An AI Data Platform for All Seasons
- By Kirk Borne, Data Leadership Group
- May 27, 2024
They (some wise anonymous folks out there) say that there is a time and place for everything. They also say there is a season for every purpose. I believe the time, place, and season for artificial intelligence (AI) data platforms have arrived. To see this, look no further than Pure Storage, whose core mission is to “empower innovators by simplifying how people consume and interact with data.”
In the past, it was sufficient to bring order to the randomness of enterprise data collection through applications of technology resources (databases and storage devices) aimed primarily at organizing, storing, indexing, and managing enterprise information assets for single purposes or single business units. However, this data was still left mostly unexploited for its maximum potential and enterprise-wide business value.
Also, in the past, it was sufficient for business automation to consist primarily of rigid rule-based robotic non-adaptive repetition of processes and fixed tasks, requiring very little (if any) new knowledge input (i.e., live data consumption) or real-time adaptation to changing business conditions.
Also, in the past, it was sufficient for AI to be relegated to academic researchers or R&D departments of big organizations, who mostly produced research reports or journal papers and not much else.
Fast-forward to 2024, and we see a totally different landscape: massive data sets feeding dynamic cross-enterprise processes, increasing automation and dynamic adaption of complex multi-step tasks, and ubiquitous value-producing applications of AI. In particular, generative AI has played a major role in the explosive development and growth of these transformations within enterprises in the past year.
Pure Storage meets the demands of enterprise AI
To support, sustain, and assure the continued success and cost-effectiveness of data-fueled AI-powered transformations in this fast changing environment, Pure Storage has stepped up its delivery of an array of award-winning AI-ready infrastructure (AIRI//S™) products and services with an AI data platform that provides the fundamental AI environment for enterprise data management (storage, access, orchestration, delivery), hyperscaled AI training, and AI inference on demand (on-prem, in data centers, at edge sites, and in micro edge devices).
One example of Pure Storage’s advantage in meeting AI’s data infrastructure requirements is demonstrated in their DirectFlash® Modules (DFMs), with an estimated lifespan of 10 years and with super-fast flash storage capacity of 75 terabytes (TB) now, to be followed up with a roadmap that is planning for capacities of 150TB, 300TB, and beyond. Another example is Pure Storage’s FlashBlade®, which was invented to help companies handle the rapidly increasing amount of unstructured data that is coming into greater use, as required in the training of multi-modal AI models. One more example is Pure Storage’s development of non-disruptive upgrades (NDUs), a feature of Pure Storage's architecture that permits upgrades and expansion of the data infrastructure with no impact on data availability or performance and with no downtime or data migrations.
Pure Storage’s announcements at GTC 2024
The preceding examples are industry-leading and exemplary, yet there's still more. At the NVIDIA GTC 2024 conference, Pure Storage announced so much more! Here are a few more details on some of those announcements. See additional references and resources at the end of this article.
A data platform for AI
Data is the fuel for AI because AI devours data—finding patterns in data that drive insights, decisions, and actions. Ease of data orchestration (ingest, cleaning, transformation, discovery, access, exploration, delivery, training, inference, deployment) is essential for data-devouring AI products and services. A data platform for AI is key to innovation and long-term affordability, scalability, sustainability, and advancement of enterprise AI applications. Anything less than a complete data platform for AI is a deal-breaker for enterprise AI.
Pure Storage provides the ideal data platform for AI, as it provides unified storage for structured and unstructured data and enterprise data services for Kubernetes, supporting the entire AI data pipeline because storage matters!
At GTC 2024, Pure demonstrated the features of their data platform for AI, specifically highlighting these benefits and features of the platform: (a) Helps organizations accelerate model training and inference; (b) Improves operational efficiency for AI/IT infrastructure teams, as well as AI/ML developers and engineers; (c) Delivers cost and energy efficiency as an enterprise scales their AI operations; (d) Provides an AI storage platform that delivers ultimate reliability and is built to handle all future AI storage needs.
Optimizing GenAI Apps with RAG—Pure Storage + NVIDIA for the win!
One of the most popular techniques associated with generative AI (GenAI) this past year has been retrieval-augmented generation (RAG). RAG is the essential link between two things: (a) the general large language models (LLMs) available in the market and (b) a specific organization's local knowledge base. In deep learning applications (including GenAI, LLMs, and computer vision), a data object (e.g., document, image, video, audio clip) is reduced (transformed) to a condensed vector representation using deep neural networks. The knowledge base then becomes the comprehensive collection of these condensed representations of the enterprise business data repositories, stored in vector format in a vector database—Vector DB being another major data technology development finding widespread adoption this past year.
As a consequence of these activities, RAG provides the bespoke use case-specific context to an organization's proprietary GenAI LLM applications. This contextualization of the GenAI LLM is not only enterprise-specific, local, and customized, but it is also proprietary—maintaining the privacy and security of the GenAI LLM application within that organization's security firewalls and policies. Additionally, RAG ensures that an organization uses its most recent data while eliminating the need for constant retraining of LLMs. Pure Storage has worked with NVIDIA (GPU memory and GPU servers) to boost the speed, accuracy, and on-prem power of such enterprise GenAI LLM applications. Here are some specific documented results:
(a) “NVIDIA GPUs are used for compute and Pure Storage FlashBlade//S provides all-flash enterprise storage for a large vector database and its associated raw data. In a specific case [presented at GTC], the raw data consisted of a large collection of public documents, typical of a public or private document repository used for RAG.”
