re:Capping AWS re:Invent 2024
Leapfrog Technology, Inc.
Building digital experiences of tomorrow by innovating better, faster.
Written by: ?Amit Joshi, Solution Architect
It has been a remarkable year for Leapfrog in strengthening our expertise in Amazon Web Services (AWS). We started with the simple goal of diving deeper into the AWS ecosystem to provide better solutions to our clients. Our most notable achievement has been reaching the Advanced Tier in the Amazon Partner Network (APN). Along the way, a significant number of our people got AWS Certified in 2024.
We also embarked on achieving various AWS Service Competencies. Although we already had years of experience working on these high-value AWS Services, it was time to bring them to the limelight. AWS Lambda was the first, but there are more on the way.
All these initiatives are just the beginning. There is more to achieve on our way to becoming a leader in AWS Solutions. With this in mind, for the first time, our team participated in AWS re:Invent 2024. There, we learned about all the exciting new innovations AWS is working on and got a close look at what is coming to AWS.
Let’s delve into some of the announcements that caught our eyes.
Amazon Nova Foundation Models
With the rise of LLMs, one of the key challenges has been the compromise between the performance and price of a model. In an attempt to solve this dilemma, Amazon Nova introduces a new generation of foundation models with advanced capabilities for processing text, images, and video. The lineup includes:
The Nova models deliver up to 75% lower costs and significantly faster performance. They support customization, fine-tuning, and diverse multimodal applications, making them ideal for businesses seeking powerful AI-driven solutions.
While speech is still missing in Nova, Amazon plans to roll out speech-to-speech and any-to-any models in 2025. Comparing Nova's feature set to cost, these models could be the key to efficient and scalable applications, ready to tackle the next wave of innovation.
Amazon Bedrock Marketplace
It hasn't been long since Bedrock became available, but I have already found myself searching for a managed model that isn’t available in Bedrock. Amazon Bedrock Marketplace offers over 100 foundation models from top providers like IBM and Nvidia, enabling easy discovery, testing, and deployment through a unified platform. It supports text, image generation, and protein research, with secure integrations, flexible pricing, and managed infrastructure for efficient AI adoption.
Amazon Bedrock Automated Reasoning
Amazon Bedrock Guardrails' Automated Reasoning uses formal logic to prevent factual errors in LLM responses, especially in sensitive areas like HR. Combined with tools like content filtering and PII redaction, it helps users test and refine policies while providing validation results and suggestions to improve AI accuracy and reliability.
More Amazon Bedrock features
Model and RAG evaluation: Provides the option to evaluate Mode and RAG for accuracy using techniques like LLM-as-a-Judge.
Prompt caching: Reduces costs by up to 90% and latency by up to 85%.
Multi-agent collaboration: Multiple AI agents working together on complex multi-step tasks.
Knowledge base enhancements: Uses Aurora Serverless with PostgreSQL Vector DB during KB quick creation, provides support for Structural Data Retriever and multimodal data processing.
领英推荐
Model distillation: Creates a student model that is 5 times faster and 75% cost saving with similar accuracy to the teacher model.
AWS Trn2 Instances
Amazon introduced EC2 Trn2 instances and Trn2 UltraServers with second-generation Trainium2 chips for AI/ML tasks. Trn2 offers a 4 times performance boost over Trn1 and 30-40% better price-performance than GPU-based P5e instances. UltraServers support trillion-parameter models with 64 Trainium2 chips and 6 TiB memory. The AWS Neuron SDK enables PyTorch, JAX, and OpenXLA compatibility, providing a cost efficient alternative to GPU instances for AI/ML works. AWS also announced Trainium3 chips to be available in late 2025, whose UltraServers are expected to be 4x performant than Trn2 UltraServers.
Amazon Q Developer
Amazon Q Developer's new features include generating documentation, supporting code reviews, and automatically generating unit tests. It could be a serious competitor to GitHub Copilot. These features have the potential to increase developer productivity by automating time-consuming tasks like documentation creation, code quality checks, and unit test generation, allowing developers to focus more on core coding tasks.
Amazon S3 Metadata Search
Amazon S3 now offers queryable object metadata for S3 buckets, allowing users to easily search for and find objects based on specific criteria. Metadata is automatically generated when S3 objects are added or modified and is stored in fully managed Apache Iceberg tables. This feature solves the challenge of finding objects that meet particular criteria in large-scale S3 buckets, eliminating the need for complex, hard-to-scale custom systems that can fall out of sync with the actual bucket state.
Personally, I've been waiting for this feature for a while as size and timestamp are key metadata for me. It might finally be time to retire my hacky scripts.
Amazon EKS Auto Mode
Kubernetes for orchestration continues to push the industry forward with a cloud-agnostic solution for deploying our application. However, regardless of the benefits Kubernetes offers, it is well-known that the management side is not as easy.
While other cloud providers have been offering some level of auto Kubernetes, I’d argue that EKS Auto Mode probably has better implementation. EKS Auto Mode automates the infrastructure and management of Kubernetes clusters, significantly reducing the operational overhead for managing Kubernetes. It extends AWS management beyond just the cluster to the supporting infrastructure, including:
Other interesting announcements
AWS re:Invent 2024 highlights Amazon’s innovative vision to provide efficient, cost-effective, secure, and curated solutions for every business need. After analyzing all the announcements, one thing is clear: GenAI is here to stay.
AWS is taking the lead in delivering options to navigate this expensive, chaotic yet crucial world of GenAI with a solution that fits everyone’s use cases.
But the focus doesn't stop there. AWS knows what customers are asking for—improvements in database, storage, computing, security, and more—and will continue to prioritize providing them with the tools they need to thrive.
What are your thoughts on this piece? Let us know in the comments, or reach out to Amit Joshi via LinkedIn.
Vice President Growth & Advisory Services @ Leapfrog Technology | AWS Advanced Partner | HCLS Healthcare and FinTech GenAI | ML | AI | SecOps | Security Services |
2 个月Great recap Amit!
Senior Consultant, Data Analytics @TTEC Digital | Data Solutions Architect | Data Science and Engineering
2 个月Great read! Looking forward to more of your content!