Blogs

RDMA over Converged Ethernet (RoCE) Explained: What Is RoCE and How to Build RoCE Networks

RDMA over Converged Ethernet (RoCE) Explained: What Is RoCE and How to Build RoCE Networks

Explore what is RDMA over Converged Ethernet (RoCE), RoCE network architecture and core components, learn how to design scalable, lossless RoCE networks for AI, HPC, and cloud data centers.

NADDOD N9570-128QC: 51.2T 128×400G Switch Powered by NVIDIA Spectrum-4 Chip

NADDOD N9570-128QC: 51.2T 128×400G Switch Powered by NVIDIA Spectrum-4 Chip

As AI training clusters scale, network latency, stability, and link utilization become critical. Learn how the NADDOD N9570-128QC leverages Spectrum-4 and ConnectX-8 to address RoCE performance challenges in large GPU environments.

Optical Transceivers

A Comprehensive Guide to 400G OSFP Ethernet Optical Transceivers

A Comprehensive Guide to 400G OSFP Ethernet Optical Transceivers

Explore 400G OSFP Ethernet optical transceivers for modern data centers, AI and HPC networks. Learn OSFP advantages, use cases, and NADDOD’s 400G OSFP solutions for high-density, high-performance connectivity.

From RoCE to Spectrum-X: The Evolution of Ethernet Networking for AI Data Centers

From RoCE to Spectrum-X: The Evolution of Ethernet Networking for AI Data Centers

This article explores the evolution of RoCE networking and the technical challenges of Ethernet in large-scale AI clusters, and explains how NVIDIA Spectrum-X improves lossless transport, congestion control, and performance predictability for AI data centers.

InfiniBand Networking

NVIDIA MetroX-2: A Remote Interconnect System for Scale-Across

NVIDIA MetroX-2: A Remote Interconnect System for Scale-Across

Against the backdrop of scale-across becoming the inevitable path for AI computing power expansion, this paper analyses how NVIDIA MetroX-2, a metropolitan-scale AI networking system, enables cross-data centre training and inference through low-latency, highly predictable remote interconnects. This facilitates genuine cross-domain computing power coordination and resource pooling.

Physical AI Analysis: From Information Intelligence to Real-World Intelligence

Physical AI Analysis: From Information Intelligence to Real-World Intelligence

Physical AI is driving artificial intelligence from the digital space into the real world. This article systematically introduces the development stages, core working mechanisms, typical applications, and NVIDIA's full-stack technology layout of physical AI.

High Performance Computing

Analyzing DGX Spark and DGX Station: NVIDIA's Deskside AI Supercomputing

Analyzing DGX Spark and DGX Station: NVIDIA's Deskside AI Supercomputing

An in-depth analysis of how DGX Spark and DGX Station bring data center–class AI computing to the Deskside , enabling local development, fine-tuning, and deployment of large models while reducing costs, enhancing security, and accelerating the adoption of AI in real-world business scenarios.

Analyzing the Impact of NVIDIA BlueField-4 on AI Context Inference

Analyzing the Impact of NVIDIA BlueField-4 on AI Context Inference

Learn how the BlueField-4 DPU optimizes context management for large language models in AI systems, achieving cross-node sharing, low-latency access, and system-level scalability through a dedicated context memory layer and high-speed key-value cache.

Introduction to the Three Key Processing Cores Inside NVIDIA GPUs

Introduction to the Three Key Processing Cores Inside NVIDIA GPUs

Starting with the SM organizational structure of NVIDIA GPUs, this paper outlines the architectural positioning and capability focus of internal computing cores such as CUDA core, Tensor core, and RT core, and explains the differences in the division of labor and applicable scenarios of different processing cores in NVIDIA GPUs.