Top Announcements of AWS re:Invent 2024

The AWS re:Invent 2024 kicked off on December 1st, Pacific Time. In the opening keynote, Peter DeSantis, Senior Vice President of AWS Utility Computing, unveiled a new AI inference acceleration feature for the fully managed AI platform Amazon Bedrock, aimed at reducing latency and enhancing performance. For example, using AWS’s own AI accelerator chip, Trainium 2, to power Trn 2 processing of the Llama 3.1 405B model inference, it takes just 3.9 seconds to process 10,000 tokens of input and generate 100 tokens.

In addition to the Llama 3.1 model, Tom Brown, co-founder and Chief Computing Officer of AI startup Anthropic, also appeared to explain how they worked with AWS to design the Bedrock inference acceleration feature. This technology can reduce the inference time of Anthropic’s latest model, Claude 3.5 Haiku, by 60%, surpassing other platforms’ speeds. This feature is available for use starting today via API. Additionally, Tom Brown revealed that they will be using hundreds of thousands of Trainium 2 chips to train the next-generation Claude models.

AWS Trainium2 Instances Now Generally Available

AWS has announced the general availability of Trainium2-powered EC2 Trn2 instances, optimized for high-performance deep learning and generative AI workloads, including large language and latent diffusion models. Trn2 instances deliver 30-40% better price-performance compared to GPU-based EC2 instances and feature 16 Trainium2 chips connected via NeuronLink technology, offering 20.8 peak petaflops of compute. Additionally, AWS introduced Trn2 UltraServers, designed for scaling generative AI workloads with up to 64 Trainium2 chips, significantly enhancing training and inference for large models. In collaboration with Anthropic, AWS is building Project Rainier, the world’s largest AI compute cluster.

AWS also unveiled Trainium3, its next-generation AI chip featuring a 3-nanometer process node, expected to quadruple the performance of Trn2 UltraServers. Trainium3 will enable faster model training and superior real-time performance for generative AI applications, with the first Trainium3-powered instances anticipated in late 2025.

Latest Amazon S3 Innovations To Remove Working Complexity

AWS announced new innovations for Amazon S3 at re:Invent, introducing fully managed Apache Iceberg support for faster analytics and easier tabular data management. These enhancements simplify data discovery with automatically generated queryable metadata and eliminate the complexity of organizing table and metadata stores on S3.

Amazon S3 Tables provide built-in support for Apache Iceberg tables, enabling up to 3x faster query performance and 10x higher transactions per second compared to standard S3 buckets. Designed for analytics workloads, S3 Tables optimize query performance and storage costs while supporting table-level access controls.

Amazon S3 Metadata generates queryable object metadata in near real-time, reducing the need for complex and expensive metadata systems. This helps customers discover, organize, and utilize their data efficiently for analytics and real-time applications. Amazon S3 Tables are now generally available, and Amazon S3 Metadata is in preview.

Amazon Aurora DSQL：The Fastest Distributed SQL Database & Enhancements to Amazon DynamoDB Global Tables

AWS CEO Matt Garman announced new database capabilities for Amazon Aurora and Amazon DynamoDB during re:Invent, targeting the most demanding multi-Region workloads requiring strong consistency, low latency, and high availability. These updates cater to both SQL-based relational databases and NoSQL systems.

Amazon Aurora DSQL: AWS introduced the fastest distributed SQL database, a serverless, PostgreSQL-compatible solution offering 99.999% multi-Region availability, strong consistency, virtually unlimited scalability, and zero infrastructure management. Aurora DSQL achieves four times faster read/write performance compared to other distributed SQL databases and overcomes challenges like low-latency strong consistency and global server synchronization with microsecond accuracy.

Amazon DynamoDB Enhancements: DynamoDB global tables now include the option for strong consistency, leveraging Aurora DSQL’s underlying technology. This enhancement ensures multi-Region applications always read the latest data without requiring application code changes, complementing DynamoDB’s existing high availability, scalability, and consistent low-latency performance.

Both Aurora DSQL and enhanced DynamoDB global tables are now available in preview.

Introducing Amazon Nova

Amazon has unveiled Amazon Nova, a new generation of foundation models (FMs) designed for advanced generative AI applications. These models support text, image, and video inputs, enabling tasks like video analysis, content creation, and multimedia generation. Built with responsible AI principles, Amazon Nova models also include safety features and transparent usage guidelines.

Amazon Nova includes several specialized models:
• Nova Micro: Text-only model for ultra-low latency and cost.
• Nova Lite: Low-cost multimodal model for text, image, and video.
• Nova Pro: High-performance multimodal model balancing speed, accuracy, and cost.
• Nova Premier: Advanced multimodal model for complex tasks, debuting in 2025.
• Nova Canvas and Nova Reel: Models for image and video generation.

Integrated with Amazon Bedrock, these models simplify experimentation, fine-tuning, and deployment while offering cost savings (75% cheaper than comparable models) and support for 200 languages. Customers can fine-tune the models using proprietary data and leverage distillation techniques for efficiency. The models excel at Retrieval Augmented Generation (RAG) and are tailored for agentic tasks requiring interaction with proprietary systems.

Amazon Nova also enhances creative output, as demonstrated by Amazon Ads, where tools like Nova Reel helped brands produce high-quality advertising content. Future plans include a speech-to-speech model and an any-to-any modality model capable of processing and generating multimodal inputs and outputs.

