Edgecore Networks is expanding its portfolio, taking aim at the emerging trend of artificial intelligence (AI) and machine learning (ML) workloads running across data center networks.

The Taiwan-based vendor is out with its new 800G-optimized switch that provides high-performance Ethernet fabrics for AI/ML deployments.

The new switch, dubbed the DCS560, packs an impressive 51.2 Tb/s of total switching capacity with 64 ports supporting 800G interfaces, and is powered by Broadcom StrataXGS Tomahawk 5 series silicon. The Tomahawk 5 was first announced by Broadcom back in 2022 in a bid to help boost hyperscaler data center operations.

"AI/ML workloads impose a unique demand on the network in terms of low entropy, high intensity, and bursty flows," Nanda Ravindran, VP product management at Edgecore Networks, told SDxCentral. "It is important for such AI workloads to use DCS560 and take advantage of adaptive routing and dynamic load-balancing features to achieve the best job completion times."

Scaling to 800G for AI

According to Edgecore, hyperscale data centers are increasingly adopting Ethernet-based fabrics in order to reduce job completion times for AI workloads. The 2RU (rack unit) DCS560 aims to meet this need for scalable, high-throughput 800G switching in an easy-to-deploy form factor.

The switch is available in two variants: one with OSFP800 optical interfaces, and the other with QSFP-DD800 interfaces, providing customers flexibility in cabling and optics. Both models support inexpensive copper cables for short reaches, as well as long-distance ZR+ pluggable optics.

Edgecore has packed the DCS560 with high-availability features, including redundant power supplies and fans. This makes it suitable for mission-critical data center deployments that require five-nines of reliability. The switch also supports a wide operating temperature range.

On the Network Operating System (NOS) side, Edgecorp's DCS560 supports the open source SONiC (Software for Open Networking in the Cloud) platform. SONiC has become increasingly popular in recent years – especially with carriers – to help power telecom cloud services.

According to Ravindran, the DCS560 provides twice the bandwidth capacity, lower power consumption per bit and better load balancing compared to previous generations of Edgecore switch platforms. Looking forward, he said Edgecore plans to expand on the DCS560 family with higher capacity switches specifically for AI/ML use cases.

With its release, Edgecore joins a cadre of vendors racing to help support large-scale AI/ML data center deployments. Arista Networks and Cisco are also among the competitive vendors racing to provide high-speed switches for AI.