Five months after closing its $7 billion purchase of networking interconnect and chip developer Mellanox Technologies, Nvidia (NVDA) – Get Report is teaming with VMware (VMW) – Get Report to expand the capabilities of Mellanox’s silicon.
On Tuesday — the start of VMware’s annual VMworld conference — Nvidia and VMware announced a pair of partnerships.
One of the partnerships involves the Nvidia GPU Cloud (NGC), a set of software tools and pre-trained deep learning models meant to help developers quickly get AI projects relying on Nvidia GPUs up and running. The other involves Mellanox’s SmartNICs, networking cards that (with the help of Mellanox’s BlueField data processing units, or DPUs) can offload networking, security and storage processing from a server’s CPUs.
The second partnership, arguably the more important of the two, will allow many of the tasks handled by VMware’s core Cloud Foundation platform — it covers VMware’s server, storage and networking virtualization software — to be offloaded to Mellanox SmartNICs.
VMware’s vSphere/ESXi server virtualization platform will still require its compute hypervisor (the software responsible for creating and running server virtual machines) to run on CPUs. However, SmartNICs from Mellanox and others will be able to offload the networking and security services delivered by VMWare’s NSX networking virtualization software, as well as data management functions for VMware’s VSAN storage virtualization software and ESXi’s host (server) management functions.
In addition, as VMware stresses, offloading such functions to SmartNICs allows NSX, VSAN and host management to be extended to “bare metal” servers that don’t run server virtualization software. That in turn helps expand VMware’s addressable market.
The capabilities are part of a broader VMware initiative known as Project Monterey, which in addition to Nvidia has the backing of Intel (INTC) – Get Report, Hewlett-Packard Enterprise (HPE) – Get Report, Lenovo and VMware parent Dell Technologies (DELL) – Get Report.
Other features that will be enabled by Monterey include letting an app running on one server to access an accelerator (for example, an Nvidia GPU or an Intel or Xilinx (XLNX) – Get Report FPGA) hosted on another server, and running “a fully-featured stateful firewall and advanced security suite” on a SmartNIC. VMware argues the latter feature will allow individual firewalls to be custom-tuned to the security needs of individual apps.
Separately, Nvidia and VMware are partnering to make it easy for NGC to be deployed on servers relying on vSphere and Cloud Foundation, as well as to support the use of NGC in tandem with VMware’s Tanzu platform for deploying app containers via the popular Kubernetes container-orchestration platform.
With VMware having a massive footprint within on-premise enterprise environments, as well as partnerships with major public cloud providers, the companies argue that this effort will help developers quickly launch AI projects in on-premise environments, while also letting them (if they care to) easily move AI workloads between on-premise and public cloud environments. Nvidia adds that its NGC software will be available on “pre-tested” servers from Dell, HPE and Lenovo that contain its flagship A100 server GPUs.
In many ways, the new VMware partnerships fit with the broader, long-term, data center strategy Nvidia CEO Jensen Huang outlined earlier this month following news of Nvidia’s deal to acquire CPU IP giant Arm.
At the time, Huang asserted that Nvidia wants to build end-to-end silicon and software platforms for servers that encompass CPUs, GPUs and DPUs. He also insisted that Nvidia doesn’t see itself as a mere chip developer, but as “a computing platform company that is able to innovate from the chips to the systems, system software, all the way to the application stacks on top.”
Nvidia’s stock is up 2.3% in Tuesday trading, adding to its massive 2020 gains. VMware’s stock is up 2.7%.