With the continued growth in AI computing demand, data center networks are rapidly transitioning from 400G to 800G. However, many data centers still deploy 100G and 200G server network interface cards (NICs), making it difficult to upgrade all servers to 800G in the short term. How can high-speed connectivity between an 800G switching network and […]
TL;DR: The 1.6T OSFP224 DAC is mechanically and electrically backward compatible with existing 400G and 800G OSFP ports, supporting signal protocols from 25G/Lane NRZ all the way to 224G/Lane PAM4. If your racks already run 400G or 800G, you don’t need to replace your switches, panels, or cabling infrastructure to step up to 1.6T—just swap […]
TL;DR: The 1.6T OSFP224 DAC is available in both Finned Top (IHS) and RHS (Riding Heat Sink) designs, with full interoperability between the two. Whether your rack uses air-cooled switches, liquid-cooled AI servers, or a mix of both, one DAC product line covers every thermal scenario—reducing SKU complexity and simplifying inventory management across heterogeneous infrastructure. […]
The XPO optical transceiver, with a single-module bandwidth of 12.8 Tbps, increases the front-panel density of switches to 4 times that of OSFP solutions. While high-density switch ports boost computing power, they also introduce new challenges for fiber optic cabling. This article briefly introduces the MPO pre-terminated trunk cables that support XPO optical modules. What […]
In modern data centers—especially in AI and 400G/800G environments—fiber cabling is no longer just about connectivity. Choosing the wrong MTP®/MPO cable can lead to link failures, wasted fibers, and costly rework. This guide focuses specifically on MTP®/MPO-12 cables, explaining their matching logic with various transceivers and providing clear selection recommendations. 1. Core Matching Rule The relationship between […]
The NVIDIA ConnectX‑9 family of SuperNICs is a next‑generation high‑performance networking solution optimized for hyperscale AI computing workloads. Supporting up to 800 Gb/s InfiniBand and Ethernet connectivity, it significantly boosts system performance in AI factories and cloud data centers. Based on official technical documentation, this article systematically outlines the supported interfaces and compatible optical modules […]
The NVIDIA ConnectX‑8 family delivers a next‑generation high‑performance networking solution for cloud and data‑center environments, building upon ConnectX‑7 with further improvements in bandwidth, interface data rates, and manageability. Based on official technical documentation, this article systematically outlines the supported interfaces and compatible optical modules for ConnectX‑8 SuperNICs. Product Models and Interface Form Factors ConnectX‑8 SuperNICs […]
The NVIDIA ConnectX‑7 family of adapter cards delivers high‑performance networking solutions for cloud and data‑center environments, integrating InfiniBand and Ethernet connectivity with best‑in‑class hardware acceleration and security features. Based on official technical documentation, this article systematically outlines the supported interfaces and technical specifications of ConnectX‑7 adapter cards. Product Models and Interface Form Factors ConnectX‑7 adapter […]
NVIDIA’s ConnectX‑6 Dx family offers three 100GbE SmartNIC models that share the same hardware platform and networking performance. The key differences lie only in security features: Secure Boot and hardware cryptographic acceleration. Common Specifications (all three models) Dual‑port 100GbE QSFP56 PCIe Gen4 x16 host interface Supports 1/10/25/40/50/100GbE speeds RoCE, SR‑IOV, storage acceleration, stateless offloads Power […]
AI is reshaping data center cabling: denser east‑west traffic, higher speeds (400G/800G+), tighter timelines, and zero tolerance for error. That means more fiber, prefabricated modular systems, meticulous testing, and evolving standards like ANSI/TIA‑942. Why AI Changes Cabling Traditional data centers handle predictable north‑south traffic. AI training needs constant, low‑latency communication between many compute nodes (east‑west). […]
