Tanushree Aggarwal for AWS Community Builders

Posted on Dec 3

CEO Keynote with Matt Garman - under 5 minutes

#aws #reinvent #techtalks #cloudcomputing

General Announcements

Billion dollars in credits announced for Startups!

Compute

NVIDIA P6 family of instances

Featuring NVIDIA Blackwell chips
Offering 25 times faster compute
Availability : early next year

Amazon EC2 Trainium 2 (TRM2) Instances Generally Available

TRM2 announced in Re:Invent2023 is now in GA!
Next generation Trainium instances, powered by Trainium2
offers best price performance for GenAI on AWS
Custom built processors by AWS for GenAI
Up to 30-40% better price performance than current generation GPU-based instances
16 Trainium2 chips
Up to 20.8 FB8 petaflops from a single compute

EC2 TRM2 Ultra Server

Connects 4 Trainium2 instances
Single Ultra node will offer over 83 petaflops of compute!
Expected to create material impact on model training clusters

Trainium 3 (TRM3)

Available later next year
2x more performance
40% more efficient

Storage

New bucket type S3 Table Buckets

Now in General Availability
Tabular data in the cloud
Up to 3 times faster query performance
Up to 10 times higher transactions per second for Apache Iceberg tables - Supporting automated table events
S3 manages removal of unreferenced files

Faster Metadata management in S3, announcing S3 Metadata

Now available in Preview
Fastest and easiest way to manage metadata in S3
S3 automatically updates object metadata in Iceberg tables

Database

Celebrating 10 years of Amazon Aurora

was announced at Re:Invent 10 years ago

Announcing Amazon Aroura DSQL

Now available in Preview
Fastest distributed SQL database
Low latency read-writes
Virtually unlimited scalability, and scales down to Zero
4x faster read-writes compared to Google Spanner
Fully managed
99.999% multi-region availability
Strong consistency
PostgreSQL compatible
Amazon DynamoDB Global Tables
Also supporting multi-region strong consistency

Artificial Intelligence

Amazon Bedrock Model Distillation

Create faster and more effective models
Send sample prompt from you application and Bedrock will do all the work for you
Transfer knowledge from a complex model to a small one with ease!
Distilled models are 500% times faster , 75% less expensive
Available in preview

Amazon Bedrock Automated Reasoning Checks

Reasoning system in Amazon Bedrock
Automated checks
If model thinks that the answer is not correct, it sends back asking for more prompts

Amazon Bedrock Multi Agent Collaboration

Easily build, deploy and orchestrate teams of agents that work together to handle complex, multi-step tasks
Orchestrate agents without complex coding
Agents work in parallel, accelerating tasks

Tune into Swami's keynote tomorrow for more Bedrock announcements!

Announcements from Amazon

Andy Jassy surprised us with 6 new Amazon Frontier models that will be available within Amazon Bedrock!

Amazon Nova

State of the art foundation models that deliver frontier intelligence
Offering industry leading price performance
Comes in 4 flavors - one text and three multimodal models.
75% more cost effective
Fasted models in their respective intelligence classes in Bedrock
Support fine-tuning to boost accuracy
Deep Integration with Bedrock Knowledge Bases for RAG
Optimized for proprietary applications that require interaction through APIs
Integration with just announced Amazon Bedrock Distillation

i) Amazon Nova Micro

Text based
Performance benchmarks equal or better than Llama and Gemini models
Now Generally available

ii) Amazon Nova Lite

Very low-cost multimodal model
Lightening fast for processing image, video and text inputs
Now Generally available

iii) Amazon Nova Pro

Highly capable multimodal model
Offers best combination of accuracy, speed and cost for wide range of tasks
Now Generally available

iv) Amazon Nova Premier

Coming in Q1 2025
Most capable multimodal model for complex reasoning tasks
best teacher for distilling custom models

Amazon Canvas

State of the art image generation model
Now Generally available
Edit images with NLP text prompts
Offers control for color scheme and layout
Built-in controls for responsible AI
Watermarking for traceability and content moderation

Amazon Nova Real

Now Generally available
State of the art video generation model

Amazon Nova Reel - coming soon

State of the art video generation model
Now Generally available
Offers full camera control - ranging from 360 degree rotation to zoom
Built-in safe AI features
Launched with 6 second videos
Will support up to 2 minute videos in the next few months

Sneak Peak

Amazon Nova Speech-to-Speech - speech to speech model, expected mid next year

Amazon Nova Any-to-Any - any-to-any model, expected mid next year

AI for Development

Amazon Q

Three new autonomous agents for Amazon Q, giving back developers valuable time!

Amazon Q develops and applies unit tests - available in GA
Use Amazon Q to create new accurate documentation – for new code or legacy code! - available in GA
Use Amazon Q to automatic code review, flag suspicious code packages - available in GA
After Slack, AWS Console and VSCode integration, now deep integration with GitLab - native support for GitLab Duo workflows! - Available in Preview

Amazon Q Developer for .NET

Modernizing Windows gets easier with Amazon Q Developer
Transform Windows to Linux in a fraction of time, 4x faster!
Agents support transforming hundreds of apps in parallel
Cost reduction by 40%
Available in Preview

Amazon Q Developer for VMWare

Modernize VMWare workloads using Amazon Q
Agents identify dependencies and transform network configs
Launches agents that can convert on-prem code to Cloud equivalent
Available in Preview

Amazon Q Developer for Mainframes

Q accelerates migration from mainframe to Cloud
Available in Preview
Offers code analysis, refactor applications
Takes legacy code and builds documentation
Available in Preview

Operations

Amazon Q Developer (for operations)

Investigate issues across AWS environment in a fraction of time
Provides guided workflows for investigating issues, by looking at CloudWatch and CloudTrail logs; checks for anomalies (eg: missing IAM permissions)
Offers possible remediations by referring runbooks and documentation
Available in Preview

PagerDuty Advance with Amazon Q

In partnership with PagerDuty
Offers unified user experience
Built using Amazon Bedrock and Claude
Leverages Amazon Bedrock Guardrails
In addition to resolving issues caused by AI, help reinforce responsible AI policies

Business Intelligence

Combining QuickSight and Amazon Q Business Data

Pull salesforce data into QuickSight
More powerful than an ordinary BI tool

ISV integration with Amazon Q index

New set of APIs for ISVs to use with Amazon Q index
Easily access data from multiple applications
You control permissions to your data
Better security with a single index
Now Generally Available

Amazon Q Business for automating complex workflows

Coming soon
Automatically build workflows based on documentation or recordings!
Navigate workflow changes in real time

AI in Analytics

Amazon SageMaker Unified Studio

Next Gen of Amazon SageMaker
Integrate with most comprehensive set of AI tools
Offers single view of enterprise data
Consolidates functionality data scientists and analysts, ML experts use
Create, share projects that use AI
Integrated data catalog and governance controls
Now in preview

Zero ETL for applications

Between Amazon Aurora, RDS for MySQL, DynamoDB, Redshift and third party SaaS apps
Now In GA

Amazon SageMaker Lakehouse

Simplifies analytics and AI with an open, unified and secure data lake house
Unifies access to data across S3, Amazon Redshift, SaaS and federated data sources
Consistent fine grain access control for data governance
Apache iceberg compatible lake house
Access through unified studio or any 3rd party tool which supports iceberg, using API
Now Generally available

Amazon SageMaker is now "Amazon SageMaker AI"

No AI was used in the creation of this blog. All information has been manually jotted down by the author during the livestream and published 10 minutes after the livestream ended. No official publications/documentation was referred before publishing this blog. In case you spot a wrong piece of information, please notify the author.