Aug 2026· Nature Reviews Electrical Engineering· Vol 3, pp. 564 - 578· 0 citations· 145 references
TL;DR
A one-chip-like standpoint for understanding how datacenter infrastructures are evolving to support AI workloads is introduced, highlighting how one-chip-like datacenter extends chip-level semantics to rack scale by combining compute express link fabrics with hardware mechanisms, thereby stabilizing latency variability along traversal paths.
It is argued that scale-up fabrics should become software-defined, as programmable switches did for Ethernet over the past decade, and that the switch is the natural place to begin.
A new class of highly scalable, effective, and resilient network-native services is made possible by Synapse, which removes the main storage bottleneck in contemporary data centers by offloading the entire telemetry capture process to the network fabric.
Seyed Hossein Ahmadpanah, A. Sahafi, S. Erfani· Journal of Supercomputing· 0 citations
This paper examines SDCs from both architectural and computing perspectives, and presents a CiM case study that quantifies that cost, and finds that mitigation effectiveness is axis-dependent, with only periodic recalibration improving all four stress axes, and the cell mapping outranks the choice of device technology...
Soyed Tuhin Ahmed, B. Farahani, Krishnendu Chakrabarty et al.· Coins· 0 citations
This survey classifies industrial AI accelerators into four architectural categories and compares their compute and memory organizations, and examines how node-, rack-, and pod-scale interconnects support collective communication, and traces architectural evolution across accelerator generations.
Data-center workloads driven by artificial-intelligence training, cloud-native services, and distributed microservices have outgrown the three-tier network designs that once served enterprise applications. Hierarchical aggregation with the Spanning Tree Protocol wastes capacity on blocked links and introduces variable...
Chetan Padshala· International journal of com...· 0 citations
Large-model parameters have grown beyond the capacity of a single xPU, dispersing across multiple xPUs spanning distinct host domains, where xPU-to-xPU communication dominates overall system efficiency. Existing scale-up interconnect remains inadequate: network-based solutions built on Ethernet--such as RoCE (RDMA over...
Zhe-Zhe Peng, Jia-Ming Huang, Chao-Fan Chen et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.