Insights from the AI Field Day: A Futurum Group Overview

Following the conclusion of AI Field Day 4, held from February 21 to February 23, 2024, I reflect on an event full of insightful discussions, groundbreaking presentations, and collaborative exchanges. They were led by the adept moderation of Stephen Foskett, organizer of Tech Field Day. They featured a stellar lineup of delegates and analysts, including myself and Allyson Klein, Ben Young, Colleen Coll, Donnie Berkholz, Ph.D., Frederic Van Haren, Gina Rosenthal, Jim Davis, Karen Lopez, Keith Townsend of The CTO Advisor, a Futurum Group Company, Ken Collins, and Ray Lucchesi, AI Field Day 4 surpassed expectations in unraveling the complexities and possibilities within the AI landscape. Participants from across the globe engaged fervently, contributing their insights and queries and shaping dynamic conversations that underscored the event's significance. As the curtain falls on AI Field Day 4, the resonance of its discussions lingers, sparking continued exploration and innovation in artificial intelligence.

VMware Private AI

The VMware session at AI Field Day 4, led by Chris Wolf, Global Head of AI & Advanced Services, delved into Private AI and its integration within VMware's ecosystem. Wolf highlighted the core challenges facing private AI adoption, emphasizing the importance of safeguarding intellectual property, data, and access. Private AI, he asserted, offers advantages in choice, cost, performance, and compliance, catering to various use cases, including code generation, contact centers, IT ops automation, and advanced information retrieval.

A notable aspect of the presentation was the discussion on VMware Cloud Foundation (VCF) serving as a platform for AI production, offering agility, scale, and lifecycle management capabilities. The introduction of RAY, an open-source method for scaling Python workloads, showcased VMware's commitment to enhancing AI capabilities within its ecosystem.

The collaboration between VMware and IBM in the realm of Private AI was highlighted, particularly in utilizing WatsonX for running data with compliance constraints that preclude public cloud usage. Wolf emphasized VMware's efforts in ensuring security and compliance through features like vSphere security and integration with Broadcom VCF.

Justin Murray, Product Marketing Engineer at VMware, provided insights into the VMware Private AI Foundation with NVIDIA solution, detailing its architecture, customer adoption, and performance monitoring capabilities. The demonstration of a Retrieval Augmented Generation (RAG) application underscored the practical application of VMware's AI solutions.

Shawn Kelly, Principal Engineer, elucidated the benefits of using VCF for AI production and detailed its integration with IBM WatsonX, facilitating on-premises deployment of IBM Watson services. Real-world use cases of Private AI at VMware, presented by Ramesh Radhakrishnan, highlighted the significant enhancements in documentation search and question-answering services, leveraging advanced models like Stanford Col(v)BERT IR.

Overall, the VMware session at AI Field Day 4 showcased the company's commitment to advancing AI capabilities within its ecosystem while addressing critical challenges in privacy, security, and performance. The collaboration with industry leaders like IBM and NVIDIA underscores VMware's role as a key player in the AI landscape, offering robust solutions for enterprise AI deployments.

Unveiling the Crucial Role of Storage in AI: Insights from Solidigm and SuperMicro

Next up on AI Field Day 4 was Solidigm and their partner Supermicro, offering valuable insights into the critical role of storage in artificial intelligence. The presentation, led by Ace Stryker, Director of Market Development at Solidigm, underscored the company's mission to accelerate the displacement of HDDs and establish itself as a premier SSD portfolio provider in the market. Stryker highlighted key trends driving Solidigm's focus, including increased chip spending, the proliferation of distributed architectures, and the rapid growth of edge computing, which outpaces traditional data center expansion. With a staggering 90% of storage being SSDs, Solidigm emphasized the importance of aligning storage solutions with the evolving demands of AI workloads.

Alan Bumgarner, Director of Strategic Planning at Solidigm, delved into the intricacies of AI data flow and storage workloads, outlining the five crucial stages of AI work: data ingest, data preparation (cleaning), training, checkpointing, and inference. By dissecting the AI pipeline, Bumgarner provided valuable insights into the storage requirements at each stage, highlighting the significance of optimized storage solutions in enhancing AI performance and efficiency.

