Kinnar Sen

Non-Uniform Memory Access (NUMA) topology shapes how every byte of model weight reaches a compute core, and misalignment silently costs throughput. This session explains why NUMA matters for memory locality when running models on EC2, and what changes across Graviton-, Intel-, and AMD-based instances.

Supercharge your AI Infrastructure by leveraging AWS broadest and deepest compute services. In this chalk talk, we'll guide you through optimizing each stage of your AI pipeline, from data processing to AI/ML workload to user applications, for maximum cost-efficiency, scalability, and performance.

Do you want to build production-grade scalable and cost efficient Agentic AI applications? You can leverage Graviton's performance, cost efficiency, and sustainability for deploying up to 80% of an AI/ML pipeline.

Organizations use their data to power everything from financial planning to generative AI. In order to be data-driven, organizations need to build highly scalable, efficient data pipelines to keep up with the rapidly-evolving modern data landscape.

Retrieval Augmented Generation (RAG) enhances generative AI inference by augmenting pretrained FMs and LLMs with current information that is contextual and relevant to the query. Learn from AWS experts about efficient RAG deployment options for use cases that optimize cloud resource usage and costs.

Organizations are adopting Kubernetes to run their workloads for flexibility, scale, and cost efficiency. Deploying production-grade workloads can be simplified through the use of operator patterns.

Machine learning (ML) pipelines are comprised of multiple sequential steps that help to create a model from raw data. The compute requirements vary with the step, can be elastic in nature, and can be cost optimized.

Dive deep into the deployment of scalable, resilient, and cost-optimized Spark applications using Kubernetes via Amazon EKS and Amazon EC2 Spot Instances. Learn how to save money on big-data workloads, and understand the best practices for preparing your big-data workloads to run on Kubernetes efficiently with Amazon EC2 Spot Instances.