Cisco and Nvidia build Ethernet-based AI infrastructure

Cisco and Nvidia build Ethernet-based AI infrastructure

Cisco and Nvidia have big plans when it comes to building an AI infrastructure for organizations. At least that’s the message both companies broadcast today at Cisco Live EMEA. Of particular note is that this is an Ethernet-based infrastructure.

If an organization wants to get started with AI, it must have an infrastructure that can handle it. Typically, existing infrastructure will not suffice. Cisco and Nvidia today announced plans to provide this AI infrastructure. In doing so, the companies promise easy deployment and management.

Ethernet is the foundation

What is most striking to us about today’s news is that this is an infrastructure based on Ethernet. This is notable because in 2019 Nvidia bought Mellanox, at the time the last independent provider of InfiniBand. That had long been the network fabric for supercomputers and was seen as superior to Ethernet in terms of performance. Today, that is no longer necessarily the case. Ethernet has been on the rise for just under a decade. The partnership announced by both parties today seems to confirm this. The fact that Cisco emphasizes several times that this is an Ethernet-based infrastructure indicates that the company is quite chuffed with this.

Cisco AI-infrastructuur.

The partnership between Cisco and Nvidia

The partnership between Cisco and Nvidia consists of a few components. First, Nvidia’s Tensor Core GPUs are available in Cisco’s M7 generation of UCS rack and blade servers. This also includes the Cisco UCS X-Series and UCS X-Series Direct. Furthermore, starting today, Cisco is putting Nvidia AI Enterprise on its price list. That is, Cisco will be selling this software platform from Nvidia, called “the OS for AI” by Nvidia, along with its own hardware.

Finally, there are Cisco Validated Designs (CVDs) that Nvidia also supports. These should make it easier for organizations to deploy AI clusters, regardless of scale. Cisco is going to offer both converged and hyperconverged options here. CVDs are already becoming available this month for FlexPod (NetApp) and FlashStack (Pure Storage) focused on GenAI inferencing using Nvidia AI Enterprise. More will follow, Cisco promises us.

Cisco hasd had a partnership with Nvidia for some time. With today’s announcements, this enters a new phase. Whether Nvidia is already giving InfiniBand a (small) kiss of death with this is not yet definite. However, it is definitely a sign of things to come if the largest networking vendor and the largest AI hardware vendor are piecing this partnership together. We wouldn’t invest too much more in InfiniBand in any case.

Also read: Nvidia dominates AI market, where are the limits to its growth?