Transitioning to Supermicro's presentation, Wendell Wenjen delved into the storage challenges associated with AIOps, emphasizing the pivotal role of storage infrastructure in enabling data-driven insights and decision-making processes. With over half of Supermicro's revenue attributed to AI-related ventures, the company's commitment to innovating storage solutions tailored for AI workloads was evident. Paul McLeod further elaborated on Supermicro's collaboration with WEKA, introducing the Supermicro DPU Direct Storage with Weka, a groundbreaking solution designed to address the evolving storage needs of AI applications.

A noteworthy aspect of Supermicro's approach was their preference for deploying entire racks for customers, highlighting a comprehensive and integrated approach to storage infrastructure deployment. While the presentations delved into storage-dense topics, they underscored the indispensable role of storage solutions in powering the entire AI lifecycle. As AI continues to evolve and proliferate across industries, the need for scalable, high-performance storage solutions remains paramount, driving innovation and collaboration within the AI ecosystem.

Unlocking Efficiency and Performance: Vast's Role in AI and HPC Data Storage

During the Vast session at AI Field Day 4, Keith Townsend engaged in a conversation with John Mao, VP of Tech Alliances, and Neeloy Bhattacharyya, Director of AI/HPC Solutions Engineering, shedding light on Vast's pivotal role in the Flash datacenter market. With a commanding 6% market share and promising earnings results, Vast has emerged as a significant player in the storage domain, boasting a deployment of 10 exabytes of data, with a considerable portion dedicated to AI and HPC workloads.

Originating as a storage platform, Vast has evolved by integrating additional features, such as a global namespace known as the "Vast database," which ensures write consistency across distributed environments. This architecture enables seamless accessibility to stored elements, regardless of their location within the data pipeline, fostering collaboration and efficiency across teams and workflows.

Vast's disaggregated shared architecture is characterized by a persistent layer where all data resides, while logic remains stateless and independent, facilitating streamlined data access and management. By leveraging insights into data access patterns, Vast optimizes performance by pre-staging data, ensuring its availability when required, thereby enhancing the overall efficiency of AI pipelines.

Furthermore, Vast's capability for data scanning at the data layer enables the identification and mitigation of inefficiencies in the data preparation phase of AI pipelines, contributing to enhanced data processing and analysis capabilities. Through these innovative approaches to storage and data management, Vast continues to position itself as a key enabler of AI and HPC solutions, driving advancements in data-driven workflows and insights.

Empowering AI Deployment: Intel's Xeon CPU Solutions Across Diverse Environments

Intel's session "Data, Model, and Deploy" at AI Field Day 4 underscored the critical focus on deploying deep learning inference models, particularly leveraging Intel Xeon CPUs across diverse environments. Ro Shah, AI Product Director at Intel, highlighted the significance of deploying AI everywhere, from edge to data center to the cloud, emphasizing the versatility and scalability of Intel Xeon CPUs in accommodating various AI workloads.

The session delineated two primary deployment scenarios: large-scale dedicated AI clusters and general-purpose AI integration into existing applications. Intel observed a trend toward utilizing Xeon CPUs for general-purpose AI deployments, particularly in scenarios where mixed workloads, such as video collaboration, demand efficient and cost-effective solutions.

A pivotal shift in the state of Generative AI was discussed, marked by the emergence of models ranging from billions to trillions of parameters. Despite the prevalence of larger models, many enterprises opt for smaller Large Language Models (LLMs), typically around 20 billion parameters, which align with Intel's focus on deploying AI on Xeon CPUs.

Shah emphasized the importance of latency in AI inference, with a target of less than 100 milliseconds for next-token latency, a threshold critical for real-time applications. Intel's roadmap for Xeon CPUs aims to optimize performance for both compute-bound and memory-bound tasks, ensuring efficient AI inference across diverse environments.

Beyond hardware capabilities, Intel highlighted its commitment to fostering a robust ecosystem and partnerships to address various AI deployment challenges. The session featured Keith Bradley, VP of IT Security and Technology at Nature Farms, sharing insights into their successful deployment of AI models leveraging Intel Xeon CPUs, demonstrating significant performance gains with each hardware upgrade.

Overall, Intel's session underscored the importance of deploying AI efficiently and effectively across diverse environments, with Intel Xeon CPUs emerging as a versatile and scalable solution catering to the evolving demands of AI inference workloads.

