High-bandwidth, low-latency, multi-terabit network switches
The Marvell network switch product line for data centers addresses general-purpose compute and accelerated compute applications for cloud, AI and converged data centers. The high-bandwidth Teralynx® Ethernet switch family offers a unified architecture that addresses ToR, spine/leaf, super spine and edge applications with capacities up to 51.2 Tbps.
- AI Cloud Network Switches
- PCIe switches
Teralynx®
AI Cloud Network Switches
The Teralynx network switch device family is built on a clean-sheet architecture that delivers a no-compromise, industry-unique combination of high bandwidth, low and predictable latency, high radix, low power and latency-transparent programmability.
Designed for both cloud data center and AI fabrics, Teralynx devices are programmatically configurable, enabling a single device to perform optimally in data center top-of-rack (ToR), spine/leaf, and edge applications, as well as frontend and backend AI cluster applications.
Teralynx switch silicon is proven and deployed at scale in the largest cloud data centers, with millions of ports shipped.
- High-bandwidth options from 6.4 Tbps to 51.2 Tbps
- Open networking support including OCP SAI and SONiC for faster development
- Ultra-low, deterministic latency reduces job completion time, eases bottlenecks
- High radix (up to 512) reduces cost and complexity by collapsing network tiers
- Programmable forwarding pipeline protects investment, enables reconfiguration
- Teralynx Flashlight™ visibility and telemetry simplifies operation, enhances visibility
Teralynx 10
51.2 Tbps, full-duplex switching processor for use in fixed switch systems targeted for cloud-scale spine and leaf, AI cluster and data center interconnect applications. (P/N: 98TX9180)
Teralynx 7
12.8 Tbps, full-duplex switching processor for use in fixed and modular switch systems targeted for cloud-scale top-of-rack, end-of-rack and spine and leaf applications. (P/N: IVM77700)
Prestera 3500 Series
L2/L3+ switch devices ideal for data center management supporting up to 48x1/2.5GbE downlink ports, 10/25GbE uplink ports and 25/50GbE stacking ports (P/N: 98DX35xx)
Structera S
PCIe Switches
The Structera S series of PCIe switches is designed to enable AI infrastructure and fanout connectivity. These devices offer ultra-low latency, high radix connectivity, and advanced fabric capabilities required for next-generation AI and HPC workloads. With an industry-first 256 lanes, the Structera S 50256 device eliminates the need for multiple smaller switches, thereby flattening the network, reducing latency and enhancing performance of AI systems. Structera S switch devices are also available for CXL applications.
- PCIe 5.0 compliance
- Configurable 16 ports (x16) or 32 ports (x8)
- Up to 2 TB/s switching capacity
- Advanced traffic management with virtual output queues (VOQ)
Structera S 50256
PCIe 5.0 switch with up to 2 TB/s switching capacity and configurable 16-port and 32-port options (P/N: XC51256_3)
Newsroom
Media
Contact Us
We believe better partnerships help to build better technologies. Let’s connect and see what we can design together!
Thank You for Your Interest
We will be in touch with you soon!
