Inside Juniper’s Ops4AI Lab: Powering the Future of AI Network Fabrics

0:00 AI is reshaping the world and data

0:03 centers are its foundation. But

0:05 designing AI infrastructure that's open,

0:08 scalable, and simple to manage, that's

0:10 no small task. That's why HP and Juniper

0:12 Networking has made a multi-million

0:14 dollar investment in the Ops for AI lab,

0:17 the industry's first lab dedicated to AI

0:20 R&D, operational innovations, and open

0:23 validated infrastructure design to

0:25 empower customers and partners with a

0:27 trusted environment to test and

0:29 fine-tune their AI models. The Ops for

0:31 AI lab combines Juniper's cutting edge

0:34 AI native networking solutions,

0:36 industryleading AI models, and a

0:39 powerful ecosystem of technology

0:40 partners to solve one critical

0:42 challenge. How to build AI data centers

0:45 that are open, scale efficiently,

0:48 deliver consistent performance, and

0:50 remain simple to operate. At the heart

0:52 of the AI lab is a production scale

0:54 topology leveraging QFX series switches

0:57 including the 800 gig capable QFX 5240

1:01 running AI optimized Junos OS software

1:03 and orchestrated by Juniper Appstster

1:05 data center director and data center

1:07 assurance for intentbased automation and

1:10 AI ops across multi- vendor

1:12 environments. gain real-time visibility

1:15 into the flows between GPUs and take

1:18 action when needed. To accelerate AI

1:20 deployments, we created Juniper

1:23 validated designs to support distributed

1:25 AI training and inference clusters, GPU

1:28 as a service platforms and hybrid AI

1:30 clouds, making them perfect for

1:32 enterprise and hypers scale

1:34 environments. We're collaborating with

1:36 leading partners to validate

1:38 performance, test real a IML models, and

1:41 design future ready architectures that

1:43 scale with your business. Today, our lab

1:46 hosts over a 100 GPUs, a mix of Nvidia

1:49 and AMD, including Nvidia DGX. The

1:52 Nvidia based systems are running with a

1:54 combination of ConnectX6 and Connect X7

1:57 Nick. In addition to that, we're also

1:59 utilizing a Ksite Aries 1S Rocky VDU

2:03 traffic generator to simulate RDMA

2:06 traffic. For storage, we have a Wacka

2:09 cluster and a BAS storage cluster. From

2:11 LLM inference to image classification

2:14 and self-supervised learning, we test

2:16 how real world AI workloads interact

2:19 with the network to ensure your

2:20 infrastructure flawlessly to deliver

2:23 optimal performance for your AI

2:25 workloads. With the ops for AI lab,

2:28 Juniper offers AI native operations,

2:30 automated life cycle management, and

2:32 telemetry rich visibility so you can

2:34 focus on results, not complexity.

2:37 Purpose-built AI infrastructure

2:39 validated by Juniper, ready for what's

2:42 next.