Skip to content

Documentation update log

2026-10-11

  • Update: History of CARC computing has a new section on the Maui High Performance Computing Center (1993–2001), which UNM ran for the Department of Defense and whose funds built the Albuquerque center. It covers the center's TOP500 systems, UNM's 2025 purchase of the MHPCC building, and six new timeline rows, with the MHPCC logo and sign added to assets/images/history/. The news releases and articles reposted on David A. Bader's site are now cited by original publisher, title and date, with links to the archived originals. They add Roadrunner's 1998 prototype and funding, the 2001 NSF phylogeny grants, the 2003 IBM PERCS collaboration and the 2003 HPC@UNM steering committee, and a reported $1.5 million cost for Los Lobos.

2026-10-10

  • Update: The web archive pages are now stored as UTF-8 with LF line endings. 186 of the 1990s pages were Latin-1 or Windows-1252, which Cascade changes on upload, so the carc.unm.edu publish stopped at its read-back check. scripts/build_web_archive.py now writes UTF-8 itself.

  • Update: History of CARC computing now has a "Capacity over time" chart and table under The machines: total peak performance, CPU cores, memory and storage of every machine in service each year from 1995 to 2026, built by scripts/build_capacity_timeline.py (many inputs are estimates, listed under the chart). Tyson Swetnam's start date as Director is now Jul 1, 2026.

  • Creation: Added the Web archive of the old center websites (draft): ten browsable snapshots of the HPCERC, AHPCC, Alliance Supercluster, HPC@UNM and first CARC websites (1998–2016), about 820 pages rebuilt from Internet Archive captures by scripts/build_web_archive.py. Pages keep their original HTML with an archive banner; links and images point at the local copies, and anything not saved points at the nearest Wayback Machine capture. The history page links to it.

  • Creation: Added History of CARC computing (draft), a sourced record of every supercomputer in the Galles Building machine room from Roadrunner and Los Lobos (1999–2000) through the HPC@UNM machines on the machine-room wall plaques (Pequin, Poblano, Ristra, Hammer, Yenta, Vista Azul, Delphi, Rio Bravo) to Easley. It covers install and retirement dates, manufacturers, grants, peak performance and costs (estimates and unknowns are marked), a timeline, past directors and associate directors, and the PIs who contributed hardware. The plaque photo and 17 logos and photos from Internet Archive snapshots of the old AHPCC, HPCERC, Alliance, HPC@UNM and CARC sites are in assets/images/history/; those snapshots also supplied the 1995–1998 IBM SP and SGI Origin machines, the 2000–2007 leadership and the 2008 and 2015 system tables.

2026-09-29

  • Update: Added static notation for the governing equation, boundary conditions, and uniform-load midpoint deflection in the Mathematica on Easley beam example, plus measured Easley scaling results for 128 symbolic calculations.
  • Update: Retitled the worked example in Mathematica on Easley as "Example: Beam Deflection."
  • Update: Published the Mathematica on Easley examples as a ZIP archive containing the Slurm scripts, Wolfram Language inputs, and kernel launcher.
  • Update: Moved the Mathematica launcher permission check into each Slurm script. The archive setup now only downloads and extracts the ZIP.
  • Correction: Restored the math-kernel.sh helper to the Mathematica on Easley examples. Its process-based license file is required by the Mathematica worker-launch workflow.
  • Update: Added the complete parallel.wl, distributed.wl, and gpu.wl sources to Mathematica on Easley, alongside the existing beam model, Slurm scripts, and downloadable archive.
  • Update: Removed the duplicate page table of contents from QuickByte Tutorials; its expandable sidebar groups are now the single navigation layer.
  • Update: Replaced Mathematica on Easley with the CARC beam-deflection tutorial and its complete, downloadable example archive: shared license launcher, Wolfram Language inputs, and single-node, multi-node, and GPU Slurm scripts.
  • Correction: Restored the confirmed Easley inventory: 9 L40S nodes, providing 36 NVIDIA L40S GPUs (44 GPUs total with the H100 GPUs).
  • Update: Split QuickByte Tutorials' programming-environment menu into Conda, JupyterHub, Julia, R, and MATLAB headings. The Conda guide now uses the current Miniconda name rather than the retired Anaconda module name.
  • Update: Split the scientific tutorial categories in QuickByte Tutorials into Bioinformatics, Chemistry & materials, Computational immunology, and Astronomy in both the sidebar and the contents page.
  • Update: Merged the former Software and Tutorials navigation into QuickByte tutorials. Its scrollable contents page now gives each active domain tutorial a home alongside the software guides.
  • Creation: Added Mathematica on Easley, a curated QuickByte covering interactive, serial, multicore, multinode, GPU, and license-server workflows with Mathematica 15.0.1 and WolframScript.
  • Update: Refreshed the systems overview from Slurm inventory collected on September 29. It distinguishes all nodes from scheduled compute nodes: Easley has 65 total nodes and 4,160 CPU cores (63 scheduled compute nodes, 8 NVIDIA H100 GPUs, 36 NVIDIA L40S GPUs, and 22.4 TB of compute-node RAM); Hopper has 68 total nodes and 2,176 CPU cores (67 scheduled compute nodes, 26 NVIDIA A100 GPUs, 2 NVIDIA V100 GPUs, and 15.1 TB of compute-node RAM). The page now lists each cluster's heterogeneous node configurations, ConnectX adapters, and switch-fabric capacities.
  • Update: Corrected the storage capacities in the systems overview: 385 TB of GPFS scratch, 1.2 PB of BeeGFS working scratch, and 133 TB of NetApp enterprise storage.

