You can’t avoid AWS re:Invent, with its 50,000+ attendees, now one of the biggest cross-industry (despite its AWS focus) events for cloud-hungry enterprise IT teams. My inbox is overflowing with announcements from AWS and dozens of its partners.

AWS keeps a comprehensive running list of announcements that it makes on its News Blog, but here are a handful of the ones I found most interesting:  

Generative AI

Generative AI is a cloud-first endeavor with the CSPs locked in a heated race to stay at the tip of the competitive spear. AWS is no different, though it tells one of the best AI stories in public cloud.

AWS long had its Trainium and Inferentia sillcon for training and inference, respectively. Its BedRock and GenAI-focused Q service have been around since 2023. This year is all about building on that momentum:

Amazon Bedrock now offers two new evaluation capabilities to streamline testing and enhance generative AI applications: RAG evaluation and LLM-as-a-judge, and is also adding custom connectors (APIs) and streaming data ingest capabilities:

• RAG evaluation (preview) in Amazon Bedrock Knowledge Bases uses LLMs to assess Retrieval Augmented Generation applications

• LLM-as-a-judge (preview) feature provides automated, humanlike model evaluations to evaluate AI applications across dimensions like correctness, helpfulness, and responsible AI criteria, with intuitive results including natural language explanations and normalized scores for easy interpretation. This one’s very cool.

• Custom Connectors all data integration from unsupported sources, like Google Drive and similar third-party platforms, without needing intermediate steps, eleiminating the requirement to transfer data to supported storage services (e.g., Amazon S3) before ingestion.

• Streaming Data Ingest now supports real-time data sources, such as news feeds or IoT sensor data, allowing developers to directly ingest streaming data into knowledge bases.

Amazon Q, AWS’s GenAI-powered assistant, gets a few new features to help streamline workflows:

• Increased Meda Support, extracting insights from visual elements embedded in documents, such as diagrams, infographics, charts, and image-based content.

• Integration with Web Browsers and Productivity Tools, enabling users to access its capabilities directly within those applications, eliminating the need to switch between platforms.

Compute

AWS EC2 pretty much defined public cloud compute, and it continues to lead the industry with new technology and feature integration. The company announced a couple of new storage-optimized instances that each leverage the latest 3^rd Generation AWS Nitro SSDs:

• EC2 I8g Graviton4 Storage-Optimized Instances: A Graviton4-powered storage-optimized instance type for high-performance, low-latency storage workloads like transactional and NoSQL databases and real-time analytics – anywhere storage performance impacts the workload its supporting. It has a nice set of configuration options, supporting up to 96 vCPUs, 768 GiB of memory, and 22.5 TB of local NVMe SSD storage. AWS showed some impressive benchmarks which we may take a closer look in a more in-depth piece.

• EC2 I7 Xeon-powered Storage-Optimized Instances: The new instance is targeted at I/O-intensive workloads like NoSQL databases, distributed file systems, and analytics, and support up to 192 vCPUs, 1.5 TiB of memory, and network speeds up to 100 Gbps. Advanced features like AVX-512 VP2INTERSECT for machine learning, AMX for deep learning, and 3x the EBS bandwidth enhance their versatility. Amazon says that this is “the highest storage density in the cloud” (we haven’t fact-checked that one).

Storage

It’s a truism that compute is only as fast the underlying storage (just ask the guys over at SNIA!), and AWS does a nice job of maintaining performance parity with nearly everyone in the storage industry, on-prem or cloud. That continues with some of today’s announcements:

• Amazon FSx Intelligent-Tiering for FSx for OpenZFS, addressing the needs of customers migrating large on-premises data sets to the cloud. Data stored is fully elastic, requires no pre-provisioning, and remains instantly retrievable across multiple AWS AZs with up to 400K IOPS and 20 GB/s throughput per file system. We’re big fans of storage tiering to help control storage costs.

• AWS Data Transfer Terminal is a secure physical location for rapid data uploads to the AWS Cloud. The first terminals are in Los Angeles and New York, with plans for global expansion. Customers can reserve time slots to upload large datasets directly to AWS services like S3 or EFS via high-throughput connections, significantly reducing data ingestion time. This also supports ingest from AWS Snowball devices. This is a much nicer, though far less cool, way of importing data than AWS’s old Snowball Snowmobile 18-wheeler. Time marches on.

• Storage Browser for Amazon S3, an open-source UI component that developers can integrate into web applications, allowing authorized end users to browse, upload, download, copy, and delete data stored in Amazon S3. This one’s been needed for a while, and, with object storage fast becoming more dominant for streaming data and archive applications, it’ll be welcome.

More to Come

This isn’t all that AWS announced. The company made a wide range of announcements covering everything from enhanced Kubernetes and contain support to a richer array of observability options, with a few things in between. We’re still digesting some of that and will write about the bits we find most interesting.

There’s more to come this week from AWS and its partners, and, once the event wraps, we’ll take a more detailed look at Amazon’s most impactful announcements. Stay tuned.