Dell Booth SC25
– Dan Meyer

ST. LOUIS – Dell Technologies added support for Nvidia’s Spectrum-X Ethernet platforms in a move that provides enterprise customers with optionality as they move to integrate AI and high-performance computing (HPC) into operations.

The support integrates Nvidia’s Spectrum-X into Dell’s broader Enterprise SONiC Distribution platform, which is the vendor’s open-source operating system based on the Software for Open Networking in the Cloud (SONiC).

Vamshi Doma, senior director for Dell Networking product management, explained to SDxCentral at this week’s SC25 event that the move was based on attempting to provide some differentiation in the market.

“The network administrators are not typically used to operating a network like a Linux shell, so we saw that as an opportunity to differentiate,” Doma said. “Nvidia has been an early adopter for SONiC and they've been a heavy contributor to SONiC, but in order to focus on what they're trying to do with AI, they've taken a step back there … so we saw this as an opportunity.”

Doma was quick to note that Dell’s integration of Spectrum-X is being done in partnership with Nvidia and “we’re not doing it in a silo.”

Dell was one of the initial partners that worked with Microsoft on the development of SONiC nearly 10 years ago. Dell rolled out its enterprise distribution platform in mid-2020, a move that further integrated SONiC into Dell’s PowerSwitch Open Networking hardware with commercial-grade services and technical support.

Doma explained that this history paved a natural progression toward the latest integration.

“For a lot of these enterprise customers, especially those who are looking to make that journey to SONiC, we already offered SONiC for our core data center product portfolio and if customers were looking at an end-to-end SONiC-based solution, it'd be great if we can offer that diversity, silicon diversity, so that's what our new launch of SONiC for Spectrum-X allows us to do,” Doma said.

Nvidia’s Spectrum-X has become a powerful and lucrative networking fabric for the vendor. It sprang out of Nvidia’s $6.9 billion purchase of Mellanox Technologies in 2019, and has become Nvidia’s push toward an end-to-end accelerated networking platform designed to improve the performance and efficiency of Ethernet-based AI clusters.

SmartFabric Manager updates and certainty

Dell’s Spectrum-X support also extends to the vendor’s SmartFabric Manager. This includes the development of blueprints from its AI Factory to help automate deployments, and the expansion of that into Dell’s rack-scale systems.

“We want to be able to offer the same capabilities consistently across the entire rack,” Doma said of the extension, citing use cases like Dell’s liquid cooling system all the way down to the GPU level.

“The rack controller will have a clear understanding of which nodes are connected to which rack,” Doma said. “The fabric knowledge is shared across into the rack controller … as well as compute and the GPU locations, the fabric can discover that. When we are displaying flows, we should be able to see an end-to-end flow, not just from a switch to switch, but all the way from GPU to GPU. So that integration kind of allows that.”

Dell’s latest work also builds on its latest AI Factory work. Those efforts took an Nvidia turn earlier this year when Dell added a handful of new integrations tied to various Nvidia compute, networking, and storage taps.

Doma noted that these Nvidia-based moves are based on meeting enterprise needs, which are increasingly around long-term networking certainty.

“[Enterprises] making significant investment in the GPU cluster, so they want to make sure they follow the GPU cluster vendor’s recommendation on the networking because they want to hold that vendor accountable for end-to-end,” Doma explained. “If I'm an enterprise and I don't have the skill set to do end-to-end troubleshooting and diagnosis, or if I'm mixing and matching and building my own thing, then I'm exposed, and that's really the inflection point.”