SUSE and NVIDIA Say ‘What If’ to a $4K AI Desktop NVIDIA Supercomputer for Enterprises

Could a $4,000 desktop AI supercomputer help busy developers, workers, channel partners, and others boost their business computing power and drive new processes, AI initiatives, and creativity inside enterprises of all kinds?

SUSE and NVIDIA believed in the idea and recently created a desktop AI supercomputer testbed to show off its promise and performance for business use.

Starting with an NVIDIA DGX Spark desktop AI supercomputer, researchers combined it with SUSE’s Rancher Kubernetes Engine 2 (RKE2) and SUSE’s K3s Kubernetes distribution to create a machine that delivers a powerful punch of compute power for any organization. The NVIDIA DGX Spark AI desktop supercomputer starts at $3,999.

“For too long, massive, sprawling data centers have tied down enterprise AI,” wrote Stacey Miller, a principal product marketing manager at SUSE, in an October 14 SUSE blog post. “NVIDIA DGX Spark changes that. It’s an AI supercomputer that’s compact enough to fit on your desk. Now, with the secure and robust foundation of SUSE’s RKE2, this powerful hardware is ready to revolutionize how companies approach AI development. This integration combines NVIDIA’s cutting-edge hardware with SUSE’s trusted enterprise-grade, open-source Kubernetes software.”

The project was done as a technical collaboration to see what could be done, and not as a commercial project, Sanjeet Singh, senior director of SUSE’s recent SUSE AI platform, told ChannelE2E. “We worked together to validate the tech stack so users know it works, but we are maintaining separate sales channels — we are not selling the hardware and they are not selling the SUSE operating system software.”

For enterprises and developer teams, the combination of the NVIDIA DGX Spark and SUSE’s software delivers the architecture and much of the capabilities of a more expensive full-size supercomputer cluster at a much lower cost, said Singh.

Instead of a company having to spend $50,000 for such technology, it can spend $4,000, he said.

“We have small MSPs that we work with, one of which is purchasing five of these for their developers,” said Singh. “And that is real democratization. What they used to do was put it in the public cloud and then time slice it, which was expensive.”

MSPs and other channel partners can benefit from such machines because enterprises need their help, he said. “Enterprises are relying on integrators to build that AI application so that they can actually build some use cases out of it.”

Using similarly equipped and configured SUSE-powered NVIDIA supercomputers, developers could build these AI use cases in-house and run them within their company environments without the need for continuous and expensive data center connections, according to Singh.

“Absolutely, you build it in your building and deploy it in the public cloud or anywhere,” he said.

The recent SUSE AI platform adds to these capabilities because it is a private AI platform that allows developers to build AI applications on top of it, said Singh. “We provide the underlying framework, we provide the scalable Kubernetes, we provide the operating system, and the full stack.”

Moving From Developer Workstations to an AI Supercomputer Brings Big Gains

The move from traditional desktop developer workstations to desktop AI supercomputers will be a big benefit for channel partners, enterprises, developers, and their end customers, said Troy Topnik, director of product management for SUSE Rancher partners and ecosystems.

“Many, if not most, large AI workloads will ultimately run on Kubernetes in data centers, but developing those AI models, agents, and applications will often be done initially on individual developer workstations,” said Topnik.

“Having a platform like NVIDIA DGX Spark provides access to smaller-scale, but still very powerful, local GPU resources, and developing those models, agents, and applications on Kubernetes locally with RKE2 or K3s facilitates the later scale-out into the data center.”

The inclusion of a scalable and consistent Kubernetes platform for developers, data centers, and edge computing requirements is another benefit, allowing for simplification of the development, training, testing, and running of AI workloads, said Topnik. “Using the same Kubernetes distribution in all environments means you can use the same tooling and knowledge to create and run workloads at a variety of scales. Both K3s and RKE2 have applications in edge deployments.”