End-to-End PCIe® over Optics for GPU Clusters: First Look Demo

For this demo, we have assembled two common configurations for reach extension: head node to GPU clusters and head node to remote, disaggregated memory systems. This system is running PCI Express® at full rate over single mode fiber for an aggregate bandwidth of 128GB/s and a reach of 20 meters in this demonstration. This reach can be easily extended to 50 meters or more based on application needs.