(b) “Document embedding, indexing, [and ingest] were completed 36% more quickly when using the Pure Storage FlashBlade//S with a native S3 interface than when using local SSDs that were inside each server, demonstrating that Pure Storage’s fast networked all-flash storage can help accelerate RAG document embedding.”
Pure Storage's RAG pipeline, in conjunction with NVIDIA GPUs and NVIDIA's NeMo Retriever collection of GenAI microservices, ensures the accuracy, currency, privacy, and relevance of proprietary enterprise LLMs. With Pure Storage, time to insight and action in AI applications is faster and better.
OVX validated reference architecture for AI-ready infrastructures
First question: What is OVX validation? OVX is NVIDIA’s standard validation paradigm for computing systems that combine high-performance GPU acceleration, graphics, and AI with fast, low-latency networking that are used to design and power complex 3D virtual worlds and digital twins that are transforming how businesses design, simulate, and optimize complex systems and processes. In this fantastic emerging realm of breathtaking technological achievements and innovations, Pure Storage has achieved OVX validation of its reference architecture for AI-ready infrastructures. At this stage, OVX validation applies directly to the increasing business demand for GenAI workloads (including RAG, LLMs, knowledge bases, and Vector DB), full-stack ready-to-run enterprise AI infrastructure, and local proprietary custom data + AI compute, storage, and networking solutions. Note: When you see “full-stack,” read “Pure Storage + NVIDIA working together seamlessly.”
Second question: What about technical debt and the cost of "lift and shift" to these new AI-ready architectures? For Pure Storage, OVX validation also certifies that Pure Storage's AI-ready infrastructure will run on NVIDIA GPUs and other vendors' servers, which is a great saving on technical debt for organizations that operate diverse server farms. OVX validation complements Pure Storage's certified reference architecture for NVIDIA DGX BasePOD, which was announced last year, as well as their FlashStack® for AI Cisco Validated Designs, which was announced here.
Since one of the only certainties about the future is its uncertainty, it is a great benefit that Pure Storage Evergreen//One™ provides storage-as-a-service (STaaS) guarantees and enables future-proof growth with non-disruptive upgrades. That means that Pure Storage owns the hardware ("the end user doesn't pay for it"), but the end user buys a subscription to the storage with the same agility and flexibility of public cloud storage and with all the security, proprietary protection, and performance of on-prem all-flash sustainable infrastructure. This is Pure Storage's SLA-guaranteed cloud-like STaaS!
More Pure Storage announcements at GTC 2024
Pure Storage's RAG development (described earlier) accelerates successful AI adoption across vertical industries. Pure Storage is accomplishing this by creating vertical-specific RAGs in collaboration with NVIDIA. First, "Pure Storage has created a financial services RAG solution to summarize and query massive data sets with higher accuracy than off-the-shelf LLMs. Financial services institutions can now gain faster insight using AI to create instant summaries and analysis from various financial documents and other sources.” Pure Storage will soon release additional RAGs for healthcare and the public sector.
Expanded investment in the AI partner ecosystem: Pure Storage is further investing in its AI partner ecosystem with NVIDIA, engaging in new partnerships with independent software vendors (ISVs). Some of these investments aim to optimize GPU utilization through advanced orchestration and scheduling, and others enable machine learning teams to build, evaluate, and govern their model development lifecycle. Additionally, Pure Storage is working closely with numerous AI-focused resellers and service partners to operationalize joint customer AI deployments further.
Looking at now and what's next
As the award-winning leader in AI-ready (and future-ready) data infrastructure, Pure Storage is collaborating with NVIDIA to empower their global customers with a proven framework to manage the high-performance data and compute requirements that these enterprises need to drive successful AI deployments, both now and into the future. Every technical leader, line of business (LOB) leader, VP of Infrastructure for AI, VP of AI/Data Science, and CDO/CTO/CAIO can now benefit from these technologies and services.
To put Pure Storage's recent accomplishments, products, services, and solutions into a single statement, I would say that its primary purpose (its North Star) is to guide and accelerate its customers' adoption of AI through the Pure Storage platform for AI.
The original article is here.
The views and opinions expressed in this article are those of the author and do not necessarily reflect those of CDOTrends. Image credit: iStockphoto/Ignitius
Kirk Borne, Data Leadership Group
Dr. Kirk Borne is the founder and owner of Data Leadership Group LLC. He is also an advisor to AI startups (including Prime.ai and Cipio.ai), an advisory board member to numerous organizations (including the IoT Community and Data Scientists Network—formerly Data Science Nigeria), a data scientist, global speaker, top influencer worldwide, astrophysicist, strategic advisor, educator, and trainer in all things Data/ML/AI.