How do SONiC and RoCE mitigate network congestion for AI workloads?

Name: PowerSwitch SN5610
Brand: Dell
SKU: powerswitch-sn5610

To mitigate network congestion for demanding, synchronized AI workloads, SONiC and RoCEv2 work together to create a multi-layered, end-to-end congestion defense system. Because AI workloads generate massive, synchronized 'elephant flows' that require lossless data transmission, these technologies utilize mechanisms like DCQCN, Priority Flow Control (PFC), and Dynamic Load Balancing to prevent packet drops and minimize latency.

Dell Networking Dell PowerSwitch SN5610

2U air-cooled switches offering 64 ports of 800GbE.

Spine800G FabricBalanced AI

Talk to a Pro

SKU: powerswitch-sn5610

Ports: 64x 800 GbE OSFP, 2x 25 GbE SFP28
Capacity: 51.2 Tbps
NPU: NVIDIA Spectrum™-4

Download Datasheet (PDF)

Better pricing available. Our experts can help you secure the lowest rate on this model.

Request a Quote

Fast shipping — same-day processing before 4:15 pm PST
Authorized Reseller — Serving customers 20+ years
Call for Bulk Pricing

Overview of AI Switches

High performance, 400-800Gb/s networking designed for massive-scale AI.

High-performance switches with low latency and high throughput for deploying advanced AI clusters effectively.

Best For:

Large-scale AI & machine learning clusters
Hyperscale data center environments
Massive east-west traffic architectures

Powerful fabrics for Al workloads

Transform your Al infrastructure with ultra-fast 800 Gbps networking, seamless InfiniBand and Ethernet connectivity, advanced congestion control, and dynamic load balancing powered by SONiC. Scale effortlessly and accelerate Al innovation with future-ready, high-performance solutions built for demanding workloads.

Overview of PowerSwitch SN5610

PowerSwitch SN5610 Key Specifications

Built for hyperscale environments, this switch supports massive data movement and large AI clusters, enabling high-speed, low-latency performance across complex, distributed architectures.

NPU

NVIDIA Spectrum™-4

Switch Capacity

51.2 Tbps

Forwarding

33.3 Bpps

Ports

64x 800 GbE OSFP, 2x 25 GbE SFP28

Ideal Workloads

AI & Machine Learning Fabrics: Purpose-built to handle the intense, non-blocking traffic requirements of GPU clusters training large language models (LLMs) and trillion-parameter generative AI.

AI Fabric Ready

Utilizes Enterprise SONiC and RoCEv2 for end-to-end congestion management, reducing P99 tail latency by up to 77% for lossless AI workloads. Read our "Building Generative AI Fabrics: Dell PowerSwitch (Ethernet) vs. NVIDIA InfiniBand (Quantum)" deep dive

Not sure if this is the right model?

If you're comparing PowerSwitch SN5610 vs PowerSwitch SN6800-LD or PowerSwitch SN6810-LD — This is designed for hyperscale — choose this only if you're running multi-rack AI training or massive east-west traffic.

View full comparison

Features

Equip your Al deployments with open, high-bandwidth Ethernet fabrics that deliver nonblocking performance, low latency, and seamless scalability for GPU clusters.

Request a Quote

Flexible Al networking

Equip your Al deployments with open, high-bandwidth Ethernet fabrics that deliver nonblocking performance, low latency, and seamless scalability for GPU clusters.

Seamless transformation

Modernize enterprise IT and drive efficiency with Al adoption. Upgrade networks for speed, throughput, and low latency to support advanced Al capabilities effectively.

Efficient innovation

Harness advanced features like RoCEv2, enhanced hashing, and optimized throughput to empower sustainable, high-performance Al deployments.

Future-ready Al fabrics

Build infrastructure that evolves with your business, featuring PowerSwitch Z-series capable of supporting massive data transfers and next-gen workloads at up to 800 GbE speeds.

