with 72 OSFP connectors. The Quantum-X Photonics CPO technology is described at the platform level and is offered in the liquid-cooled MPO12 variant Q3450-LD. Confirm your intended optics model against the NVIDIA compatibility list before ordering.

Does it support fourth-generation SHARP?

Yes. Quantum-X800 switches support SHARP for efficient offloading of compute operations to the network, boosting performance by up to 9x. The fourth generation adds FP8 precision and new collective operations such as ReduceScatter and ScatterGather.

What management software does it use?

It uses NVIDIA Networking OS Software (NVOS) for chassis management and configuration, with CLI, REST APIs, SNMP and gNMI telemetry. A dedicated OSFP InfiniBand in-band management port supports NVIDIA UFM.

Can you confirm compatibility with my optics and cables?

Yes. Please share your planned OSFP optics, cables and host adapter models, and our technical team will confirm compatibility before ordering.

Precautions

• Confirm rack type and cooling: the Q3400-RA is a 4U air-cooled system designed for standard 19-inch rack cabinets.
• Confirm power feed: 200–240 VAC for Q3400-RA; DC-powered racks should consider Q3401-RD.
• Confirm optics and cabling: the Q3400-RA uses 72 OSFP connectors and pluggable optics; validate against the NVIDIA compatibility list.
• Confirm whether your deployment requires CPO-based liquid cooling (Q3450-LD) instead of pluggable optics.
• System weight, dimensions and operating