2026-09-12

  • Update: Made the Markdown behind every page easier for AI agents to find, following the conventions of the DUST 2026 site: each rendered page now carries a "View this page as Markdown" button beside the edit and view-source buttons and a "Machine-readable versions" line at the end of the article (Markdown twin, raw source on GitHub, llms.txt, llms-full.txt); the site footer links llms.txt, llms-full.txt, and the agent guide; llms.txt is now built from the site nav and lists the Markdown twin and raw GitHub source of every page plus the corpus size; llms-full.txt and the per-page Markdown mirror have relative links rewritten to absolute URLs; robots.txt names the raw-source convention; and the agent guide explains the raw-source fallback for sandboxes that cannot reach carc.unm.edu and warns that <head> tags are invisible to text-extracting fetchers. Scripts now share scripts/okf_common.py.

2026-09-08

2026-09-01

  • Update: Reconciled the corpus against the CARC knowledge store — staff-reviewed articles distilled from real support tickets plus cluster state observed directly over SSH on 2026-07-25. Corrected stale facts: Slurm intro now lists Easley's real partitions (debug is 1 hour, not 4; there is no Easley condo partition) plus the group-gated h100/l40s GPU partitions and the default time/memory behavior; resource limits (now curated in-repo) carries observed per-cluster partition tables — the old Hopper table was a hardware generation stale — full storage quotas with file-count limits, and the job time-limit extension policy; storage and backups (now curated in-repo) documents the real paths (/carc/scratch, /easley/scratch, /projects), the 180-day Easley scratch cleanup, that Hopper has no machine-local scratch, and how to find a project's path; the retired anaconda3 module was replaced with miniconda3 across conda environments, getting R, and deep-learning packages; getting R and Gurobi with R now show the current r/4.x module tree; and modules no longer claims plain intel modules don't exist.
  • Creation: Added CUDA-aware MPI — passing GPU device pointers directly to MPI and fixing the mixed OpenMPI/UCX environment segfault, distilled from a staff-reviewed, user-confirmed ticket resolution.

  • Update: Folded ticket-derived answers into the FAQ and guides: troubleshooting covers the uid not in group permitted to use this partition rejection and the ColdFront path to GPU-partition access, jobs stuck in CG, a full out-of-memory diagnosis (per-job memory enforcement, DefMemPerCPU defaults, calibrating with seff), Python ModuleNotFoundError after loading miniconda3, quota surprises (group-ownership accounting, stale warnings), hanging transfers, and password-reset propagation; the general FAQ answers where project scratch lives, how to keep data past scratch retention, GPU-partition gating, and allocation expiration/renewal; password reset and transferring data carry the matching tips; getting started tells PIs to keep allocations current.

  • Update: The agent guide now cross-links the GPT 101 generative-AI workshop — an OKF v0.2 bundle with the same llms.txt conventions, maintained and taught by CARC.

2026-08-30

  • Update: System-status links now point straight at the live monitors — the cluster login & website status board, UNM IT alerts, perfSONAR network performance, the Easley DNS check, and XDMoD usage metrics — instead of the intermediary carc.unm.edu downtime page. The landing button and troubleshooting steps use the cluster status board; the support card and systems overview list all five.