Common Mistakes When Choosing an AI Switch

Choosing based on specs instead of workload
Fix: Start with your environment. Hyperscale switches are designed for massive east-west traffic. Make sure your workload actually demands this level of throughput.
Also consider: PowerSwitch SN6800-LD, PowerSwitch SN6810-LD
Overbuilding for AI when running primarily VMware
Fix: Focus on balanced performance and scalability instead of ultra-high throughput.
Better fit: PowerSwitch SN5400, PowerSwitch Z9664F-ON
Deploying hyperscale where mid-scale is sufficient
Fix: Not every rack needs 800 GbE. Use hyperscale at the spine and right-size your leaf switches.
For leaf / ToR: PowerSwitch SN5400, PowerSwitch Z9664F-ON

Technical Details

Hardware Specifications

Complete technical specifications for the PowerSwitch SN5610.

PERFORMANCE

NPUNVIDIA Spectrum™-4

Switch Capacity51.2 Tbps

Forwarding Capacity33.3 Bpps

Latency<850 ns

CONNECTIVITY

Ports64x 800 GbE OSFP, 2x 25 GbE SFP28

Buffer Size160 MB

For full specifications, please reference the datasheet (PDF).

Frequently Asked Questions

Expand All | Collapse All

Dell offers high-density Ethernet up to 800 GbE (e.g., PowerSwitch Z9864F-ON with 64x 800 GbE) and 400 GbE options (e.g., Z9664F-ON with 64x 400 GbE). For InfiniBand, NVIDIA Quantum-X800 delivers 800 Gb/s per port with up to 144 ports per chassis, while Quantum-2 provides 400 Gb/s. This gives you the flexibility to scale training and inference fabrics as your GPU footprint grows.

Yes. Dell AI Ethernet switches support RoCEv2 and advanced congestion control features designed for consistent, low-latency performance across GPU clusters. Enhanced hashing and optimized throughput help maintain stable job completion times under load.

The PowerSwitch Z9864F-ON delivers sub-microsecond latency (<700 ns), with nonblocking throughput across 64x 800 GbE ports. This performance helps minimize step times in distributed training and reduces fabric-induced stalls.

Yes. Many environments use Ethernet for the broader data center and storage access, and InfiniBand for dedicated training clusters. Dell supports both approaches, and you can interoperate via gateways or data pipeline design while maintaining separate, optimized fabrics.

Start with nonblocking leaf-spine tiers sized to your GPU node counts and east-west traffic. Use 400/800 GbE on leafs and spines with balanced uplinks, and follow validated designs for hashing, buffer tuning, and RoCEv2 settings. Dell publishes reference architectures to accelerate and de-risk deployments.

800 GbE platforms like Z9864F-ON use OSFP112 ports, while 400 GbE systems like Z9664F-ON use QSFP56-DD. Your choice of DACs, AOCs, or optics will depend on distance and power budgets, and Dell provides qualified transceivers and cables for each platform.

Yes. Options include NVIDIA Spectrum SN2201 (48x 1G BASE-T with QSFP28 uplinks) and Dell PowerSwitch S3248T-ON (48x 1G with higher-speed uplinks). Dedicated OOB networks improve reliability, upgrade safety, and troubleshooting for production fabrics.

Dell AI switches are validated with PowerEdge AI servers and PowerScale AI storage to maintain high throughput and predictable latency end-to-end. You can use published reference designs to size links, choose topology, and align storage bandwidth to GPU demand.

Yes. Dell Design Services for AI Networking tailors the fabric to your models, GPU counts, and growth plan, and ProDeploy accelerates rollout with best practices. ProSupport Infrastructure Suite delivers 24/7 assistance to keep your AI network performing at its best.

Yes. PowerSwitch Z9864F-ON earned TMC's 2025 Unified Communications Product of the Year, with Z9664F-ON and Z9432F-ON also recognized in prior years. Independent industry accolades reflect performance and innovation for high-capacity data center networking.

Product Comparison

Compare Ethernet switches

See how the PowerSwitch SN5610 stacks up against other models in the lineup.

