Brand: NVIDIA | Category: Network Switches
SKU: NVIDIA-920-23686-2560-000 | Part #: 920-23686-2560-000 | MPN: 920-23686-2560-000
Contact for Pricing — Request a Quote
The spine tier in a two-tier non-blocking fat-tree fabric determines layer 2 and layer 3 switching behavior, with NVLink Switch Chip 2 technology enabling the high-radix interconnect required for AI factory data centers. The NVIDIA NVLink Switch System (NVL72 Spine), manufacturer part number 920-23686-2560-000, is purpose-built as the second tier of the GB200 NVL72 Rack-Scale AI Platform, delivering fifth-generation NVLink 5 connectivity across 72 NVIDIA Grace Blackwell Superchips (GB200) per rack. This spine subsystem is not a standalone network switch; rather, it operates as an integrated component within the NVIDIA GB200 NVL72 rack architecture, paired with NVLink leaf switches to create a unified fabric topology optimized for massive GPU-to-GPU bandwidth and minimal latency in hyperscale compute environments.
AI infrastructure teams, network architects, and hyperscale data center operators purchase the NVIDIA NVLink Switch System to enable the extreme all-to-all connectivity required by large-scale foundation model training and inference workloads. Organizations deploying the GB200 NVL72 platform rely on this spine-tier switching fabric to eliminate bandwidth bottlenecks between GPU clusters, ensuring that 72 GB200 superchips within a single rack can communicate at line rate without contention. The spine system integrates seamlessly with NVIDIA Base Command and NVLink fabric management tools, allowing centralized monitoring and orchestration across the entire rack-scale AI platform without requiring separate network management consoles.
Omnixon Global supplies the NVIDIA NVLink Switch System (NVL72 Spine), part number 920-23686-2560-000, to leading AI research institutions and cloud infrastructure operators across the UAE and broader EMEA region. For organizations planning to deploy or expand GB200 NVL72 rack infrastructure, contact our enterprise sales team to request a detailed quote, technical configuration review, and lead-time assessment for this critical fabric component.
| Brand | NVIDIA |
| Category | Network Switches |
| SKU | NVIDIA-920-23686-2560-000 |
| Part Number | 920-23686-2560-000 |
| Condition | New |
| Capacity | 1.8TB |
| Manufacturer Part Number | 920-23686-2560-000 |
| Product Name | NVIDIA NVLink Switch System (NVL72 Spine) |
| Product Family | GB200 NVL72 Rack-Scale AI Platform |
| NVLink Generation | NVLink 5 (fifth generation) |
| Switch Tier | Spine (second tier in two-tier NVLink switch fabric) |
| Interconnect Technology | NVLink Switch Chip 2 |
| Fabric Topology | Two-tier non-blocking fat-tree (leaf + spine) |
| GPU Connectivity (per rack) | 72 NVIDIA Grace Blackwell Superchips (GB200) |
| Form Factor | Rack-integrated subsystem within GB200 NVL72 rack |
| Deployment Context | NVIDIA GB200 NVL72 rack; not a standalone network switch |
| Target Infrastructure | AI factory data centers, hyperscale GPU compute facilities |
| Management Integration | Managed via NVIDIA Base Command and NVLink fabric management tools |
| Compatibility | NVIDIA GB200 NVL72 rack platform with NVLink leaf switches |
The NVIDIA NVLink Switch System (NVL72 Spine) is enterprise networking hardware for data-center, campus, and branch deployments. Common roles include core/distribution switching, server uplinks, top-of-rack fabric, and edge connectivity in mixed Cisco/Juniper/Arista environments.
Key specifications for the NVIDIA NVLink Switch System (NVL72 Spine): capacity 1.8TB; 3 years support; new condition; performance 8TB/s all-to-all GPU bandwidth within GB200 NVL72 rack systems; support 3 years standard (manufacturer). Manufacturer part number 920-23686-2560-000. For the full datasheet with electrical, environmental, and compliance details, contact our pre-sales engineering team.
The NVIDIA NVLink Switch System (NVL72 Spine) interoperates with standard switching protocols (LACP, MLAG/VPC, 802.1Q, OSPF/BGP). For specific cross-vendor scenarios — Cisco-to-Arista MLAG peering, fabric integration with Cumulus/SONiC, or migration from legacy Catalyst — our network engineers map a compatibility plan as part of the RFQ.