  • Update: The header logo is now a Googie starburst — the same 12-ray construction as the homepage hero's atomic bursts (alternating ray lengths, tip dots, cycling colors), in the cream/turquoise/white subset that reads on the cherry header.

  • Update: Completed the retirement of Wheeler, Taos, Gibbs, and Xena across the corpus. The "Legacy content" admonitions are gone, and every active page now reads against the current clusters: hostnames and prompts point at Hopper, retired-only sections were removed (the Wheeler/PBS Orca variant, the Xena AlphaFold script, Xena-specific partition flags), and the historical Xena and Wheeler queue tables moved from resource limits into the legacy cluster reference. The migration pipeline now enforces this: it fails if a retired system name appears outside the sanctioned legacy pages (provenance frontmatter and the changelog stay truthful).

  • Update: Converted the remaining PBS-era material on active pages to Slurm (storage example script and wording, R package installs interactive-session request) and fenced all file paths on the storage page. Fixed the SSH config example (now a single well-formed block covering Hopper and Easley).

  • Update: Footer social links: removed the X/Twitter icon (account no longer exists) and pointed YouTube at the main UNM CARC channel.

2026-08-29

  • Update: Made the deployed site directly consumable by AI agents: every page's Markdown source (OKF frontmatter intact) is now served at its URL plus index.md; rendered pages advertise it via link rel=alternate and okf:* meta tags (type, status, trust tier, generated-at); robots.txt points crawlers at llms.txt, the full corpus, and the mirror convention (scripts/postbuild_agent_surface.py, wired into CI). Added the For AI agents guide and a repository AGENTS.md/CLAUDE.md for coding harnesses.

  • Update: Embedded CARC YouTube recordings across the site: the Video tutorials page now carries the full QuickBytes playlist, CARC Annual Meeting 2025 talks, and research presentations from the UNMCARC channel; ten guide pages (logging in, Slurm intro, storage, transfers, modules, conda, X11, GNU Parallel, SimCov, parallel R) embed their matching walkthrough via the migration pipeline.

  • Creation: Rebuilt the Workshops and slides page (previously a stub) as a curated catalog of 33 slide decks: the Introduction to CARC series, domain-focused workshops, course guest lectures, and legacy material.

  • Update: The Xena cluster has been retired. Removed Xena from the Systems overview, Facilities description, FAQ, and landing page; added it to the retired-systems list in Cluster specifications; marked Xena-specific GPU guides (PyTorch, MATLAB GPU/deep learning, deep-learning packages) as status: draft with legacy notices pending review against Hopper and Easley GPUs.

  • Update: Annotated all code across the corpus: tab-indented QuickBytes code now renders as language-fenced blocks with syntax highlighting; restructured Installing deep learning packages (now curated in-repo); annotated inline code references in Parallel R with the future package.

  • Creation: Added the Interactive computing section (Open OnDemand, JupyterHub) and the FAQ section (General FAQ, Troubleshooting).

  • Creation: Added Contributing to these docs — the OKF frontmatter contract, verification workflow, and local build instructions for CARC staff.

  • Update: Added machine-readable llms.txt and llms-full.txt indexes generated from OKF frontmatter (scripts/gen_llms_txt.py, enforced in CI); moved the page table of contents into the left sidebar.

  • Initialization: Created this documentation bundle with Zensical, structured as an Open Knowledge Format (OKF v0.2) knowledge bundle.

  • Migration: Migrated 56 tutorials and guides from UNM-CARC/QuickBytes and UNM-CARC/webinfo with provenance frontmatter (generated, sources, per-file last_modified from git history). All migrated pages are unverified pending CARC staff review.
  • Creation: Wrote the Getting started overview, Good Neighbor Use Policy, Systems overview, Video tutorials, Getting help, Acknowledging CARC, Mission and vision, Facilities description, and Partner cyberinfrastructure pages from carc.unm.edu content.
  • Deprecation: Marked Cluster specifications (retired Wheeler, Taos, and Gibbs systems) and R batch jobs with PBS as deprecated; both are kept for history and links.

Machine-readable versions of this page: Markdown twin · raw source on GitHub · llms.txt · llms-full.txt (whole site). See For AI agents.