Xsight Labs Announces X2 Programmable SDN Ethernet Switches for Hyperscale and Edge Data Centers Optimized for the AI Factory Era

X2 delivers performance and flexibility to address evolving AI and Compute workloads at the lowest Total Cost of Ownership

SAN JOSE, Calif. & TEL AVIV, Israel–(BUSINESS WIRE)–Xsight Labs, a fabless semiconductor company delivering end-to-end connectivity solutions for next-generation AI Factory, Traditional hyperscale IaaS, Next Generation Edge and private data center networks, today announced the availability of the X2, the second-generation X-Switch family of fully programmable SDN Ethernet switches.


X2 is designed to enable enterprise and cloud service providers to operate their networks at peak efficiency by addressing the key bottlenecks associated with data movement in edge and AI clusters. A monolithic die in 5nm, the X2 supports 12.8 Terabits per second (Tbps) throughput and is optimized for interconnecting AI, storage and compute clusters in data centers.

“Xsight Labs is proud to unveil X2, an ultra-low-power, fully programmable SDN data center switch,” said Yossi Meyouhas, Xsight Labs CEO. “The X2 is optimized for cloud networks transitioning to native 100G SerDes connectivity to the host and 800G optics to the fabric. At 200W, X2 delivers a 40% improvement in power consumption compared to incumbent 12.8Tbps top-of-rack switches in the network today.”

“Oxide has selected the Xsight Labs X2 12.8T for the next-generation switch in our Oxide cloud computer. We strongly believe that programmability is the future of network data planes, and we believe that the deeply programmable X2 leads the industry. Moreover, the commitment of the Xsight Labs team to a crisp, documented hardware/software interface allows us to better build the compiler, tooling and system software at the heart of the cloud computer. We have said that switching silicon is awaiting its ‘x86 moment’: a committed instruction set that allows for not just a world-beating chip but also unlocks open systems software innovation. We believe the moment may have finally arrived with the X2, and we at Oxide are thrilled to be among its early adopters,” said Bryan Cantrill, Oxide Computer Co., Co-Founder and CTO.

“Credo has been supporting Xsight Labs for multiple years with our HiWire AECs, PHYs and Gearbox ICs,” said Don Barnetson, VP of Product at Credo Semiconductor. “The X2 12.8T TOR refresh employs 128x100G and Credo is excited to continue this partnership. Our HiWire SHIFT AECs allow the X2 to serve legacy NICs, DPUs and accelerators while enabling migration to modern hosts at 112G using Credo’s HiWire SPAN and CLOS AECs.”

“X2 supports a broad range of SKUs from 3.2Tbps at 140W to 12.8Tbps at 200W,” said Gal Malach, Co-Founder and CTO at Xsight Labs, “This provides hyperscale and enterprise cloud customers with an extensive portfolio of programmable, power-efficient and cost-effective switches for their 25G, 50G and 100G SerDes-based connectivity requirements.”

The X-Switch family delivers application-optimized switching for the 100G SerDes ecosystem. The X2 is a scalable, power-efficient, flexible solution supporting advanced features like congestion-aware routing, fast link recovery, and Ultra Ethernet Consortium (UEC) compliance to improve the overall efficiency of the network for AI workloads. The X2 enables best-in-class network visibility empowering operators to troubleshoot network failures with little to no downtime, leading to CapEx/OpEx savings for cloud service providers and edge operators.

X2 Key Features:

  • Up to 12.8Tbps full-duplex throughput
  • Supports a full portfolio of switch configurations from 3.2Tbps to 12.8Tbps
  • Ultra-low power: From 3.2Tbps @ ~140W to 12.8Tbps @ ~200W
  • Low latency: below 700nsec first-bit-in-first-bit-out for typical data center applications
  • 100G SerDes with in-rack DAC connectivity without re-timers
  • Extensive tool suite optimized for AI at the edge:

    • X-IQ™ Engine: Ultra Ethernet Consortium (UEC)-ready programmable congestion measurement and signaling for dynamic load balancing
    • X-LB™: UEC-ready end-to-end congestion awareness and dynamic load balancing for reducing tail latency
    • X-FC™: Incast tolerance mechanism with Head-of-Line (HoL) blocking prevention
  • Full programmability and real-time performance insights:

    • X-PND™ Engine: Elastic resource allocation of control tables, maximizing adaptation to application needs
    • X-VIEW™: Telemetry suite for network troubleshooting and optimization
    • X-MON™: In-field device health monitor for improved reliability

Software

X2 is accompanied by a comprehensive Software Development Kit (SDK) that incorporates multi-layer APIs, allowing full control of the X2 silicon. The modular SDK enables support for customers’ proprietary network operating systems (NOS), is fully compatible with OCP SAI and SONiC network operating systems and comes with a set of reference applications.

For the full list of features, view the X2 product brief.

Availability

X2 is sampling to customers now. Please contact sales@xsightlabs.com for more information.

About Xsight Labs

Xsight Labs is a fabless semiconductor company that enables end-to-end connectivity for AI and hyperscale data centers. Xsight Labs’ technology delivers exponential bandwidth growth and unmatched versatility while lowering power and total cost of ownership. Founded in 2017, Xsight Labs is headquartered in Kiryat Gat, Israel with additional offices in Tel-Aviv and Haifa, and international offices in Yerevan, Armenia, as well as in Boston, Raleigh, and San Jose in the United States. For additional information, visit www.xsightlabs.com.

Contacts

Matt Baxter

Nereus for Xsight Labs

xsightlabs@nereus-worldwide.com

Artículos Relacionados