Review of 2023 Amazon Cloud Technology re:Invent, innovative AI, walking together all the way

As the world's leading cloud computing company, Amazon Cloud Technology held the 2023 Amazon Cloud Technology re in Las Vegas from November 27 to December 1, 2023. :Invent, from the Amazon Cloud Technology re:Invent global conference held in 2012 to the current 2023 Amazon Cloud Technology re:Invent, looking back at the previous re:Invent conferences, Amazon Cloud Technology has played an important role in every re:Invent conference. New products will be launched at each conference, and the innovative concepts promoted have increasingly become the benchmark for innovation in the global cloud computing industry.

There is no limit to the expansion of data scale, leading the way to build Serverless

Peter, Senior Vice President of Amazon Cloud Technology, announced three serverless innovations in databases and applications on the first day of re:Invent, enabling customers to more quickly and easily scale their data infrastructure to support demanding business use cases. .

New shoots from old branches? Three new serverless services launched

**New Amazon Aurora Limitless Database** automatically scales to millions of write transactions per second and manages petabytes of data while maintaining the simplicity of operating a single database;
New Amazon ElastiCache Serverless Makes it faster and easier to create highly available cache services and instantly scale to meet application needs;
New Amazon Redshift Serverless uses AI Model to predict workloads and automatically scale and optimize, providing up to 10 times cost-effective improvement.

Amazon Aurora Limitless Database data expansion without limits

Recalling that in recent years, Amazon Cloud Technology has built large-scale Serverless services on its huge cloud service system, including Amazon S3 for storage, Amazon Lambda for computing, and Amazon DynamoDB for databases. Later in 2018, Amazon Cloud Technology launched Amazon Aurora Serverless, which can automatically and seamlessly adjust database capacity according to workload needs without any adjustments or failovers to the database. Although Amazon Aurora Serverless was launched, Peter concluded: "We are still limited by the size of the physical server, which is not Serverless. However, database sharding is a well-known improvement technology.Consider horizontally dividing data into subsets and distributing them to a bunch of physically independent database servers (called shards) to take advantage of the database performance of a single server." , so Amazon Cloud Technology launchedNew Amazon Aurora Limitless Database: Customers can easily expand their databases beyond the capacity limits of a single server and achieve high performance through database sharding. **With unlimited databases, there is no need to worry about new additions that need to be managed. With a new database, your application has just one database entry available.

Amazon ElastiCache Serverless instantly builds the backbone of the cloud

Amazon ElastiCache Serverless is a new serverless option that allows customers to create a cache in a minute and instantly scale capacity based on application traffic patterns. Amazon ElastiCache Serverless is compatible with two popular open source caching solutions, Redis and Memcached.
Judging from the author's own experience, when using Elasticache to create Redis or Memcached in the past, it would take at least 4-5 minutes, and deletion would also take a long time. If For the test environment, it is a very time-consuming step. Now that Elaticache Serverless was launched, the author immediately went to experience Elasticache Serverless. In actual operation, the startup speed is not necessarily within 1 minute as officially claimed. Of course, Yes, but in most cases, the created cache database will become available within the interval of 1 minute to 2 minutes.
One section

The speed has indeed improved. It is speculated that it may be because of the amount of resources that the speed may not be that fast.
A major feature of Serverless is that it automatically expands the amount of resources. By default, resources are not limited and can grow on their own. Of course, customers are also given the option to set the maximum amount of resources themselves.

Amazon Redshift Serverless AI helps save time and effort

Amazon Cloud Technology launched Amazon Redshift Serverless in 2021, but for some demanding workloads, manual intervention is still required. Peter demonstrates these issues by introducing some challenging data warehouse patterns

Consistently low sub-second latency is required for small queries for dashboards and reports;
For large ETL batch jobs, they need to run efficiently without interfering with other workloads. These jobs may process billions of rows;
For special complex analytical queries, performance needs to be optimized without impacting production workloads. These queries can be 10-100 times the size of typical queries.

Based on the above issues, Amazon Cloud Technology has launched a new Amazon Redshift Serverless feature: next-generation AI-driven scaling and optimization. **This function uses AI models to predict workloads and automatically expand and optimize resources. It upgrades the existing query-based resource allocation to query-based, data-level calculation complexity and other dimensions, and uses machine learning models to calculate Resources are dynamically allocated to help customers achieve cost-effectiveness goals.

