DriveNets Network Cloud-AI​

With the fast growth of AI workloads, network solutions used in the fabric of AI clusters need to evolve to maximize the utilization of costly AI resources and support standard connectivity that enables vendor interoperability.​

DriveNets Network Cloud-AI which is based on the largest-scale DDC (Distributed Disaggregated Chassis) architectures in the world, provides predictable, lossless back-end cluster connectivity and 10%-30% improvement in JCT (Job Completion Time) of high-scale, high-performance AI workloads. It maintains a GPU, ASIC and ODM agnostic architecture and supports over 100% ROI at day-1.

Performance at Scale​

Up to 32K GPUs per cluster (100G-800G ports)​
Performance validated in trials by hyperscalers​
10-30% improved JCT performance: may lead to 100% system ROI

Trusted Solution​

Based on a certified OCP DDC (Distributed Disaggregated Chassis) architecture​
Proven in the world’s Largest DDC Networks​
Well-Known protocol – 600 million Ethernet ports per year

Architectural Flexibility​

Ethernet-based - enables vendor interoperability​
GPU, ASIC & ODM agnostic​
Seamless internet connectivity

Description

To best support AI workloads, new networking solutions will need to reduce AI implementation costs, enable greater throughput and 100% utilization, and support Ethernet-based implementations that enable network interoperability and multivendor support.

Early trials by some of the world’s most prominent hyperscalers have proven Network Cloud for AI’s ability to achieve a 30% improvement in JCT (Job Completion Time) over all other industry Ethernet solutions. Network Cloud-AI is the industry’s only solution capable of delivering the high performance of a proprietary solution with a standards-based Ethernet implementation; providing unrivaled performance while avoiding vendor lock-in.

  • Best JCT performance
  • Up to 32K GPUs (100-800Gbps) per cluster
  • Scheduled fabric – lossless connectivity, low latency & jitter
  • Highest bisectional bandwidth
  • Seamless failover and fastest recovery​
  • No vendor lock
  • Support diverse Apps

Network Cloud-AI vs. The Alternatives

Network Cloud-AI is a Trusted Solution

  • Well-Known Protocol – 600M Ethernet Ports Per Year.

Network Cloud-AI has Architectural Flexibility

  • GPU, ASIC & ODM agnostic
  • One technology for front-end and back-end networking

Network Cloud-AI shows Performance at Scale

  • Up to 32K GPUs per Cluster (100G-800G ports) ​
  • Up to 30% higher performance in AI training​
  • No performance impact on network convergence (lossless, seamless failover)
  • Cell-based fabric – single-hop GPU communication and scalable QoS (optimal packet-loss, latency, and jitter performance)

Network Cloud-AI shows Performance at Scale

  • Up to 32K GPUs (100G-800G ports) per Cluster
  • Scale with no performance impact

Network Cloud-AI has Architectural Flexibility

  • GPU, ASIC & ODM agnostic

Trusted Solution

  • Powering the world’s largest DDC Network (>52% of AT&T’s backbone)​
  • Performance proven by hyperscalers​

Network Cloud-AI has Architectural Flexibility​

  • Open and disaggregated – Based on a certified OCP architecture ​
  • GPU, ASIC & ODM agnostic
Related Content
The Next Network Bottleneck – Part 1

The network as a bottleneck that slows down applications and prevents new use cases (and business) from becoming feasible seems...

Read more
The Compute-Networking Gap: The Next Network Bottleneck – Part 2

In our last blog, we talked about external and internal I/O mechanisms becoming interchangeable as they offer (roughly) the same...

Read more
AI Back-End Network Bottleneck Resolved by Network Cloud-AI: The Next Network Bottleneck – Part 3

As established in our previous blog post, the AI back-end network has some unique requirements derived from its: huge scale...

Read more