Unlocking AI Potential: VMware's Intel AMX CPU Integration in vSphere Environments

VMware presented an insightful session at AI Field Day 4, focusing on the utilization of Intel AMX CPUs for AI workloads within the vSphere environment. Earl Ruby, an R&D Engineer at Broadcom (VCF), led the discussion, highlighting the potential of AI without GPUs and leveraging Intel AMX CPUs on vSphere, particularly for Large Language Models (LLMs).

The integration of Intel AMX, comprising advanced matrix extensions embedded into every core, provides significant capabilities for AI and machine learning tasks. As Intel AMX is integrated into the latest Xeon chips, customers benefit from enhanced performance improvements for both traditional computing and AI workloads.

vSphere 8, with its hardware version 20 support, becomes a prerequisite for utilizing Intel AMX capabilities effectively. Developers can explore the Intel AI Tools Selector website for streamlined integration and optimization of AI workloads on Intel hardware within the vSphere environment.

During the session, a demonstration showcased the deployment of AI models on Ice Lake and Sapphire Rapids systems, illustrating substantial performance improvements with the latter. Ruby emphasized the efficiency of CPU utilization, highlighting scenarios where CPUs suffice for AI tasks, while GPUs remain essential for specific use cases demanding lower latency and accelerated processing.

Moreover, the session delved into the deployment of AI workloads on vSphere with Tanzu Kubernetes, emphasizing the importance of aligning the right versions and configurations to leverage AMX capabilities effectively. By utilizing OpenVINO on vSphere 8 and integrating Tanzu Kubernetes releases, organizations can explore AI deployment without the need for additional accelerator cards, leveraging the power of modern Xeon chips and vSphere infrastructure.

Overall, VMware's session showcased the evolving landscape of AI deployment, emphasizing the potential of Intel AMX CPUs within the vSphere environment and providing insights into leveraging Tanzu Kubernetes for seamless AI workload deployment and management.

Google's Innovations in AI Platforms and Infrastructures: Unlocking the Potential of Gemma and Gemini Models

During the Google session on Applied AI, Platforms, and Infrastructures at AI Field Day, Brandon Royal, Product Manager for AI Infrastructure and Cloud Runtime, along with Ameer Abbas, Product Manager, provided insights into Google's innovative approach to AI platforms and infrastructures.

The discussion centered around Gemma, a family of lightweight open models derived from extensive research and technology aimed at creating Gemini models. Google emphasized Kubernetes as the foundation for Open Large Language Models (LLMs), highlighting its flexibility, performance, and efficiency in supporting AI workloads. The Google Kubernetes Engine was spotlighted as a cloud-native infrastructure optimized for AI training and inference, offering a robust ecosystem that integrates both first and third-party solutions.

The session also introduced Vertex, a managed platform option that provides fully managed, out-of-the-box open models for Gemma and Gemini, offering end-to-end support for training and inference processes. Customers seeking a lower level of choice can explore options beyond Vertex, leveraging Google Cloud TPUs for enhanced performance and scalability.

Google Cloud TPUs were highlighted for their role in enhancing AI capabilities, with a focus on Google Kubernetes Engine offering two modes: Standard mode and Autopilot model. Ameer Abbas discussed Duet for Devs and VertexAI for builders, emphasizing Google's commitment to providing intuitive and accessible AI solutions for developers and users alike.

Overall, the Google session highlighted the company's dedication to advancing AI platforms and infrastructures, leveraging cutting-edge technologies like Kubernetes and TPUs to empower developers and organizations in building and deploying AI solutions efficiently and effectively.

Accelerating AI Pipelines: Hammerspace's Solutions for Unstructured Data Orchestration

During AI Field Day #AIFD4, Chad Smith and Floyd Christofferson of Hammerspace presented innovative solutions for accelerating AI pipelines and orchestrating unstructured data. They addressed the challenges of managing unstructured data across diverse storage systems and locations by introducing the concept of a global data environment with a parallel global file system. This environment allows data to remain in place while providing high-performance access essential for AI workloads.

Hammerspace's approach mitigates the silo problem in AI pipelines by assimilating file system metadata from existing storage, preventing copy sprawl, and maintaining data governance. They unveiled the Hammerspace #HyperscaleNAS, a storage-agnostic solution offering HPC-class parallel file system performance without altering existing infrastructure.

