What technical features and architectural components allow InfiniBand fabrics to deliver the low latency and high bisection bandwidth required for AI and deep learning supercomputers? (Select all that apply.)
Select all correct answers, then click Submit.
Short Explanation and Infographic
Think of traditional Ethernet like a crowded party where everyone is trying to speak over the same microphone—that's your classic shared medium bus architecture, and it's terrible for high-performance computing. InfiniBand does things differently. It's a switched fabric, meaning every connection is a dedicated point-to-point path. To make things even faster, InfiniBand uses a Subnet Manager to coordinate paths, supports huge speeds ranging from old 10 Gbps SDR to modern 800 Gbps NDR, and relies on RDMA (along with GPUDirect RDMA) to let GPUs and memory communicate directly across the network without asking the CPU for permission. Got that? It's all about direct, dedicated pathways and hardware offloads. Keep that shared bus distractor out of your answers!
Full explanation below image
Full Explanation
InfiniBand is designed from the ground up for low-latency, high-throughput applications in cluster architectures. First, its Switched Fabric design utilizes point-to-point serial links rather than traditional Ethernet-style shared media, maximizing data isolation, throughput, and scalability. Second, the Subnet Manager functions as the control plane, handling topology discovery, configuring routing tables, and executing rapid, automatic path failovers. Third, Remote Direct Memory Access (RDMA) bypasses the local CPU and OS kernel, enabling direct memory writes and reads between hosts to eliminate context-switching latency. Fourth, GPUDirect RDMA enables direct PCIe-to-network paths for GPU memory transfers across nodes, eliminating the need to copy data to system RAM first. Fifth, it supports scalable transmission rates spanning from Single Data Rate (SDR) at 10 Gbps up to Next Data Rate (NDR) at 800 Gbps per link. Choice A is incorrect because InfiniBand uses a point-to-point switched fabric architecture, not a shared-bus or shared-medium topology.