Model

NPU

Capacity

Ports

PowerSwitch SN6800-LD

NVIDIA Spectrum™-6

409.6 Tbps

512 x MMC-12

PowerSwitch SN6810-LD

NVIDIA Spectrum™-6

102.4 Tbps

128 x MMC-12

PowerSwitch SN6600-LD

NVIDIA Spectrum™-6

102.4 Tbps

64 x OSFP224

PowerSwitch SN5610 Current

NVIDIA Spectrum™-4

51.2 Tbps

64x 800 GbE OSFP, 2x 25 GbE SFP28

PowerSwitch SN5600D

NVIDIA Spectrum™-4

51.2 Tbps

64x 800 GbE OSFP, 1x 25 GbE SFP28

PowerSwitch SN5400

NVIDIA Spectrum™-4

25.6 Tbps

64x 400 GbE QSFP-DD, 2x 25GbE SFP28

PowerSwitch Z9864F-ON

Broadcom Tomahawk™ 5

51.2 Tbps

64x 800 GbE OSFP112, 2x SFP+

PowerSwitch Z9664F-ON

Broadcom Tomahawk™ 4

25.6 Tbps

64x 400 GbE QSFP56-DD, 2x 10 GbE SFP+

PowerSwitch Z9432F-ON

Broadcom Trident4-X11™

12.8 Tbps

32x 400 GbE QSFP56-DD, 2x 10 GbE SFP+

PowerSwitch SN4700

NVIDIA Spectrum™-3

12.8 Tbps

32x400 GbE QSFP-DD

InfiniBand Switches

High-performance, high-density multi-rate aggregation and AI fabric switches.

Explore InfiniBand Switches

Out-of-Band Management

Dedicated switches for management, orchestration, and troubleshooting of production infrastructure.

Explore Out-of-Band Management

Downloads

Documentation & Resources

Access datasheets and technical resources for the PowerSwitch SN5610.

Hardware Spec Sheet

Complete hardware specifications, port configurations, and performance data.

Download PDF

Z9864F-ON vs. Quantum-X800

Head-to-head technical assessment of 800GbE Ethernet vs. NDR InfiniBand.

Read Assessment

Building AI Fabrics

Deep dive into DCQCN, PFC, SHARP, and 800G fabric technologies.

Read Guide

Scaling to 800G

Strategic blueprint for trillion-parameter AI: topology, CPO, and roadmap.

Read Blueprint

Need Help?

Our specialists can help you find the right configuration.

Get a Quote

Request competitive pricing on the PowerSwitch SN5610.

REQUEST QUOTE

Call Us

Talk to a networking specialist right now.

833-335-0426

Related Models

Also Consider

PowerSwitch SN6800-LD

Core Multi-ASIC Hyperscale AI

NPU: NVIDIA Spectrum™-6
Capacity: 409.6 Tbps
Ports: 512 x MMC-12

Get a Quote More Details

PowerSwitch SN6810-LD

Spine CPO Optics High Efficiency

NPU: NVIDIA Spectrum™-6
Capacity: 102.4 Tbps
Ports: 128 x MMC-12

Get a Quote More Details

PowerSwitch SN5400

Leaf 400G Fabric AI Clusters

NPU: NVIDIA Spectrum™-4
Capacity: 25.6 Tbps
Ports: 64x 400 GbE QSFP-DD, 2x 25GbE SFP28

Get a Quote More Details

Dell Networking Dell PowerSwitch SN5610

High performance, 400-800Gb/s networking designed for massive-scale AI.

Powerful fabrics for Al workloads

PowerSwitch SN5610 Key Specifications

Ideal Workloads

AI Fabric Ready

Not sure if this is the right model?

Common Mistakes When Choosing an AI Switch

Hardware Specifications

PERFORMANCE

CONNECTIVITY

Frequently Asked Questions

What speeds and port densities do Dell AI switches support?

Do these switches support RoCEv2 for AI/ML fabrics?

How low is the latency on 800 GbE?

Can I mix Ethernet and InfiniBand in one AI environment?

How do I design a leaf-spine for AI training?

Which cables and optics are supported at 800 GbE and 400 GbE?

Do you offer out-of-band (OOB) management switches?

How does this integrate with my servers and storage for AI?

Do you provide services to design and deploy AI fabrics?

Are there third-party reviews or awards I can reference?

Compare Ethernet switches

InfiniBand Switches

Out-of-Band Management

Documentation & Resources

Also Consider

PowerSwitch SN6800-LD

PowerSwitch SN6810-LD

PowerSwitch SN5400