Accelerate your Gen AI journey
with Nutanix Unified Storage

Artificial Intelligence (AI) and Machine Learning (ML) have become integral parts of modern technology, powering everything from personalized recommendations to predictive analytics. Generative AI (Gen AI) elevates AI/ML to unprecedented levels. Instead of simply finding and classifying ‘existing’ content, Gen AI generates ‘new’ content, unlocking possibilities across a wide range of industries. In a recent CIO survey, 70% of respondents believe that AI/ML is a game-changing technology, poised to rapidly democratize digital delivery beyond the IT function¹.

Understanding the AI/ML lifecycle: Raw data to intelligent models

At the core of AI/ML is data. The journey of data from its raw state to intelligent AI/ML models constitutes the AI/ML lifecycle. Broadly, this lifecycle consists of the following stages:

1. Data Collection: During this stage, raw data is collected from various sources such as recorders, sensors, databases, and logs. Often this data is collected at the edge where it is being generated. This poses challenges in deploying a ‘right-sized’ and secure solution at the edge, managing multiple edge locations, and providing secure access and visibility to the data as it is collected.   
2. Data Processing: Raw data from various edge locations is aggregated at the core datacenter where it needs to be prepared for model creation or customization. Challenges at this stage include securely and efficiently consolidating data without multiple hops and implementing a dense, scalable storage solution for storing all raw data.  
3. Model Customization: After processing, data must be trained and iterated to develop a tailored AI/ML model. Training often demands significant computational resources and GPU acceleration, ideally achieved in the public cloud. In most cases, users will simply get an off-the-shelf trained model and tune it with use case specific data. Irrespective of the approach taken, the challenges at this stage include supporting fast, low-latency reads of bulk data by the GPU servers, streamlining data transfer to the public cloud, and ensuring storage flexibility between on-premises and cloud environments. 
4. Inferencing: Implementing and operationalizing trained models for real-time inferencing can occur either at the edge or in the datacenter. For Gen AI models like Retrieval-Augmented Generation (RAG), enhancing results may involve incorporating additional contextual data. Challenges at this stage include securely deploying AI models and context data with limited GPU resources and storage for real-time inferencing, keeping models and data updated, and gathering log data for continual model refinement.  
5. Data Archiving: Finally, raw and processed data require archiving for future analysis and compliance. If opting for on-premises archiving, a cost-effective object storage solution is required. Alternatively, if archiving in the public cloud, bulk data must securely and efficiently transition from on-premises datacenters to public cloud object storage, such as AWS Glacier.

Nutanix Unified Storage: Data services for the entire AI/ML lifecycle 

Most high-performance storage solutions for AI/ML are designed for specific stages of the AI/ML lifecycle, necessitating users to deploy and manage multiple storage solutions. This introduces risks, complexities, and inefficiencies stemming from managing disparate storage solutions, using third-party tools for data movement with multiple hops, and lacking visibility as data progresses through the AI/ML lifecycle.  

The Nutanix Unified Storage (NUS) solution is uniquely positioned to handle AI/ML workloads across all stages of the AI/ML lifecycle in the following ways:

• Dynamic Scalability: NUS offers unmatched linear scalability of capacity and performance, ranging from small scale edge deployments to hyperscale infrastructure in the datacenter or public cloud. Starting from as low as 1 TiB, NUS with Global Namespace can scale to hundreds of PBs. With the addition of 6500 ION platform by Micron Technology Inc., which delivers industry-leading sequential write speeds on high capacity 30TB SSDs, a single NUS node will boast over 550 TB of all-NVMe capacity, combining the cost and density benefits of QLC with the performance and endurance characteristics of TLC  – all while using 20% less power than competing High Capacity NVMe SSDs.

• Deployment Flexibility: NUS offers software-defined file, object, and block storage services, providing the flexibility to deploy as hyperconverged infrastructure (HCI) or dedicated high-performance dense storage, adapting to diverse AI/ML needs. Depending on the stage of the AI/ML lifecycle, NUS can be deployed at the edge, in the datacenter or in the public cloud. 
• Global Data Management: Nutanix Unified Storage (NUS) enables the seamless management of multiple clusters via a single global namespace, thereby streamlining data management and fostering collaboration across teams. Data scientists benefit from a unified view of data, regardless of its location, while granular IAM-based access controls ensure secure data access. Similarly, from an IT administration perspective, Nutanix clusters in different locations and stages of the AI/ML lifecycle can be managed centrally through the Nutanix Central SaaS-based global management portal.  
• Comprehensive Data Security and Insights: Security, privacy, and governance are paramount concerns as data traverses locations throughout the AI/ML lifecycle. The Nutanix Data Lens cyber storage solution offers integrated cyber resilience through proactive ransomware defense, auditing, and data insights, ensuring robust data security and governance at every stage. As a SaaS-based solution, Nutanix Data Lens provides a unified global portal for viewing data across multiple locations in different AI/ML lifecycle stages.
• Linear Performance Scaling: High throughput and linear performance scaling enable faster data processing, optimizing GPU utilization and accelerating AI/ML workflows. With Nutanix Unified Storage (NUS), a single node can support up to 10GB/s of sequential read throughput, approaching the line speed for a 100GbE network port. As an example, a 30-node cluster can linearly scale up to 300GB/s of aggregated sequential read throughput.    
• Edge to Cloud Data Mobility: The AI/ML lifecycle involves data movement across edge locations, datacenters and the public cloud. Built-in N:1 files and objects replication in NUS facilitate seamless data consolidation from multiple edge locations to a central location. Likewise, built-in cloud data replication can be leveraged to move data seamlessly to the public cloud, such as AWS S3 or the Nutanix Cloud Clusters (NC2) platform, for model training or data archiving.   

GPT-in-a-Box: Turnkey Gen AI Solution

To streamline infrastructure deployment for Gen AI and AI/ML workloads, we've introduced the Nutanix GPT-in-a-Box turnkey solution, which combines GPU-based compute, an opinionated AI software stack, Kubernetes^®, and Nutanix Unified Storage (NUS) for efficient inference and model tuning. GPU acceleration facilitates real-time inferencing and processing, while the storage can host the AI/ML model and any additional context data, such as that used with a Regenerative-Augmented Generation (RAG) model. Furthermore, NUS's built-in replication can be utilized for deploying and updating the data model and context data.

Nutanix Unified Storage for AI/ML: Delivering an unified experience

In conclusion, Nutanix Unified Storage (NUS) is a transformative solution for organizations embarking on their Gen AI and AI/ML journey. Its scalability, flexibility, simplicity, performance, data mobility and security features make it uniquely suitable for all stages of the AI/ML lifecycle. 

By offering a single platform for the entire AI/ML lifecycle and addressing the challenges inherent in AI/ML operations, NUS empowers organizations to accelerate, innovate, and succeed in their Gen AI and AI/ML journey.

Learn how Nutanix and Micron are collaborating to deliver high-capacity storage innovations - Read the blog

Learn more about Nutanix Unified Storage

• 10 Key Success Factors for Software-Defined Storage with Modern Data Services Platforms - Access PeerSpot Report
• Business Value of Nutanix Unified Storage - White Paper
• The Use of Object Storage for Modern Data-Intensive Workloads Requiring High Performance: Access the white paper.
• Get Data Lens for Free: Build a Fortress with Data Lens for Free!

¹Gartner-Survey-of-Over-2400-CIOs. Business Insider. https://www.businesswire.com/news/home/20231017475990/en/Gartner-Survey-of-Over-2400-CIOs-Reveals-That-45-of-CIOs-are-Driving-a-Shift-to-Co-Ownership-of-Digital-Leadership/