VMware and Nvidia introduce the VMware Private AI Foundation with Nvidia: a joint offering that integrates Nvidia’s AI platform along with the necessary hardware into VMware’s cloud platform.

Keep VMware and Nvidia VMware Private AI Foundation with Nvidia above the baptismal font. The solution is exactly what the name suggests. The two companies are collaborating to make Nvidia’s AI offerings available within the VMware platform.

Adoption is critical to VMware. The company’s goal is to provide a vendor-independent cloud platform with its VMware Cloud and VMware Cloud Foundation that enables customers to centrally manage a hybrid multi-cloud environment. VMware’s platform layer runs on individual hyperscalers, giving organizations some independence from these giants.

That’s an attractive proposition as long as the capabilities of the VMware platform are competitive with those of hyperscalers. The AI ​​hype brings with it a challenge. If companies want to experiment with AI, they need to be able to do so within the VMware platform. Otherwise, individual solutions from hyperscalers with the associated manufacturer commitment and fragmentation of the multicloud IT environment are tempting.

(Planned) functionality

VMware Private AI Foundation with Nvidia combines VMware Cloud Foundation with the Nvidia AI Enterprise software stack. Of course, GPU integration is also included. The two parties point to a number of advantages:

• privacy: The solution enables customers to run their AI workloads close to where their data resides while maintaining data protection.
• Free choice: You can train models to walk anywhere you want. This will initially be done on hardware from manufacturers, and the public cloud will be added at a later date.
• Performance: It goes without saying that workloads that can be accelerated will benefit.
• scalability: The solution allows you to scale AI workloads up to 16 (v)GPUs, with a single virtual machine spanning multiple nodes.
• Lower costs: The two parties claim that their solution maximizes the available computing power of GPUs and CPUs. This is possible thanks to the solution’s HCI fundamentals, where processing power, storage and network capacity can be configured according to the needs of a virtual machine.
• storage: VMware references its vSAN Express Storage architecture, which supports direct I/O to the GPU via GPUDirect over RDMA without wasting CPU time.
• network acceleration: VMware vSphere and Nvidia NVSwitch get deep integration.
• speed: VMware and Nvidia provide images for vSphere ready to launch AI workloads with the necessary frameworks and libraries already installed.

The offer is compatible with Nvidia NeMo out of the box. This framework is part of Nvidia AI Enterprise and ensures that companies can relatively quickly train or adapt models according to their specific requirements. Llama-2 is also directly integrated and there is an integration of Hugging Face, a kind of GitHub for open source LLMs.

Availability

VMware Private AI Foundation with Nvidia will be available in early 2024, with it being clear what functionality the new offering will receive as of now. Nvidia and VMware explicitly name the manufacturers whose servers will initially support the new solution. These include Dell, HPE and Lenovo.

The focus is initially clearly on on-premise environments. Support for the entire hybrid multicloud is planned for a later date, in which customers can also integrate the capabilities of hyperscalers.

Everyone must come

Nvidia CEO Jensen Huang is a welcome guest at technology conferences this year. He previously spoke at the Snowflake Summit in Las Vegas, where Huang highlighted a similar offering. Snowflake also wants to offer an alternative that is detached from the hyperscaler offering, but the focus of the offering is on data. Again, integrating Nvidia and AI capabilities was key to keeping the offering relevant.