Sign in
AI Opt-Out Mechanics & Self-Service HPC
workFree

AI Opt-Out Mechanics & Self-Service HPC

Thursday 1 October 2026, 18:00Venue time (Budapest)
High Five BudapestDirections
FreeTickets sold by meetup.com
Book · Freevia meetup.com
Add to calendar

Downloads an .ics file · Times are in Europe/Budapest · Google Calendar

Join us for a night of cloud guardrails and supercomputing at scale. We’re unpacking the fine print and resolution mechanics of AWS’s 31-service AI Opt-Out scope, paired with a post-mortem-heavy deep dive into self-service HPC—covering the multi-node MPI failures, mount races, and run-away costs that only happen on real production setups. Pizza and beer included! 🍕🍺 Agenda 18:00 - Arrival 18:15 - One Policy, Thirty Services: What Changed in AWS's AI Opt-Out Scope 18:50 - Building a Self-Service HPC Platform on AWS 19:25 - Networking, Pizza, Drinks One Policy, Thirty Services: What Changed in AWS's AI Opt-Out Scope Nico Herzhauser In 2025, the AI services opt-out policy covered a handful of AI services. Today the official list counts 31 — CloudWatch, GuardDuty, Security Hub, Glue, DMS, and more. This is a Level 300/400 walk through one AWS Organizations policy most accounts never set: how it resolves, why it is not an SCP, what happens to your data with and without it, and where it silently does nothing. You leave with a checklist, a verification script, and a clear view of the edge cases the policy never reaches. Building a Self-Service HPC Platform on AWS István Szukács HPC in the cloud promises elastic supercomputing on demand, but the gap between "a cluster launched" and "a cluster that actually runs your job" is where the hard lessons live. This session walks through building a self-service HPC platform on AWS — ParallelCluster/PCS, Graviton compute, Lustre scratch, and EFA networking, provisioned and torn down on demand with full cost controls. Then we deconstruct the failures that only surface on real infrastructure: a security-group rule that silently broke every multi-node MPI job over EFA, nodes accepting work before their filesystems mounted, and clusters that billed forever because an expiry date was written but never read. We share how each was root-caused live, the guardrails that now prevent them, and the Graviton price-performance trade-offs that make the whole thing worth it.

See and connect with other Palaner members who are going

Sign up to see who else is going and to connect with them
Ask Palaner

Going to AI Opt-Out Mechanics & Self-Service HPC? Ask me anything about it.

I read the organiser's pages and answer in a few seconds.

Answers are AI-generated · Privacy