Author
Alastair Waite
For the past few months, I’ve worked with my colleagues at CommScope to create an Artificial Intelligence (AI) ordering guide to help our customers and partners to navigate the myriad of structured cabling solutions available. This new world is both exciting and extremely dynamic in terms of technology choices.
In the AI industry, every day can bring a new technology announcement, a new product launch and new predictions from the industry press. Some of these developments align to one another, whilst others can seem to contradict, making it difficult for C-suite executives and data center operations directors to plan without second guessing themselves about whether they’ve made the right choice.
Discussing this business challenge with a colleague, they introduced me to a poem titled “The Road Not Taken” by the American poet Robert Lee Frost. In that poem, Frost takes a walk in the woods. When he comes to a fork in the road, he is forced to make a choice, to follow either path A or path B.
He can’t see what is at the end of either road; the view is obscured by undergrowth—meaning that a decision could only be based on what they could be seen directly in front of him. Having made a choice and getting to the end of the road, Frost begins to question if he made the right choice, or if he would ever have the chance to revisit the decision and where the other road might have led him.
I see this scenario reflected in the conversations and questions about AI technology I hear amongst data center owners on a regular basis. They often ask, what they should deploy? Will that decision support their company’s objectives? Will their infrastructure be future proofed—and where can they learn more? Like Frost, they can’t yet see the end of their proposed path and ponder the implications of their decisions based on available information.
To help answer some of these questions, we created an extensive ordering guide called “Data Center Cabling Solutions for NVIDIA AI Networks.” This document is targeted at data center owners and operators, network integrators and systems providers. These are the people who have responsibility for making AI data centers happen, and this document guides the reader through the cabling solutions (including optical fiber, twisted pair copper and fiber raceway) that are essential to meeting the demands for AI computing.
Highlighting CommScope’s solutions for AI applications in a single document simplifies the selection process for a network designer by showing how each of the solutions work together. It also defines where the NVIDIA transceivers are in the network, their interface speeds and the corresponding cabling options they require.
For further simplification, we thought hard about how best to design the document for easy navigation, so we made it interactive—enabling the reader to move quickly and easily through each of its sections via a tool bar at the bottom of each page. From here, you can choose to navigate by either link speeds (e.g. 200G, 400G or 800G), or via an architectural type (direct connect versus a structured cabling approach). We also included reference designs, giving examples of how to cable an NVIDIA DGX H100 scalable unit (SU), right up to the impressive NVIDIA DGX H100 Super POD.
When you find a solution and product that matches your specific needs, we’ve provided hyperlinks that will take you directly from the document to the product web pages on CommScope.com, where you can find live information and data sheets—which you can save to project BoMs that can be built using the “My Product Lists” section of our website.
Recognizing that there is more than one road to take, we offer systems-level guidance based on our collective experiences and customer insights from around the globe, including:
- The flexibility that can be brought to your networking backbone design by selecting an MPO-16 fiber infrastructure over a legacy MPO-8 system—and the additional densification benefits that this choice can bring to your pathways around the data hall and within data racks.
- How using mesh architectures in your data center design can support reduced patching complexity and simplify redundancy planning.
- When ultra low-loss (ULL) systems can assist in migrating to higher data rate applications in an AI network.
- Why choosing FiberGuide® optical raceway is an essential tool to support the distributed nature of networking equipment in a power-constrained AI network.
To me, the poem “The Road Not Taken” captures the current dilemma that many in the AI data center construction industry face, that is, that decisions need to be made which are never easy, especially if the view of either road is not clear, and that the opportunity to revisit your decision may not be possible. My advice is to choose your partners well, a trusted partner who has travelled many roads, one that has the experience to explain the implications of taking one decision over another.
Finally, take a guide with you that can give instant access to the type of detail that you’re going to need to simplify your AI journey. Download our guide: Data Center Cabling Solution for NVIDIA AI Networks, here.
And be sure to check out our Generative AI resources and articles here, where we regularly provide fresh insight and updated content about the subject.
