Rhett Butler via Flickr

Welcome to the November Chameleon newsletter! This month brings exciting updates including enhanced hardware browser integration, CHI@Edge stability improvements, and expanded support for distributed training on AMD GPUs, along with announcements about upcoming webinars and our sixth user meeting at NCAR’s campus in Boulder.

Changelog

Changes to testbed capabilities and services this month.

New Hardware Browser Integration. The hardware browser is the tool that lets you easily search and explore all of the hardware resources across Chameleon’s baremetal sites, letting you find nodes with e.g., specific CPU or GPU models. This month, we are excited to announce improvements in how the hardware browser integrates with these sites. From the node detail view in the hardware browser, you can simply click a button to check that node's availability or to reserve it. We have also improved the site availability calendar, which shows when each node is reserved, so that clicking on a node name will automatically fill in the lease create form to reserve it. Additionally, we fixed an issue where deep links did not work if you were logged out of a site. Now, links to specific pages (such as the site calendar) will redirect you properly after you log in.

CHI@Edge improvements. We’ve been working to improve the stability and usability of CHI@Edge over the last few months, and wanted to highlight a few of these improvements. First, we’ve fixed issues with CHI@Edge floating IP traffic, where sometimes traffic wouldn’t flow to newly associated IPs.. Additionally, we’ve fixed several cases where containers would move to an “error” state instead of launching correctly, particularly when peripherals are mounted. Lastly, we’ve migrated the CHI@Edge control-plane to a new, more powerful, controller node with fast disks to resolve cases where the kubernetes backend was timing out.   Internally, we’re now deploying the k3s control-plane with chi-in-a-box + kolla-ansible, improving isolation and our ability to update configuration and versions without downtime.

Updated AMD ROCm image to support distributed training. CHI@TACC has a wide variety of hardware, with 18 node types. This includes 8 gpu_mi100, which are the only nodes with AMD GPUs on Chameleon, with each node having 2 MI100 GPUs. Similar to our CUDA images for Nvidia GPUs, we support a CC-Ubuntu24.04-ROCm image which is preconfigured with settings and software to work with these MI100 GPUs. This month, we’ve updated the ROCm image, setting the IOMMU to the recommended passthrough mode, which makes it easier to run distributed training on these GPUs.

Upcoming Webinars

The Chameleon team hosts webinars on a variety of topics ranging from how to use the testbed to showcasing specific tools, research workflows, and educational projects that users built on Chameleon.

Keep an eye on our webinar calendar as we announce new webinars for upcoming months.

Community News & Resources

Save the Date! Sixth Chameleon User Meeting on April 15-16, 2026 (Boulder, CO): Our sixth user meeting will take place at the National Center for Atmospheric Research (NCAR) in Boulder, CO this year! The Chameleon User Meeting is an in-person forum held over two days for users to discuss their research and education projects, share experiences of working with the Chameleon testbed, solve challenges together, and propose new features that will make their experiments and education projects easier. Mark the dates on your calendar and watch our webpage for incoming updates.

Machine Learning Webinar (Recap): Fraida Fund presented her work running a 190-student class on machine learning operations (MLOps) to a group of educators and Chameleon users on Nov. 25. We had almost 100 sign-ups and close to 50 attendees. Watch the presentation here and download the slides.

Chameleon at SC25: Four students who worked with us over the summer (Saieda Ali Zada, Hudson Reynolds, Alex Tuecke, and Zahra Temori) presented posters at Supercomputing (SC) 2025 this year. Read about their work in our student highlight. Kate Keahey also presented a paper (co-authored with Fraida Fund from NYU) highlighting Fund’s open-source material for teaching machine learning operations (MLOps) at scale.

Tips&Tricks Blog of the Month: From GitHub to Publication: Using Trovi Effectively | Chameleon

User Experiment Blog of the Month: Introducing MINCER’s Performance Measurement and Reproducibility Appliance | Chameleon 

Upcoming Maintenance

Nothing to announce!