DCS864E

Highlights

51.2 T switch with 64 x 800 G OSFP or 128 x 400 G/512 x 100 G in 2 RU
World’s fastest switch with port-to-port latency under 560ns for 800 GE ports
Max TDP of 2200 W with 64 x 800 G SR8 ports under full traffic load
Large on chip buffer of 200+MB for better RoCE (RDMA over Converged Ethernet) performance
10ns PTP and SyncE performance supports strictly synchronized AI parallel computing
Advanced INT (in-band network telemetry) for packet delays, drops, and path traversal, enabling more advanced congestion control algorithms
Open SeraNOS based on SONiC with best SAI support, more robust and reliable.
Compatible with heterogeneous GPUs and SmartNICs from all leading vendors
Line-rate programmability to support evolving UEC (Ultra Ethernet Consortium) standards

Product Applications

AI Training Fabric
AI Inference Fabric
Data Center Fabric
Ethernet Storage Fabric
HPC (High Performance Computing)
Supercomputer

OVERVIEW Features SPECIFICATIONS Ordering Information

Overview

Engineered to meet the demands of AI training/inference, high-performance computing (HPC) and cloud computing/storage, the Sera DCS864E Open Ultra Ethernet Switch is a cutting-edge data center solution. Boasting a massive switching capacity of 51.2 T and port density of 64x800G, it supports ultra-scale data centers and delivers end-to-end performance on par with InfiniBand networks.
Compliant with UEC standards, it features comprehensive open APIs, facilitating seamless integration for data centers and HPC clusters. As a vendor-neutral network device, it is compatible with heterogeneous GPUs and network cards from various vendors. With industry-leading low latency and high reliability, the Sera DCS864E Open Ultra Ethernet Switch stands as the premier choice for data centers in the AI era.

Hardware

Industry-leading MARVELL TERALYNX 10 (TL10) ultra-low latency ASIC chips with 51.2 Tbps switching capacity
Up to 14.4 Bpps of L2/L3 full wire-speed forwarding
512 x 112 G Long Reach (LR) best-in-class SerDes, supporting 64 x 800 G, 128 x 400 G, or 512 x 100 G interfaces via optical modules/splitters
Load-aware per flowlet/packet balancing to prevent congestion and ensure efficient utilization of available bandwidth
Intel Xeon 8-core CPU for line-rate programmability, enabling future networking protocols through software upgrades
200+ MB of on-chip buffer, dynamically shared by every port, ensures superior network quality with fewer packet drops
Active queue management avoid traffic congestion by ECN
Support power and fan redundancy to ensure continuous business operation
Air flows in from the front panel, and exhausts from the rear panel

Networking

8 x CX864E support 512 GPUs interconnection with 400 G per port speed
192 X CX864E support 8192 GPUs interconnection with 400 G per port speed
192 X CX864E support 128K ML/AI nodes interconnection with 100 G per port speed
Support for all leading collective communication libraries, including NCCL, RCCL, oneCCL, BCL, MPI, and Gloo
Optimize collective operations (Allreduce, Allgather, All-to-all, Broadcast) with built-in multicast
Support various network topologies, including spine-leaf, fat-tree, dragonfly, butterfly, 3D mesh/torus, hyper-x and hypercube

Operating System

Pre-installed SeraNOS is an enterprise version of SONiC with SAI as the kernel
Integrated rich L2/L3 network features, complete support for network virtualization, QoS policies and other services
Advanced functional containerized & event-driven system architecture to accelerate network service development/customization
Provides an open REST API interface that can be centrally managed and invoked by third-party applications such as cloud management
platform
Provides KLISH command line for traditional network engineers

High Availability

1+1 hot-swappable power supplies and N+1 hot-swappable fans
BFD for BGP and OSPF in 50ms
BGP multi-homing for multiple connections to a server with automatic load balancing and failover
Multi-chassis link aggregation group (MC-LAG) for active/active L2 multipathing
Up to 8192-way equal-cost multi-path (ECMP) routing for load balancing and redundancy

Automated O&M

Integrated with Python and Ansible to support automated operation and maintenance
Supports ZTP which enables devices to automatically obtain and load deployment files

Features

Layer-2 Features

4k VLANs
VLAN-Based MAC learning
LAG/LACP
MAC dynamic learning
MAC priority
MAC migration
LLDP neighbor discovery and aging
LLDP custom TLV
STP
MSTP
IGMP
IGMP over VXLAN
IGMP Snooping
9K Jumbo frame

Layer-3 FeaturesIPv6 NDP

IPv6 ND proxy
Static ARP
Dynamic ARP
Gratuitous ARP
ARP proxy
Routing protocols: OSPF, BGP, MP-BGP
Static routes
8192-path Equal Cost Multipath Routing (ECMP)
ECMP elastic hash
DHCPv4/v6
VRF
Uplink and downlink association (monitor-link)

High Reliability

Containerized system
MC-LAG
BFD: BGP association, OSPF association
SLA
Track
ICPM-echo
BGP multi-homing

Security & Control

ACL for IPv4/v6
DSCP mapping, marking, classify
IACL, EACL
IPv4/v6 CoPP
Policy speed limited
TACACS+

QoS

Multi-Queue scheduling
Flow classification
DSCP mapping queue
Speed limit policy
Bandwidth guarantee
Tail drop
WRED
Data center network: PFC, ECN, ETS, DCBX
Congestion control: QCN, DCQCN, DCTCP
RoCEv2
Visibility
In-band network telemetry (INT)
SPAN
ERSPAN
Detailed telemetry on packet delays (HDC), packet drops (BDC) and path traversed by a packet (IPT)
Time Synchronization
PTP (IEEE 1588v2)
NTP

System Management

CLI, WEBUI, controller
DevOps: Ansible, Python
Out-of-Band management interface
System information monitoring
ONIE install
ZTP
SONiC upgrade
Warm reboot
Fast reboot
SNMP
Syslog
REST API
gRPC
CRM

SPECIFICATIONS

Product Model		DCS864E
Network Interface	800 GE(OSFP)	64
	400 GE(QSFP-DD)	128
	200 GE(QSFP56)	–
	100 GE(QSFP28)	512
Switching Chip	Chip Model	MARVELL TERALYNX 10
Switching Chip	Switching Capacity	51.2 Tbps
Forwarding Performance	Packet Forwarding Rate	14.4 Bpps of L2/L3 forwarding
	Cut-through latency	560ns
	Packet Buffer	200+ MB
Control and Manage CPU (COMe)	CPU	Intel Xeon 8-core
	RAM	32 GB SODIMM
	SSD	256 GB M.2 SATA
Management Interface	USB	1 x USB 2.0
	Console	1 x Console RJ45
	MGMT	1 x MGMT GE RJ45
Electrical Characteristics	Fan Redundancy (Hot-pluggable)	3+1
	Power Redundancy (Hot-pluggable)	1+1
	Input Voltage	100 – 240 V AC
	Input Voltage	HVDC 180 V~300 V
	Maximum Power Consumption	2200 W (with full ports of 800 G-SR8)
Dimensions	Height	2 U
Dimensions	Dimensions (WxHxD, mm)	440 x 87 x 660
Operating Conditions	Operating Temperature	0 – 40℃
Operating Conditions	Relative humidity	10% – 90% (non-condensing)

Ordering Information

Part Number	Description
DCS864E	Ultra Ethernet Switch, 64 x 800 G OSFP, hot-pluggable 1+1 PSU, hot-pluggable 3+1 FANs, front-to-power airflow, 2 RU height