Jekyll2026-07-16T14:17:42-04:00https://www.excl.ornl.gov/feed.xmlWebsite for ExCL.{"name"=>nil, "avatar"=>nil, "bio"=>nil, "location"=>nil, "email"=>nil, "googlescholar"=>nil, "orcid"=>nil, "bitbucket"=>nil, "github"=>nil, "linkedin"=>nil}2026-07-16T00:00:00-04:002026-07-16T14:17:34-04:00https://www.excl.ornl.gov/excl/july-meeting-2026<p>July 2026 ExCL meeting slides.</p> <!--more--> <object data="/assets/presentations/2026-07-july-excl-meeting.pdf" type="application/pdf" width="560" height="700"><p>It appears that you don't have a PDF plugin for this browser. You can <a href="/assets/presentations/2026-07-july-excl-meeting.pdf">click here to download the PDF file.</a></p></object> <hr /> <p>Below is a LLM-generated summary from the meeting slides.</p> <hr /> <h1 id="-excl-monthly-update--july-2026">🚀 ExCL Monthly Update — July 2026</h1> <p>This month’s meeting covered Claude Code usage on ExCL systems, the completed integration of the Maxwell system, an in-progress OpenMPI deployment, an operations update, and an open discussion on access and tooling.</p> <hr /> <h2 id="-claude-code-in-excl">🤖 Claude Code in ExCL</h2> <ul> <li>See the <a href="https://docs.excl.ornl.gov/quick-start-guides/claude-code">Claude Code Quick-Start Guide</a> in the ExCL User Docs</li> <li>A dedicated VM for running Claude is coming soon; in the meantime, launch Claude sessions on Explorer and use Slurm to reach specific accelerators</li> <li>Claude tends to launch sub-agents and other long-running jobs that can persist after exiting the TUI, so keep that in mind when closing claude</li> </ul> <hr /> <h2 id="️-maxwell">🖥️ Maxwell</h2> <ul> <li>Maxwell (formerly ORC HYP100/HYP65) has completed its move into ExCL <ul> <li>4x Nvidia A100, 4x AMD MI100</li> <li>255 HT cores (one core failure), 1TB memory, 3TB /scratch storage</li> <li>Fully integrated into ExCL authentication, file services, and Slurm</li> <li>Standard access applies — free for all unless someone requires solo access for performance measurement</li> </ul> </li> </ul> <hr /> <h2 id="️-openmpi-deployment">🛠️ OpenMPI Deployment</h2> <ul> <li>Generic build is complete and being deployed <ul> <li>The IRIS project is the initial driver for this deployment</li> </ul> </li> <li>Hardware-specific (accelerator) builds are next; each hardware/software combination requires a separate build — expected to take time but not be problematic</li> <li>The module system will handle placing the correct build in your path</li> <li>UCX deployment is under way; RoCE is generally supported on ExCL hardware, though some network provisioning work remains</li> </ul> <hr /> <h2 id="️-operations">⚙️ Operations</h2> <ul> <li>With few exceptions, all research systems have been updated to Ubuntu 24.04, simplifying Slurm deployment</li> <li>All host operating systems are up to date</li> <li>Migration to new hypervisors will happen over the next few weeks; minimal user impact expected outside of migration windows, and this will also offer an alternative to home Claude VMs</li> <li>DR tape backups of home directories are currently not working (an ITSD issue under investigation); backup to HA disk storage continues to work as expected, with no change to snapshot operations <ul> <li>As always, keep code in version control and pushed to a remote repository</li> </ul> </li> <li>Outages will continue on the first Tuesday of each month as needed</li> </ul> <hr /> <h2 id="-questions--projects--comments--discussions">💬 Questions / Projects / Comments / Discussions</h2> <ul> <li>Recordings are available via the link in the newsletter</li> <li>Is there interest in Ubuntu 26.04 LTS support?</li> <li>Questions on license use.</li> <li>Request for a quick load/user-count command (pdsh-like tool)</li> <li>Interest in Parsec access in ExCL — individual user vs. team access, and cost</li> </ul> <hr /> <h2 id="-summary">📌 Summary</h2> <p>July’s meeting introduced Claude Code usage guidance on ExCL, confirmed Maxwell’s full integration into the lab, previewed the in-progress OpenMPI/UCX deployment, gave an operations update (Ubuntu 24.04, hypervisor migration, a DR tape backup issue), and closed with user questions on LTS support, licensing, tooling, and Parsec access.</p>{"name"=>nil, "avatar"=>nil, "bio"=>nil, "location"=>nil, "email"=>nil, "googlescholar"=>nil, "orcid"=>nil, "bitbucket"=>nil, "github"=>nil, "linkedin"=>nil}

July 2026 ExCL meeting slides.