New Amazon Q Developer Capabilities

AWS has announced new capabilities for Amazon Q Developer, a generative AI-powered assistant designed to simplify complex application migration and modernization projects. These enhancements aim to save organizations significant time and costs by automating processes such as source code analysis, code generation, testing, and execution.

Amazon Q Developer has already proven its efficiency by migrating thousands of production applications from older Java versions to Java 17, saving an estimated 4,500 years of development work and achieving $260 million in annual cost savings.

The latest updates expand its capabilities to support large-scale legacy transformations, including:
• Windows .NET to Linux Modernization: Achieve migrations up to 4x faster while reducing licensing costs by up to 40%.
• VMware Workload Transformation: Convert on-premises VMware network configurations to AWS cloud-native architectures within hours instead of weeks.
• Mainframe Modernization: Streamline processes such as code analysis, documentation, and refactoring to accelerate mainframe workload migration.

These capabilities are available through a new Amazon Q Developer web application, offering an optimized collaboration platform for transformation projects. VMware and mainframe transformations are exclusive to the web application, while Windows .NET transformations can also be performed within developers’ IDEs. All new Amazon Q capabilities are available in preview today.

Enhanced Amazon Q Developer

AWS announced enhancements to Amazon Q Developer, a generative AI-powered assistant that accelerates software development and operations. With its new features, Amazon Q Developer automates tedious tasks, allowing developers to focus on strategic work and improve productivity. It also offers unmatched capabilities to resolve operational issues faster, benefiting developers and operators alike.

Key enhancements include:
• Automated Unit Testing: Identifies and generates unit tests automatically, improving test coverage and reducing developer effort, enabling faster feature delivery with higher code reliability.
• Automated Documentation: Creates and updates accurate project documentation, simplifying onboarding for new developers and maintaining clarity across growing projects.
• Automated Code Reviews: Performs initial code reviews to catch quality, style, and security issues, streamlining the review process and reducing time spent on revisions.
• Operational Issue Resolution: Leverages AWS expertise to investigate and resolve production issues across AWS environments quickly. By analyzing vast data points, Amazon Q identifies anomalies, proposes root cause hypotheses, and guides users to fixes while compiling findings in Amazon CloudWatch for team collaboration.

Amazon Q Developer integrates with the AWS Management Console, GitLab, and IDEs, offering developers flexibility in where they work. The new agentic capabilities are available in IDEs today and in preview via GitLab, while the operational capability is in preview.

Amazon Q Business Is Transforming How Employees work

AWS has announced new features for Amazon Q Business, a generative AI-powered assistant that helps employees find information, gain insights, and take action efficiently. These updates streamline workflows, enhance productivity, and integrate enterprise data across multiple applications, enabling employees to focus on impactful work.

Key innovations include:
• Unified Insights Across Data Sources: Amazon Q Business integrates with Amazon QuickSight to provide insights and analysis from structured data (databases, data lakes) and unstructured data (documents, emails) in one place. Employees can make faster decisions using visually rich data presented through QuickSight.
• Cross-App Index for Enhanced Experiences: A new cross-app index allows independent software vendors (ISVs) to access unified organizational data through a single API, enhancing generative AI-powered experiences across apps like Zoom, Asana, and PagerDuty. Fine-grained permissions ensure data security and control.
• Library of 50+ New Actions: Employees can perform tasks directly within popular third-party apps, such as creating Asana tasks or sending Microsoft Teams messages, without switching between tools.
• Automation of Complex Workflows: Generative AI agents now automate complex tasks like invoice processing, customer support ticket management, and employee on-boarding, further reducing manual effort.

Amazon Q Business capabilities like the cross-app index and 50 new actions are available now. Insights from QuickSight are in preview, while the new automation capabilities will launch in 2025.

Amazon SageMaker: Unified Platform for Data, Analytics, and AI

AWS announced the next generation of Amazon SageMaker, providing a unified platform for SQL analytics, big data processing, machine learning (ML) development, and generative AI. This platform integrates tools and data, simplifying collaboration and enabling teams to act on insights efficiently.

Key features include:
• SageMaker Unified Studio: A single development environment combining analytics, ML, and AI tools, offering seamless data access and collaboration across an organization.
• SageMaker Catalog: A governance framework that ensures secure access to data, models, and artifacts. Built-in safeguards include data classification, toxicity detection, and responsible AI policies.
• SageMaker Lakehouse: Provides unified access to data stored across lakes, warehouses, and federated sources, supporting Apache Iceberg standards for simplified data processing. Integrated access controls enable secure data sharing.
• Zero-ETL Integrations: Simplifies data integration by eliminating the need for manual ETL processes, allowing direct access to SaaS application data (e.g., Zendesk, SAP) for analytics and AI.

These features streamline workflows, reduce silos, and enhance collaboration across data and AI projects. SageMaker Unified Studio is currently in preview, while other features are generally available.

The AWS re:Invent conference in 2024, with the new UltraServer Trainium2, takes only 3.9 seconds to process 10,000 Token inputs and generate 100 tokens, greatly accelerating the processing speed and helping enterprises create more business value. Want to know more about AWS’s new services and the latest cloud technologies? Nextlink will continue to update you with the latest cloud technology information, don’t miss it!