The University of Nebraska–Lincoln hosted the HEPiX Fall 2026 Workshop from Oct. 19–23 in the Willa S. Cather Dining Complex, in partnership with the Holland Computing Center and the Office of Research and Innovation.
HEPiX, founded in 1991, is the twice-yearly forum where the IT staff who run computing for high-energy and nuclear physics laboratories compare notes. This year's meeting drew participants from CERN, KEK in Japan, the Rutherford Appleton Laboratory in the United Kingdom, PIC in Spain, the Swiss National Supercomputing Centre, Brookhaven and Fermi national laboratories, ESnet, more than a dozen U.S. universities, and industry partners IBM and Lenovo, for roughly 50 talks over five days.
The Holland Computing Center 's Garhan Attebury chaired the workshop and led the local organizing team of Caughlin Bohn, Carl Lundstedt, and Natasha Pavlovikj. Lenovo and AMD sponsored at the gold level, and Dell and Advanced HPC at the silver level. Slides from every talk are public on the workshop's Indico page.
Nebraska on the program
Attebury opened the site-report session with the state of Nebraska's Tier-2 center for the CMS experiment, about 16,000 CPU cores and 16 petabytes of storage that help process data from CERN's Large Hadron Collider, and the story of moving that data center across campus this fall in the days between home football games.
Derek Weitzel, research associate professor in the School of Computing and Holland Computing Center presented the National Research Platform, a Kubernetes cluster spanning 129 sites with about 1,400 GPUs that HCC helps operate. His talk covered how NRP serves open large language models to researchers, how it uses LLMs to help users debug their own jobs, and what it has learned about security on a cluster shared by dozens of institutions.
The big themes
- Preparing for the High-Luminosity LHC. The first experiments for the High-Luminosity LHC begin in 2030, with the recorded particle collision events are expected to need two to four times today's processing, two to three times the disk, and twice the tape. A worldwide data challenge in February 2027 will rehearse at half that scale.
- Hardware is getting more expensive, not cheaper. Speakers from CERN and Brookhaven described a market in which AI data centers set the pace. The price CERN pays for server memory has risen four-to-six-fold since 2023, its storage cost per terabyte rose 68 percent in a year, and memory and hard-drive makers have sold out production through 2027 and 2028. The community's long-standing assumption that computing gets cheaper over time may no longer hold, and labs are responding by keeping hardware in service for a decade or more.
- AI in both directions. A new track on applied AI showed labs hosting their own LLM services, using AI to help operate infrastructure, and building benchmarks to test AI coding agents on their own software. The same tools are flooding security teams: KEK reported 86 percent more published vulnerabilities in 2026 than in 2025, much of it from AI-driven vulnerability mining.
- Energy and sustainability. CERN's new Prévessin Data Centre runs well below the industry average for power overhead and recovers four megawatts of waste heat for district heating. Sites are beginning to account for the carbon cost of individual compute jobs.
For the School of Computing
Strip away the physics and the week was computer science at scale: distributed systems, Kubernetes, exabyte storage, terabit networking, security, and AI operations. The Holland Computing Center does this work on campus, and students, staff, and faculty can use the National Research Platform, including its GPUs and hosted LLM service, free of charge for research and education.
The full program and slides are at indico.cern.ch/event/1679145.