Amazon Cloud Technologies builds on pioneering work in serverless technology to help customers manage data of any scale and dramatically simplify their operations so they can focus on innovating for end users without spending time and energy configuring, managing and expand its data infrastructure. Mainly includes the following capabilities:

Predictive models based on machine learning can predict future workload patterns and adjust resource capacity in advance.
Real-time query analyzer uses machine learning to estimate the resource requirements of each query and allocate them appropriately. The system analyzes more than 50 unique features of each query.
Optimize every query to reduce cost or improve performance based on customer needs. Queries have different scaling modes such as linear, sublinear and superlinear.

Use both soft and hard tools, AI support, and full-stack generative AI capabilities

Amazon Cloud Technologies and Anthropic deepen partnership

Adam emphasized the importance of flexible customer choice in the rapidly evolving field of artificial intelligence. To achieve this goal, Amazon Cloud Technologies continues to expand its cooperation with industry-leading innovative companies, such as groundbreaking AI startup Anthropic. Through a new partnership with Anthropic, the latter will leverage Amazon Cloud’s dedicated machine learning chip, Trainium, to train their next generation of complex Claude models. Amazon Bedrock customers will also receive exclusive early access to advanced Claude customization and fine-tuning model capabilities not available elsewhere.

Amazon AI works again!

Amazon Bedrock releases more model choices and new development tools to help securely build and scale generative AI applications

Amazon Bedrock now supports new base model versions including: Anthropic Claude 2.1, Meta Llama 2 70B, Amazon Titan Family, and more

New features of Amazon Bedrock, including model fine-tuning, retrieval-augmented generation (RAG), and pre-training based on the Amazon Titan large model

Another key layer on top of generative AI is providing customers with simple, fast and secure access to APIs for a variety of underlying models through Amazon Bedrock. Amazon Bedrock launched two months ago and has already attracted more than 10,000 active customers across a variety of industries using it to quickly build and scale generative AI applications. Adam also released new features of Amazon Bedrock, including model fine-tuning, retrieval-augmented generation (RAG), and pre-training based on the Amazon Titan large model.

Agents for Amazon Bedrock

**With the new GA Agents for Amazon Bedrock, users can create and deploy fully managed Agents in a few simple steps, and perform complex business tasks by dynamically calling APIs. **Amazon Bedrock provides the API architecture required to complete tasks based on natural language instructions provided by users.

Guardrails for Amazon Bedrock

At Amazon Cloud Technology, we are committed to developing AI in a responsible manner. Adam has released a new preview version of Guardrails for Amazon Bedrock for the security of generative AI, customizing safeguards based on application requirements and AI policies. Guardrails can provide a consistent level of AI security for all applications across underlying models, block unwanted topics in generative AI applications, filter harmful content based on AI policies, and more. Generative AI must be safe and responsible. This is Amazon Cloud Technology’s Job Zero.

A new generation of Graviton processors - Amazon Graviton4

Looking back at the recent Amazon Cloud Technology Re:Invent conferences, almost all of them will launch hardware

In 2018, AWS released the first version of Graviton for open source and non-performance-critical scripting workloads as part of its A1 instance family.
The second-generation AWS Graviton2 was released in December 2019. As the first of its sixth-generation instances, AWS promises a 40% improvement in price/performance and an average 72% reduction in power consumption compared to fifth-generation Intel and AMD instances.
In May 2022, AWS offered Graviton3 processors as part of its seventh-generation EC2 instances, delivering a further 25% improvement in compute performance over Graviton2.

Now, in 2023, Amazon Cloud Technology Re:Invent launched a new generation of Graviton—Amazon Graviton 4: 96 Neoverse V2 cores, 2MB of L2 cache per core, and 12 DDR5-5600 channels work together to make Graviton 4 and Compared with Graviton3, the speed of processing databases is increased by up to 40%, the speed of processing Web applications is increased by 30%, and the speed of processing large Java applications is increased by 45%.
R8G instances supporting Amazon Graviton 4 will be available in multiple sizes, with 3x the vCPU count and 3x the Amount of memory.
David Brown, vice president of computing and networking at Amazon Cloud Technologies, said: "By focusing our silicon designs on the real-world workloads that matter to our customers, we are able to provide them with the most advanced cloud foundation. facilities." "Graviton4 marks our fourth generation of silicon in just five years and is the most powerful and energy-efficient silicon we have ever built for a broad range of workloads."