2026-06-18T00:00:00-04:002026-06-18T14:27:17-04:00https://www.excl.ornl.gov/excl/june-meeting-2026<p>June 2026 ExCL meeting slides.</p> <!--more--> <object data="/assets/presentations/2026-06-june-excl-meeting.pdf" type="application/pdf" width="560" height="700"><p>It appears that you don't have a PDF plugin for this browser. You can <a href="/assets/presentations/2026-06-june-excl-meeting.pdf">click here to download the PDF file.</a></p></object> <hr /> <p>Below is a LLM-generated summary from the meeting slides.</p> <hr /> <h1 id="-excl-monthly-update--june-2026">🚀 ExCL Monthly Update — June 2026</h1> <p>This month’s meeting featured spotlights on data versioning tools for ML workflows, an operations and procurement update, a hardware redeployment note, and an open discussion on performance-monitoring access.</p> <hr /> <h2 id="-data-versioning-tool-spotlights">📁 Data Versioning Tool Spotlights</h2> <p><strong>DVC (Data Version Control)</strong> — <a href="https://dvc.org/">dvc.org</a></p> <ul> <li>Works well with ExCL: data can be stored in a shared cache in the project folder, which can also serve as remote data storage outside of ExCL</li> <li><code class="language-plaintext highlighter-rouge">dvc repro</code> reproduces a pipeline by figuring out what changed and rerunning only the affected stages</li> <li>See the <a href="https://docs.excl.ornl.gov/quick-start-guides/dvc">DVC Quick-Start Guide</a> in the ExCL User Docs</li> </ul> <p><strong>LakeFS</strong> — <a href="https://lakefs.io/">lakefs.io</a></p> <ul> <li>Git-like version control for a data lake</li> <li>Provides a unified, high-performance, secure API over versioned, reproducible, and auditable storage</li> <li>Integrates with common data/ML tooling (Spark, pandas, Jupyter, MLflow, TensorFlow, langchain, and more) and backs onto POSIX, S3, GCS, Azure Blob, and other S3-compatible storage</li> </ul> <hr /> <h2 id="️-operations-and-procurement">🛠️ Operations and Procurement</h2> <p>Most recent operations work has been internal and infrastructure-focused:</p> <ul> <li>Updates to respond to well-known kernel vulnerabilities</li> <li>More frequent update schedules going forward, likely aligning to a monthly update/reboot cadence</li> <li>Extensive work on account management to reduce staff overhead and the opportunity for errors <ul> <li>Accounts are being reviewed for PAS compliance; PAS will become a requirement on all accounts (low effort for ORNL staff members)</li> <li>PAS requests for foreign national researchers have substantial lead times (about four weeks)</li> </ul> </li> <li>Procurement proposals are in motion — now is the time to get research-enabling suggestions and requests to your systems engineering team</li> </ul> <hr /> <h2 id="️-excl-updates">🖥️ ExCL Updates</h2> <ul> <li>The former HYP100 system (4x V100, 4x MK100) is now in the ExCL laboratory envelope (in K200)</li> <li>Maxwell will be redeployed next week as an ExCL-standard configured system; emergency access is available before redeployment is complete</li> </ul> <hr /> <h2 id="-questions--projects--comments--discussions">💬 Questions / Projects / Comments / Discussions</h2> <ul> <li>Recordings are available via the link in the newsletter</li> <li>Discussion on how to provide solo access for performance analysis, and how that would affect GitLab runners</li> <li>Interest in performance monitoring via <code class="language-plaintext highlighter-rouge">btop</code>, and whether to extend user access to Grafana/check_mk</li> </ul> <hr /> <h2 id="-summary">📌 Summary</h2> <p>June’s meeting highlighted two data-versioning tools (DVC and LakeFS) for ML and data-lake workflows, continued operations and account-management hardening, a hardware redeployment (Maxwell), and an open discussion on expanding performance-monitoring access for users.</p>{"name"=>nil, "avatar"=>nil, "bio"=>nil, "location"=>nil, "email"=>nil, "googlescholar"=>nil, "orcid"=>nil, "bitbucket"=>nil, "github"=>nil, "linkedin"=>nil}

June 2026 ExCL meeting slides.

