Disaggregated Scale-Up Architecture

Deliver extended reach, low latency, and high data throughput with multi-rack PCIe® 7 connectivity for large GPU clusters required to address the growing complexity and size of AI models.

Similar Solutions

Challenges

  • Disaggregation of scale-up clusters across multiple racks of compute, switch, and memory means high-bandwidth links must extend to reaches of 50+ meters
  • Low-latency PCIe 7 links at these long reaches are required to ensure maximum GPU utilization and lower total cost of ownership
  • High-radix switch and GPU architectures in disaggregation demand connectivity solutions with higher density than front panel pluggables
  • Heterogeneous platforms increase integration complexity, requiring interoperability, validation, and system-level visibility across diverse GPUs, CPUs, XPUs, and endpoints

Solution: Aries PCIe Smart Optical Drivers and TIAs

  • Aries PCIe Smart Optical Drivers and TIAs enable high-density, low latency, low-power linear pluggable optics (LPO) and near-packaged optics (NPO) links extending beyond 50 meters across multi-rack systems
  • High-density optical drivers and TIAs combine with Scorpio™ Smart Fabric Switches to provide high-radix, low-latency PCIe fabric connectivity for scaling disaggregated GPU architectures
  • The Aries Smart Signal Conditioner family – the industry’s first and broadest PCIe 7 signal conditioning portfolio – extends reach across retimer, redriver, cable, and optical links while providing flexibility in copper and optical connectivity
  • COSMOS Optical Intelligence delivers end-to-end telemetry, diagnostics, security, and dynamic link tuning to accelerate deployment and simplify fleet management