Categories – Featured

Chameleon Newsletter & Changelog June 2026

Welcome to the Chameleon June 2026 Newsletter!

This month we're reminding you not to miss out on limited-time Ponte Vecchio (Intel GPU) nodes and the plentiful hardware available at CHI@NCAR, previewing a redesigned Resource Discovery browser, adding ccauth support to python-chi, improving long-running commands on CHI@Edge, retiring the vendor-data service, and refreshing our documentation. In the community, Kate Keahey delivered a keynote at the ISC High Performance 2026 Reproducibility Workshop.

I/O Analysis is All You Need: An I/O Analysis for Long-Sequence Attention

An I/O-driven accelerator that achieves up to 3.4× speedup over FlashAttention-2 by minimizing data movement in long-sequence self-attention

Modern AI systems are often limited not only by how much computation they perform, but also by how much data they need to move. AttenIO, developed by Xiaoyang Lu at Illinois Institute of Technology, is an I/O-driven accelerator for exact long-sequence self-attention that achieves up to 3.4× speedup over FlashAttention-2 by applying principled I/O analysis to minimize data movement across the memory hierarchy.

Redesigned Chameleon Resource Discovery in Preview Now

Resource discovery on Chameleon now supporting availability-aware browsing and more!

Chameleon's redesigned resource discovery service is now in preview, offering availability-aware browsing across all CHI sites. Find free bare metal nodes in real time, generate ready-to-run reservation commands, and get conflict detection — all in one unified interface.