Note: The tables filled with "/" indicate that the author has not found relevant data to fill in.

processor	Graviton1	Graviton2	Graviton3(E)	Graviton4
Example	A1	M6g/M6gd
C6g/C6gd/C6gn
R6g/R6gd
T4g
x2gd
G5g
I4g/Im4gn/Is4gen	C7g/C7gd/C7gn
M7g/M7gd
R7g/R7gd
Hpc7g	R8G
core	Cortex-A72	Neoverse-N1	Neoverse-V1	Neoverse V2
maximum frequency	2000MHz	2500MHz	2600MHz
Architecture revision	ARMv8.0	ARMv8.2-a	ARMv8.4-a	ARMv9.0-A
Number of cores	16	64	64	96
L1 cache (per core)	48KB inst / 32KB data	64KB inst / 64KB data	64KB inst / 64KB data	/
L2 cache (per core)	512KB	1MB	1MB	2MB
LLC (shared)	/	32MB	32MB	/
DRAM	/	8x DDR4	8x DDR5	12x DDR5-5600
DDR encryption	/	yes	yes	/

A new generation of Amazon Trainium2 chips

**Based on the successful experience of the training chip Trainium, Adam officially released a new generation of Trainium2 chip. **It has optimized basic model training with hundreds of billions or even trillions of parameters. Its performance is 4 times higher than that of the previous generation chip. It has 65 EFlops and can provide performance support on demand. Star generative AI company Anthropic plans to build models using Trainium2 chips.

When innovating, we will interact between various FMs and APIs. Amazon Bedrock can provide a variety of models, such as AI21 labs, Anthropic, Cohere, and Meta. There is also an Amazon Titan model. In addition, Amazon Cloud Technology is also the first cloud vendor to integrate the Meta Llama 2 model.

Amazon Cloud Technology becomes the first cloud vendor to launch NVIDIA GH200 NVL32 instances

NVIDIA founder and CEO Jensen Huang announced that Amazon Cloud Technology has become the first cloud vendor to launch NVIDIA GH200 NVL32 instances. This instance comes with 32 GH200 super chips, which are interconnected through up to 900GB/s NVLink network, forming an instance with up to 20TB of shared memory, which can be used to accelerate the training of large AI models with 1 trillion parameters.

Where there is Q, there must be A, Amazon Q, the fried products of Amazon Q

If there is a question, there must be an answer, Amazon Cloud Technology launched the hottest product at 2023 Re:Invent **Amazon Q: **A product based on generative artificial intelligence (AI) A new assistant designed to assist with work and can be tailored to a customer's business, supporting developers and IT professionals, available in multiple areas of AWS, and quickly accessible no matter where you work Answers and Ideas
Offering a variety of features and usage scenarios

With Amazon Q, AI experts are on hand to answer questions, write code faster, troubleshoot issues, Optimize workloads and even help you write new features. These capabilities simplify all phases of building applications on AWS.
If you need additional assistance, Amazon Q also allows you to interact with an AWS Support agent directly from the Q interface, eliminating any pain points in the customer self-service experience. Integration with AWS Support is available in the console and provides benefits included in AWS Support plans.

Amazon Cloud Technology’s Technical Philosophy

Amazon CTO Dr. Werner Vogels’s keynote speech at re:Invent this year took Amazon as a best practice from a special perspective and showed us how to design a cost-first technology architecture.

"Identify the areas with the most profit potential in business activities, and build and optimize the entire business structure around this." This sentence actually reveals Amazon's Technology philosophy means that technology is always built around business, and its purpose is to make business run more efficiently. In this way, it is not difficult to see why Amazon first popularized cloud technology to users, and it has always represented the most advanced trend of thought in the field of cloud technology.

