CloudEngine 16800: Built for AI

Unleashing Computing Power with New Ethernet

Entering a fully connected, intelligent world, data centers are shifting from the cloud era into the age of intelligence, powering nascent technologies such as 5G and Artificial Intelligence (AI).

Stepping into this arena, Huawei CloudEngine 16800 is a next generation series of 400 GE data center switches, purpose built for intelligence. Featuring 400 GE ultra-broadband, lossless Ethernet, and full-lifecycle automated management, CloudEngine 16800 revitalizes digital transformation across industries.

More Specifications

400 GE Ultra-Broadband Unleashes Mass Computing Power

With an upgraded hardware switching platform, orthogonal architecture, and no backplane, CloudEngine 16800 breaks through multiple technical challenges, from high-speed signal transmission to heat dissipation and power supply. It features a high-density 48-port 400 GE line card per slot and 768-port 400 GE switching capacity — five times the industry average — meeting the exponential growth of traffic in the intelligent era. In addition, power consumption per bit is reduced by 50%, ensuring greener operations.

Eighty 36 x 100G line cards

100G platform switch in the cloud era
5x
Sixteen 48 x 400G line cards

CloudEngine 16800

400 GE optical module

SuperFast: Submicron smooth copper foil

SuperFast: Ultra-low-loss intra-board cable

SuperPower: Highly efficient power supply

SuperCooling: Board-level phase-change heat dissipation

SuperCooling: Device-level mixed-flow fan

400 GE Quad Small Form Factor Pluggable-Double Density (QSFP-DD) optical modules use Pulse Amplitude Modulation 4-level (PAM4) signals for high-speed data transmission. Compared with 100GE Non-Return to Zero (NRZ) signals, PAM4 signals have four different signal levels and can transmit 2-bit information in each clock cycle, doubling transmission capability. In addition, the amplitude of PAM4 signals is just one third of NRZ signals.

Watch Video

Huawei uses submicron smooth copper foil that is processed by a special electroplating technique. This material has a roughness of just 0.3 microns, five times better than the industry's next best. This future-oriented material improves the transmission efficiency of electrical signals by 30%, laying a solid foundation for high-density 400 GE transmission and smooth evolution to 800 GE.

Watch Video

The cross-sectional circumference of the intra-board cable is twice that of a Printed Circuit Board (PCB) , providing 70% lower signal loss. Port density is increased per unit space through optical module mirroring. All of this builds a 48 x 400 GE Line Processing Unit (LPU) with the highest port density in the industry.

Watch Video

Magnetic blow-out and large exciter technologies are used to realize the fast switching of two independent power inputs of a single power module within milliseconds. The power modules support intelligent switching between two power inputs, significantly saving space in the equipment room and improving power supply efficiency by an astonishing 95%.

Watch Video

A new carbon nanotube thermal pad and Vapor Chamber (VC) phase-change radiator enable four times higher heat dissipation efficiency than the industry's next best, alongside 20% higher device reliability.

Watch Video

An innovative mixed-flow fan design and magnetic permeability motor deliver higher air volumes with less noise, also consuming 50% less power than traditional fans. That represents a saving of 320,000 kWh of electricity each year, equivalent to saving approximately US$40,000 in electricity costs and cutting carbon emissions by 250 tonnes.

Watch Video

Lossless Ethernet Unleashes 100% of AI Computing Power

CloudEngine 16800 is a great choice for building lossless networks with maximum throughput and minimum latency. This is achieved thanks to Huawei's innovative iLossless algorithm that:
• Enables automatic adaptation and optimization of traffic models.
• Unleashes 100% of AI computing power, compared to 50% over traditional Ethernet.
• Improves data storage Input/Output Operations Per Second (IOPS) by 30%.

27%

Data calculation efficiency

375/s

478/s

(Number of AI training samples [iteration/s])
30%

Data calculation efficiency

1256/s

11,670/s

(Storage IOPS)

Full-Lifecycle Automated Management Enables an L3 ADN for Data Centers

With CloudEngine 16800 and iMaster NCE- Fabric — Huawei's intelligent network management and control system — Huawei has successfully built an L3 Autonomous Driving Network (ADN) for data centers, the first in the industry that the independent assessment organization, the Tolly Group, has ever verified. This future-proofed network achieves service provisioning in seconds, implementing error-free network changes. Huawei has also innovated knowledge graphs for network Operations and Maintenance (O&M), allowing networks to rapidly locate faults in 3 minutes, as well as proactively predict up to 90% of risks.

SDN

Semi-automated

Network design: Expert experience-dependent + tool-assisted.
Configuration change: Manual comparison.
Troubleshooting: Expert experience-dependent (location) + manual rectification.
ADN

Fully automated

Intent-driven network recommendation, enabling zero-touch service rollout.
Pre-event simulation + post-event verification, enabling error-free network changes.
Network self-optimization, self-healing, and self-governance, eliminating service interruptions.

Resources

Login

CloudEngine 16800

Unleashing Computing Power with New Ethernet

400 GE Ultra-Broadband Unleashes Mass Computing Power

Eighty 36 x 100G line cards

Sixteen 48 x 400G line cards

Lossless Ethernet Unleashes 100% of AI Computing Power

27%

30%

Full-Lifecycle Automated Management Enables an L3 ADN for Data Centers