This month we're highlighting the last chance to register for the Sixth Chameleon User Meeting, a new webinar recording from UTEP's MINCER team, platform updates including multi-instance GPU support, new cloud traces, switch performance improvements, and several testbed usability enhancements.
Data centers waste enormous performance potential because software evolves faster than profiles can be collected. Wax, developed by Tawhid Bhuiyan and colleagues at Columbia University, uses source code and debug information to recover 65–93% of fresh-profile performance gains from stale profiles—outperforming the state of the art by up to 7.86 percentage points across five real-world data center applications.
Chameleon's stitch port feature lets you connect bare-metal GPU servers directly to FABRIC's network fabric — no public IP required. This post walks through a Trovi artifact that provisions a Chameleon GPU node, links it to a FABRIC slice over fabnet, and runs LLM inference across the stitched network. Grab the notebook and adapt it to your own cross-testbed workflows.
This April, we're bringing together researchers, educators, and infrastructure builders for two days of honest conversation about AI on research computing infrastructure. We're thrilled to kick things off with a keynote from one of HPC's leading voices.
Teaching cloud computing without tying students to a single vendor is harder than it sounds. Dr. Michael MacLeod at Austin Community College found a solution in Chameleon, and the results speak for themselves: 14 courses, 420 students, and a curriculum that's still growing. Read more in the blog.