Issue # 130: July 4, 2025 to July 18, 2025

Hey Reader.

Welcome to Issue # 130 of AWS Graviton Weekly, which will be focused on sharing everything that happened in the past week related to AWS Silicon: from July 4, 2025 to July 18, 2025.

Before continue, I just wanted to say thank you to the incredible CAST.AI team for sponsoring our work here.

You guys rock, and we are more than happy to continue spreading the world about the platform.

Thanks again for helping us to become the second place where anyone can be informed about AWS Silicon.

Back to the usual content.

Enjoy.

Recommendation of the week: Cut your Kubernetes costs in half with CAST.AI

Book a demo

NEWS

ARTICLES AND TUTORIALS

Amazon EKS enables ultra scale AI/ML workloads with support for 100K nodes per cluster

Under the hood: Amazon EKS ultra scale clusters, by Shyam Jeedigunta (AWS); Apoorva Kulkarni (AWS) and Raghav Tripathi (AWS)
Introducing Kiro, by Deepak Singh (AWS) and Nikhil Swaminathan (AWS)

Accelerate generative AI inference with NVIDIA Dynamo and Amazon EKS, by Baladithya Balamurugan (AWS), Anton Alexander (AWS), Arun Raman (NVIDIA), Kshitiz Gupta (NVIDIA), Wenhan Tan (NVIDIA), Brian Kreitzer (AWS), Eliuth Triana Isaza (NVIDIA), Harish Rao (AWS), and Jiahong Liu (NVIDIA)

Scaling to the Future: Kubernetes Reimagined with Graviton Processors, by Sukhpinder Singh

Carlos Manzanedo Rueda and Arthur Petitpierre from AWS will be in London and are planning to run a "Meet the EC2/Performance expert" session on upcoming August 1st.

Here's the description of the session:

Getting the best compute performance is a bit of an art, and knowing about low level details can help a lot. This session will provide a small group an opportunity to talk to EC2 Specialist and Performance experts. They will cover all the details you’ve always wanted to know about, such as memory topology, how to get access to hardware counters, how to take into account the side-effects of hyperthreading, and how to properly run performance tests and optimize your latency. The session will help attendees to build a mental model on modern compute and EC2 differentiators and best practices on how to use tools like aperf and techniques like Utilization Saturation and Errors (USE), Top-down Microarchitecture analysis (TMA) and what are the actionable things that you can do as a developer in the iterative cycle of performance tuning.

Feel free to reach Carlos on LinkedIn on this post