Real-world examples demonstrated the effectiveness of Hammerspace's solutions. For instance, a hyperscaler with a large AI training and inferencing environment achieved scalability, while a visual effects customer met rendering performance requirements without changing their storage setup.

In a subsequent session, Christofferson and Smith delved into the capabilities of Hammerspace's Global Data Environment software for automating unstructured data orchestration across multi-vendor, multi-site, and multi-cloud storage environments. They highlighted the separation of file system metadata from actual data, creating a global metadata control plane that enables transparent and automated data orchestration without disrupting user access.

The software supports multi-protocol access, objective-based policies, and seamless integration across platforms. It finds applications in various industries, as demonstrated through examples from online gaming, Blue Origin, and a London-based data center optimizing costs through render job orchestration.

These presentations collectively showcased Hammerspace's commitment to enhancing AI workflows and unstructured data management with practical and scalable solutions, addressing critical challenges in the AI ecosystem.

Empowering Data Insights: Qlik's AI Strategy Unveiled

The Qlik AI Strategy presentation, spearheaded by Miranda Foster, VP of Corporate Communications, and Mary Kern, VP Portfolio Marketing, provided valuable insights into Qlik's journey in the realm of data analytics and artificial intelligence (AI). As a leading data analytics company, Qlik has been delivering AI/ML solutions for over five years, with offerings like predictive analytics and AutoML capabilities enabling customers to build over 170K AI models.

Through an analysis of customer needs and preferences, Qlik identified a growing demand for AI solutions that reduce risk, address complexity, and scale the impact of AI initiatives. Leveraging their existing data foundations and pipelines, Qlik has developed turn-key AI-enhanced solutions, including auto ML and self-service capabilities, to help customers embrace AI complexity effectively.

The Qlik Staige solution integrates various components to enable users, even those unfamiliar with AI, to derive value from their data using well-established and trusted algorithms. Furthermore, Qlik's collaboration with Snowflake enhances its capabilities to provide comprehensive AI-driven insights and analytics.

In the AI Roadmap segment, presented by Nick Magnuson, Head of AI, and Ryan Welsh, Field CTO Generative AI, Qlik outlined its forward-looking approach to AI adoption. Recognizing the power of AI to derive insights from both structured and unstructured data, Qlik is expanding its capabilities to accommodate the analysis of unstructured data, leveraging acquisitions like Talon and Kyndi to enhance data acquisition and processing.

With an emphasis on generative AI and the integration of enterprise contextual data, Qlik aims to empower enterprises with AI-driven insights to solve complex business challenges effectively. Real-world use cases and demonstrations showcased the practical applications of Qlik's AI solutions, including predicting customer churn, understanding global operations, and incident identification and collaboration using Microsoft 365 Teams integration.

Overall, Qlik's AI Strategy presentation underscores the company's commitment to democratizing AI and enabling customers to effectively derive actionable insights from their data. It offers a glimpse into the future of AI-driven analytics and decision-making.

Summarizing Insights from AI Field Day Sponsors: Advancing AI

The AI Field Day event, featuring sponsors such as Intel, VMware by Broadcom, Vast, Hammerspace, Solidigm, Google Cloud, Qlik, and Kamiwaza, showcased diverse cutting-edge technologies and innovative solutions to revolutionize the AI landscape.

Intel and VMware by Broadcom presented groundbreaking insights into AI deployment, leveraging Intel AMX CPUs and VMware's infrastructure to accelerate AI pipelines and enhance data orchestration. Vast introduced solutions for managing massive volumes of unstructured data, while Hammerspace showcased innovative approaches to orchestrating AI workflows across diverse storage environments.

Solidigm and SuperMicro provided valuable insights into the crucial role of storage in AI, emphasizing the importance of high-performance storage solutions for AI workloads. Google Cloud demonstrated its commitment to advancing AI platforms and infrastructures, offering scalable AI development and deployment solutions.

Qlik's AI Strategy underscored the company's dedication to empowering organizations with AI-driven insights, while Kamiwaza presented innovative solutions for AI-driven automation and optimization.

Overall, the AI Field Day event served as a testament to the rapid evolution and transformative potential of AI technologies. With sponsors at the forefront of innovation, the event provided invaluable insights into the future of AI and its implications across various industries. As AI continues to permeate every aspect of our lives, events like AI Field Day play a crucial role in fostering collaboration, knowledge sharing, and advancements in AI technology.