The NVIDIA GB200 NVLink Switch System is a critical component of the Blackwell-based AI infrastructure, enabling seamless, high-speed communication between GPUs. It facilitates the creation of a massive, single-GPU domain for large-scale AI training and inference workloads.
- Based on the NVIDIA Blackwell architecture for extreme-scale AI clusters.
- Delivers up to 1.8TB/s of bidirectional throughput per GPU via fifth-generation NVLink.
- Supports NVLink Switch Tray configurations for up to 72-GPU domains in a single rack.
- Features integrated SHARP v4 (Scalable Hierarchical Aggregation and Reduction Protocol).
- Provides 130TB/s of aggregate bandwidth in NVL72 configurations.
- High-density 1U rackmount form factor optimized for data center efficiency.
- Advanced telemetry and monitoring for large-scale cluster management.
- Optimized for liquid-cooled and air-cooled rack environments.
- Essential for multi-node scaling of Large Language Models (LLMs).
- Reduces communication overhead and latency in distributed AI training.
NVIDIA
NVIDIA - GB200 NVLink Switch System - High-Performance AI Interconnect Switch
For a quote, contact us at info@tropical.com
- SKU:
- GB200 NVLink Switch System
- Weight:
- 25.00 LBS
- Width:
- 17.30 (in)
- Height:
- 1.75 (in)
- Depth:
- 30.00 (in)
Related Products
A relatable product
NVIDIA
NVIDIA - DGX H200 NVLink System - AI Supercomputing Infrastructure
The NVIDIA DGX H200 is the world’s premier AI platform, purpose-built for the most demanding generative AI and high-performance computing (HPC) workloads. It leverages the NVIDIA H200 Tensor Core GPU...
NVIDIA
NVIDIA - GB200 Liquid Cooling Module - High-Performance AI Infrastructure Cooling
The NVIDIA GB200 Liquid Cooling Module is a critical thermal management component designed for the Blackwell architecture, specifically the GB200 Grace Blackwell Superchip. It enables high-density AI...
NVIDIA - DGX GB200 - Rack-Scale AI Infrastructure System
The NVIDIA DGX GB200 is a rack-scale AI factory purpose-built for training and inferencing trillion-parameter generative AI models. It integrates 36 NVIDIA GB200 Grace Blackwell Superchips into a...
NVIDIA
NVIDIA - GB200 - Grace Blackwell Enterprise AI Superchip
The NVIDIA GB200 Grace Blackwell Superchip is a premier AI module engineered for the most demanding trillion-parameter large language model (LLM) workloads. It combines the power of the NVIDIA Grace...
NVIDIA - DGX SuperPOD (GB200) - AI Infrastructure Supercomputer
The NVIDIA DGX SuperPOD with GB200 systems is a leadership-class, liquid-cooled AI infrastructure designed for the most challenging generative AI training and inference workloads. It provides a...