UPDATED 10:21 EDT / SEPTEMBER 15 2020

INFRA

Lenovo, Nvidia partnership bridges HPC and enterprise AI with switches for optimized networking

Artificial intelligence is fast becoming a part of the everyday enterprise workflow, but the computing infrastructure to support such a data-intense task must modernize. As businesses transform to better leverage data intelligence and become more agile through cloud-native processes, high-performance networking becomes priority. But investing in the InfiniBand standard for high-performance computing network switches has been a hard sell for information-technology departments with an existing Ethernet fabric in place.

Enabling enterprise to catch the fast train to intelligent business operations are long-time partners Nvidia Corp. and Lenovo Group Ltd.

“We love, from an HPC perspective, to use InfiniBand,” said Scott Tease (pictured, right), general manager of HPC and AI at Lenovo. “But most enterprise clients are using Ethernet. So where do we go? We go to a partner that we’ve trusted for a very long time. And we selected the Nvidia Mellanox Ethernet switch family.”

Tease and Kevin Deierling (pictured, left), senior vice president of marketing at Nvidia, spoke with Stu Miniman, co-host of theCUBE, SiliconANGLE’s mobile livestreaming studio, for a digital CUBE Conversation on how intelligent workloads are driving the need for high-performance Ethernet switches for enterprise customers. (* Disclosure below.)

Ethernet speeds jump 40,000x in 40 years

It’s been 40 years since the first specification of Ethernet standards was published. At that time, high speed was defined as “a data rate of 10 megabits per second.” In contrast, today’s switches are designed to handle speeds up to 400GbE. “We’re about 40,000 times faster,” Tease stated.

Another change is a focus on GPU and networking performance rather than the CPU. “Trying to design a platform that’s solely based on a CPU and then jam these other items on top of it — it no longer works,” Tease said. “You have to design these systems in a holistic manner, where you’re designing for the GPU, you’re designing for the network.”

Lenovo is the world’s leading supercomputer provider. By integrating Nvidia’s Mellanox Spectrum-3 SN4000 Open Ethernet Switches into its solutions, the company brings the “rock-solid” interoperability and pre-tested capability of its InfiniBand switch to enterprise AI workloads.

“We’ve shown in HPC that the days of just taking an Ethernet card or an InfiniBand card, plugging it in the system, and having it work properly are gone,” Tease stated. “You really need a system that’s engineered for whatever task the customer is going to use.”

Nvidia invested heavily in the software ecosystem that’s built on top of the GPU and the networking, including it’s just-announced acquistion of Arm Holdings Ltd., according to Deierling. “By integrating all of that together on a platform, we can really accelerate the time to market for enterprises that want to leverage these modern workloads,” he said.

Regarding the engineering side of the solution, Tease said: “We can do all that upfront engineering to make sure that the platform, the systems; the solution as a whole works exactly how the customer is going to expect it to work,” he said. “When we’re selling these solutions, like an SAP solution, for instance, the customer is not buying a server anymore, they’re buying a solution, they’re buying a functionality. … So any of the systems that are going to be coming from us, pre-configured, pre-tested, are all going to have Nvidia networking inside of them.”

Meeting the demands of AI requires engineered solutions

The partnership in enterprise solutions is a natural extension of Lenovo and Nvidia’s work together in supercomputing. Lenovo was the first to water-cool an InfiniBand card and one of the first companies to deploy the highly scalable networking method of dragonfly topology.

“We’re looking forward to doing a lot of that same exact kind of innovation inside of our systems as we look to Ethernet,” Tease stated. He predicts a customer shift from InfiniBand to Ethernet as switch speeds continue to increase.

“Having both of these offerings inside of our lineup is going to make it really easy for customers to choose what’s best for them over time,” Tease added.

The demand for “different networks for different workloads,” will continue, according to Deierling. He predicts the rise of two different cases: One using Ethernet with an InfiniBand network and a second for new AI-centric, cloud native enterprise workloads.

“You have all of the infrastructure in place with our Spectrum ASICs, our ConnectX adapters, and now integrated with GPUs, that we’ll be able to deliver solutions rather than just complements,” Deierling said.

While the overall Ethernet switch market showed a small decline in 1Q20, due to the COVID-19 pandemic, the highest speed switching platforms saw growth. According to industry analysis by International Data Corp., 100Gb revenues increased to $1.28 billion in 1Q20. This was a 9.9% year over year increase, with high-speed switch sales accounting for 20.8% of market revenue.

Those higher speeds are required to keep up with the massive amounts of data crunched by modern GPUs and AI workloads.

“Where you need that networking, like we’ve been used to an HPC, is when you start looking at deep learning in training, scale out training,” Tease stated.

Without an optimized network, companies find themselves restricted to running these workloads on a single workstation. “They haven’t been able to figure out how to spread that workload out and chop it up, like we’ve been doing in HPC, because they’ve been running into networking issues,” Tease said.

The goal of Nvidia’s new Ethernet switches is to bring enterprise customers the same speeds and flexibility that HPC customers have taken for granted. “Workload management, distribution of workload, chopping jobs up into smaller portions, and feeding them out to a cluster … the same kind of thing we’ve been doing forever with Mellanox in the past, now Nvidia networking, we’re just going to take that to the enterprise,” Tease said.

“I think it’s a proven set of components that together forms a solution,” Deierling added. “That’s the real key … delivering a solution, not just piece-parts. We have a platform — that software, hardware, all of it integrated.”

Watch the complete video interview below, and be sure to check out more of SiliconANGLE’s and theCUBE’s CUBE Conversations. (* Disclosure: Nvidia Corp. sponsored this segment of theCUBE. Neither Nvidia nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE

A message from John Furrier, co-founder of SiliconANGLE:

Your vote of support is important to us and it helps us keep the content FREE.

One click below supports our mission to provide free, deep, and relevant content.  

Join our community on YouTube

Join the community that includes more than 15,000 #CubeAlumni experts, including Amazon.com CEO Andy Jassy, Dell Technologies founder and CEO Michael Dell, Intel CEO Pat Gelsinger, and many more luminaries and experts.

“TheCUBE is an important partner to the industry. You guys really are a part of our events and we really appreciate you coming and I know people appreciate the content you create as well” – Andy Jassy

THANK YOU