Nutanix is building on its Cisco deal, Broadcom’s VMware acquisition, the GenAI boom, containerization, PostgreSQL interest, and green initiatives with a slew of announcements at its .NEXT conference in Barcelona.
The announcements concern GPT-in-a-Box, AI inferencing and training, adoption of in-place Cisco UCS servers and vSAN nodes, a PostgreSQL partnership with EDB, expanded Kubernetes support, and electricity power monitoring.
We’ll cover the AI angles here, look at UCS server support and vSAN node adoption in a second article, with EDB, Kubernetes, and power monitoring combined in a third story.
Nutanix and AI
“Enterprise is the new frontier for GenAI,” said Thomas Cornely, Nutanix SVP of Product Management, “and we’re excited to work with our fast growing ecosystem of partners to make it as simple as possible to run GenAI applications on premises at scale while maintaining control on privacy and cost.”
Nutanix has upgraded its GPT-in-a-Box offering to integrate with Nvidia’s NIM microservices and the Hugging Face large language model (LLM) library, and intends to add support for Nvidia’s GPUDirect file delivery protocol so Nutanix’s HCI systems can pump file data to Nvidia GPU farms for AI model training.
Joint customers will be able to use GPT-in-a-Box 2.0 to consume validated LLMs from Hugging Face and execute them. Nutanix and Hugging Face will develop a custom integration with Text Generation Inference, the Hugging Face open source library for production LLM deployment, and enable text-generation models available on the Hugging Face Hub within GPT-in-a-Box 2.0.
There will be a jointly validated and supported workflow for Hugging Face libraries and LLMs, ensuring customers have a single point of management for consistent model inference.
Nutanix says many enterprises find it challenging to make all the decisions needed to set up AI apps, such as choosing among hundreds of thousands of models, serving engines, and supporting infrastructure. They lack “the new skill sets needed to deliver GenAI solutions to their customers.” Its collaboration with Nvidia and GPT-in-a-Box v2 enhancements help simplify the customer experience.
Tarkan Maner, chief commercial officer at Nutanix, said: “Enterprises are looking to simplify GenAI adoption, and Nutanix enables customers to move to production more easily while maintaining control, privacy, and cost.”
Nvidia NIM
NIM microservices, running on top of the Nutanix Cloud Platform, enable retrieval-augmented generation (RAG) whereby generally trained GenAI LLMs get access to a proprietary and private user data: spreadsheets, presentations, documents, mails, POs, whitepapers, etc. They enable AI inferencing on a wide range of models, including open source community models, Nvidia AI Foundation models, and custom models, leveraging industry-standard APIs.
Users will be able to build scalable and secure NIM-enhanced GenAI apps by accessing the catalog of NIM microservices.
Nutanix announced certification for the Nvidia AI Enterprise 5.0 software platform for streamlining the development and deployment of production-grade AI, including Nvidia NIM.
Manuvir Das, Nvidia VP of Enterprise Computing, said: “The integration of Nvidia NIM into Nutanix GPT-in-a-Box gives enterprises an AI-ready solution for rapidly deploying optimized models in production.”
GPT-in-a-Box 2.0 will include:
- Unified user interface for foundation model management, API endpoint creation, end-user access key management
- Point-and-click-user interface to deploy and configure Nvidia NIM
- Integrated Nutanix Files and Objects, plus Nvidia Tensor Core GPUs
- Automated deployment and running of inference endpoints for a range of AI models and secure access to the model using fine-grained access control and auditing
- Support for AI-optimized GPUs, including Nvidia’s L40S, H100, L40, and L4
- Support for density-optimized GPU systems from Dell, HPE, and Lenovo to help lower TCO by allowing customers to deploy a smaller number of systems to meet workload demands
NUS, Data Lens, MGX and partners
Nutanix enhanced its unstructured data platform for AI/ML and other applications. Nutanix Unified Storage (NUS) now supports a new 550-plus terabytes dense low-cost all-NVMe platform and up to 10 GBps sequential read throughput from a single node (close to line speed for a 100 GbitE port), enabling faster data reads and more efficient use of GPU resources.
Nutanix’s Data Lens cloud-based data governance and ransomware hunting service has extended its cyber resilience to Objects in addition to Files data. A new Data Lens point of presence in Frankfurt enables broader adoption for EU customers, meeting their own compliance needs.
Nutanix has planned support for NX-9151, which is based on the Nvidia MGX modular server reference architecture. This allows for different configurations of GPUs, CPUs, and DPUs – including Nvidia Grace, x86 or other Arm CPU servers, and OVX systems.
There is a new Nutanix AI Partner Program providing customers with simplified access to an expanded ecosystem of AI partners. Partners will help customers build, run, manage, and secure third-party and homegrown GenAI apps on top of Nutanix Cloud Platform and GPT-in-a-Box, targeted at prominent AI use cases. Initial partners include Codeium, DataRobot, DKube, Instabase, Lamini, Neural Magic, Robust Intelligence, RunAI, and UbiOps.
Nutanix GPT-in-a-Box 2.0 is expected to be available in the second half of 2024. More information can be found here. Support for Nvidia GPUDirect and NX-9151 are under development. Additional features announced in NUS as well as Data Lens are available. More information here.
