Resources on HPC InfiniBand & AI, Data Center Networking

OSFP IHS vs OSFP RHS: Thermal Design and Key Differences Analysis

This article introduces two thermal designs for OSFP IHS and OSFP RHS optical modules, explaining their main differences in structure, heat dissipation methods, and system integration. It helps users understand their respective characteristics and application scenarios to deploy optical transceivers appropriately in different environments.

Unveiling Multi-plane Networks: Principles, Physical Design, Advantages, and Applications

This paper analyzes how multi-plane networks, through parallel planes, load balancing, and shuffle box cabling, combined with architectural design, physical implementation, and intelligent scheduling mechanisms, achieve high bandwidth, low latency, and better cost, providing a reliable and efficient network foundation for large-scale AI and HPC clusters.

NVIDIA ConnectX-9 SuperNIC Analysis: 1x800G Ethernet Breakthrough

The ConnectX-9 SuperNIC achieves three major breakthroughs: providing 1x800G on a single port in Ethernet mode, integrating post-quantum cryptography for security, and deep integration with the Rubin GPU platform. Understand how the ConnectX-9 SuperNIC defines the next generation of AI infrastructure. Unlock the future of networking now.

AI Networking

CPO Switch: Next-Generation Integrated Optical Networking Interconnection Solutions

CPO switches shorten the electrical signal path, reduce power consumption, and decrease the number of pluggable modules by co-packaging optical modules with switch chips, while improving link reliability. This breaks through the bottlenecks of traditional switch architecture in terms of transmission performance and energy consumption, and provides a technical path for future high-bandwidth computing.

Neo

Dec 5, 2025

Google TPU: The AI Chip for the AI Inference Era

Google TPU vs NVIDIA GPU: Who Wins? This article provides a comprehensive analysis of the architectural evolution and system-level advantages of Google TPUs, delves into Ironwood, memory, and interconnect design, and discusses the differences between TPUs and NVIDIA GPUs, helping enterprises evaluate the best computing power solution for the era of large-scale computing.

Deep Dive into NVIDIA DGX B300

This article provides a comprehensive analysis of NVIDIA DGX B300, covering aspects such as computing performance, NVLink interconnect, ConnectX-8 network architecture, power supply and data center integration, and Mission Control operations management. Click to read and gain a complete understanding of the DGX B300.

The Key Role of High-quality Optical Transceivers in AI Networks

This article analyzes the potential faults and cost risks brought about by low-quality optical modules in AI networks, and introduces how high-performance optical modules can provide stable, low-latency interconnection under high-load environments, improve GPU cluster utilization, reduce downtime losses, and build highly reliable and scalable AI infrastructure.

Blogs