Rule No. 1: Treat cost as a non-functional requirement
Consider costs early and continuously when designing, developing, and operating systems issues to balance functionality, time to market, and efficiency.
Second Law: The durability of a system depends on how well its costs match the business model
Design systems that align with the profitability levers of the business model to leverage Economies of scale, going with the trend, ensuring continuous growth while ensuring profits. Unlimited growth without profitability will diminish value.
The third rule: architectural design is a set of trade-offs
Every design decision is accompanied by trade-offs. It is critical to regularly re-evaluate technical and business trade-offs and commit resources to meet business needs.
Law 4: Unobservable systems have immeasurable costs
Although monitoring systems require upfront investment, they allow organizations to pinpoint wasteful behavior, optimize workflow, and allocate resources to high-priority tasks in an orderly manner.
Law 5: Cost-aware architecture enables cost control
With a strong monitoring system in place, you can take action in areas where you identify opportunities for improvement. By implementing detailed controls, you can achieve the best balance between cost and user experience.
The sixth rule: Cost optimization is a gradual process
The pursuit of cost efficiency is a continuous process. Monitor your system to understand its patterns and eliminate inefficiencies. Continuous optimization requires revisiting the system to find more room for improvement.
Law 7: Unchallenged success leads to blind confidence
Continue to question what has worked in the past. Revisit previously successful methods and tools. As Grace Hopper (the famous computer scientist and the mother of COBOL) said, one of the most dangerous phrases in the English language is: "We've always done it this way."

Looking to the future

Starting from 2020, Amazon Cloud Technology has begun to make predictions for the coming year and actively respond to future developments

For 2021: From schooling to space: eight predictions on how technology will continue to change our lives in the year ahead

Cloud will be everywhere
The Internet of Machine Learning
In 2021, pictures, videos, and audio will trump text
Technology will transform our physical world as much as the digital world
Distance learning wins its place in education
Small businesses will race to the cloud, with Southeast Asia and sub-Saharan Africa leading the way
Quantum computing is starting to boom

For 2022:

Artificial Intelligence-Powered Software Development Takes Dominance
Ubiquitous cloud has advantages
The rise of smart spaces, especially in aged care
Sustainable development has its own structure
New waves of connectivity will bring new application categories

For 2023:

Cloud technology will redefine sports as we know it
The analog world will reshape how we experiment
Smart energy innovation wave
The coming supply chain transformation
Customized chips become mainstream

for 2024

Generative AI will become culturally aware
Femintech is finally taking off
AI assistant reshapes developer productivity
Educational reform keeps pace with technological innovation

From the predictions and social development in recent years, it is clear that Amazon Cloud Technology has a foresight for the future and has been actively responding to future development, developing its own chips, assisting education and women's development, accelerating AI innovation, etc.

Event Preview: 2023 Amazon Cloud Technology re:Invent China Tour - 10 city tour is coming

In order to further bring the essence of 2023 Amazon Cloud Technology re:Invent and experience to Chinese customers and cloud computing enthusiasts, we are specially organizing the 2023 Amazon Cloud Technology re:Invent China Tour event! The city tour will cover 10 cities including Beijing, Shanghai, Guangzhou, Shenzhen, Chengdu, Qingdao, Nanjing, Xi'an, Hangzhou and Changsha. We sincerely invite you to experience the latest technologies and products, use AI solutions to improve your skills and build new applications, and achieve industry empowerment to obtain practical results. A global event, shared by China, Amazon Cloud Technology looks forward to your visit!

2023 Amazon Cloud Technology re:Invent China trip re:Invent 2023 keynote speech sharing

In the 2023 Amazon Cloud Technology re:Invent China City Tour, we will also invitemysterious guests to give you a panoramic view Interpret the many major releases and application innovations of Amazon Cloud Technology re:Invent in 2023! Want to keep up with the big names and fully explore the frontiers of the conference? Sharing is not to be missed, so stay tuned!

2023 Amazon Cloud Technology re:Invent China Tour Unlocks Hidden Easter Eggs in Beijing Station

During the city tour, Beijing Station also carefully prepared an early-bird course on getting started with generative AI for business/technical decision-makers. This course will provide you with an overview of generative AI, as well as methods for planning generative AI projects and building a generative AI-ready organization. Participants can learn what generative AI is, how it solves executive concerns and problems, and how it can power business growth and revolutionize a wide range of industries.

Touring exhibitions in 10 major cities, scan the QR code to join the "group" immediately!

Starting from December 12, 2023, the 2023 Amazon Cloud Technology re:Invent China City Tour will be launched in 10 major cities, covering 10 cities: Beijing, Shanghai, Guangzhou, Shenzhen, Chengdu, Qingdao, Nanjing, Xi'an, Hangzhou, and Changsha ,Looking forward to your arrival!

For more exciting content, please see Amazon Cloud Technology official account, search on WeChat: Amazon Cloud Technology

Reference information