XConn Technologies Unveils Game-Changing Hybrid Switch for High-Performance Computing and AI Applications


In a leap forward for the realm of high-performance computing and artificial intelligence, XConn Technologies (XConn) has unveiled a groundbreaking innovation that is set to redefine data center interconnect technology. The industry pioneer has introduced the world’s inaugural hybrid switch, seamlessly combining Compute Express Link (CXL) 2.0 and Peripheral Component Interconnect Express (PCIe) Gen 5 interconnect technologies within a single, powerful 256-lane System on Chip (SoC). Named the XC50256 switch, or “Apollo,” this marvel of engineering ushers in unparalleled advancements in latency, power efficiency, and cost-effectiveness.

Emerging from its “stealth mode” debut, XConn chose the prestigious Flash Memory Summit, held in Santa Clara, California, to unveil the Apollo switch. This exceptional technological achievement marks a significant stride towards reshaping the landscape of high-performance computing and AI applications.

Designed with meticulous attention to detail and purpose-built for artificial intelligence (AI), machine learning (ML), and high-performance computing (HPC) applications, the Apollo switch represents a quantum leap forward in interconnect technology. Embodying the fusion of CXL 2.0 and PCIe Gen 5, this innovative switch empowers system designs with the capability of both CXL memory pooling and expansion functions, leveraging the potency of CXL 1.1 hardware while ensuring future compatibility with CXL 2.0. Notably, the Apollo switch’s hybrid mode accommodates both CXL and PCIe devices within the same system, offering unparalleled flexibility for system vendors to optimize components for AI system design while seamlessly transitioning from PCIe to CXL in a heterogeneous computing environment.

XConn’s announcement also extends to the unveiling of the XC51256 switch, a PCIe Gen 5.0-only variant that holds the distinction of being the most-dense PCIe Gen 5.0 switch in the industry. With a staggering 256 lanes, this product emerges as the ultimate solution for Just-a-Bunch-Of-GPUs (JBOG) and Just-a-Bunch-Of-Accelerators (JBOA) processing configurations. Notably, the XC51256 switch stands out by minimizing latency and power consumption per port while conserving valuable board space—a remarkable feat achieved by enabling single-chip usage models that outperform competitors reliant on multiple chips.

The unveiling of the Apollo switch by XConn signifies a crucial breakthrough in addressing the connectivity challenges posed by data-intensive workloads, particularly in AI/ML and genomics processing. By nearly doubling the switch lanes provided by rival solutions and halving port-to-port latency and power consumption per port, the Apollo switch could represent a significant stride forward in transcending bandwidth barriers and accelerating processing speeds.

XConn’s remarkable achievement stems from the collaborative efforts of a team of interconnect veterans, who founded the company in 2020. Boasting decades of expertise in the data center interconnect and switching domains, XConn’s leadership team also comprises active contributors to esteemed organizations such as the CXL Consortium, Open Compute Project, and PCI-SIG, further underscoring the company’s commitment to industry advancement and innovation.

The unveiling of the Apollo switch at the Flash Memory Summit not only reaffirms XConn’s dedication to driving technological progress but also signals a seismic shift in high-performance computing and AI, inspiring a new era of interconnect possibilities that will undoubtedly shape the future of the digital landscape.

News Desk

Print Friendly, PDF & Email