2026-05-21T00:00:00-04:002026-05-21T13:40:31-04:00https://www.excl.ornl.gov/excl/may-meeting-2026<p>May 2026 ExCL meeting slides.</p> <!--more--> <object data="/assets/presentations/2026-05-may-excl-meeting.pdf" type="application/pdf" width="560" height="700"><p>It appears that you don't have a PDF plugin for this browser. You can <a href="/assets/presentations/2026-05-may-excl-meeting.pdf">click here to download the PDF file.</a></p></object> <hr /> <p>Below is a LLM-generated summary from the meeting slides.</p> <hr /> <h1 id="-excl-monthly-update--may-2026">🚀 ExCL Monthly Update — May 2026</h1> <p>This month’s meeting featured tool and workflow spotlights on file management and Git workflows, a recap of a recent planned power outage, and a discussion of the hardware wishlist for AI and agentic development.</p> <hr /> <h2 id="-file-management-favorites-aaron-young">📁 File Management Favorites (Aaron Young)</h2> <p>Aaron shared a roundup of tried-and-true and newly discovered file management tools across platforms:</p> <p><strong>Cross-platform</strong></p> <ul> <li><strong>Beyond Compare</strong> — folder/file diffing, Git difftool/mergetool integration, SFTP folder access, image and Excel diff. A long-time favorite.</li> </ul> <p><strong>Windows-specific</strong></p> <ul> <li><strong>OneCommander</strong> — a more performant Explorer alternative with dual-pane/tabbed browsing, flexible shortcuts, file age heatmap, and a plaintext config file.</li> <li><strong>Voidtools Everything</strong> — instant indexed filename search with expressive regex syntax.</li> </ul> <p><strong>Terminal / Cross-platform</strong></p> <ul> <li><strong>Yazi</strong> — a blazing-fast Rust-based terminal file manager with Vim-style navigation, multi-tab support, bulk rename, and integrations with <code class="language-plaintext highlighter-rouge">jq</code>, <code class="language-plaintext highlighter-rouge">fd</code>, <code class="language-plaintext highlighter-rouge">fzf</code>, and <code class="language-plaintext highlighter-rouge">zoxide</code>.</li> </ul> <p><strong>Linux / macOS</strong></p> <ul> <li><strong>ncdu</strong> — NCurses disk usage TUI</li> <li><strong>plocate</strong> (<code class="language-plaintext highlighter-rouge">locate</code>) — indexed file search</li> </ul> <hr /> <h2 id="-new-git-tools-aaron-young">🔀 New Git Tools (Aaron Young)</h2> <p>Two tools for modernizing Git workflows:</p> <p><strong>Lazygit</strong></p> <ul> <li>Visualizes a Git repository in a TUI</li> <li>Maps Git commands to vim-style keypresses for fast, keyboard-driven operation</li> </ul> <p><strong>GitButler</strong></p> <ul> <li>Very easy rebasing and commit editing — great for crafting clean PRs/MRs</li> <li>Supports stacked and parallel branches with git forge integration</li> <li>Easy to set up and tear down on existing repos</li> <li><em>Con:</em> Still in early development</li> </ul> <hr /> <h2 id="-favorite-workflows-aaron-young">🧠 Favorite Workflows (Aaron Young)</h2> <p><strong>Note-taking / Personal Knowledge Base / Second Brain</strong></p> <ul> <li>Combines <strong>OneNote</strong> (Teams/Outlook capture) + <strong>Obsidian</strong> (local linked notes) + the <strong>PARA method</strong> (Projects, Areas, Resources, Archive) + the <em>Linking Your Thinking</em> framework</li> </ul> <p><strong>Remote Linux Development</strong></p> <ul> <li><strong>MobaXterm</strong> + <strong>tmux</strong> + <strong>VS Code</strong> + <strong>ThinLinc</strong> — a stack for effective remote work on ExCL systems</li> </ul> <hr /> <h2 id="-planned-power-outage--may-1517">⚡ Planned Power Outage — May 15–17</h2> <p>A planned outage affected 5600 K200 (research) and L200 (moderate infrastructure) racks over the weekend of May 15–17 to deploy a new UPS5 and complete triennial electrical maintenance.</p> <p>Key outcomes:</p> <ul> <li>Both ExCL racks in K200 were affected, including all file servers, hypervisors, Triple Crown, and several compute servers</li> <li>The outage ended Sunday evening on schedule (no automated notification was received)</li> <li>Most ExCL systems were restored by 11 AM Monday</li> <li>A small number of follow-on issues appeared; post-maintenance procedures are being updated</li> <li>Staff used the window to patch systems and remediate <strong>Copy Fail</strong> and <strong>Dirty Frag</strong> vulnerabilities</li> </ul> <hr /> <h2 id="️-hardware-wishlist">🖥️ Hardware Wishlist</h2> <p>ExCL is developing a new hardware wishlist and actively seeking user feedback and requests. The following systems are under consideration:</p> <table> <thead> <tr> <th>System</th> <th>Details</th> </tr> </thead> <tbody> <tr> <td><strong>TensTorrent BlackHole P150</strong></td> <td>PCIe accelerator card</td> </tr> <tr> <td><strong>NVIDIA DGX Spark</strong></td> <td>NVIDIA GB10 Grace Blackwell superchip</td> </tr> <tr> <td><strong>HP ZGX Nano</strong></td> <td>NVIDIA GB10 Grace Blackwell superchip</td> </tr> <tr> <td><strong>Framework Desktop Mini AI Cluster</strong></td> <td>AMD Ryzen AI Max+ 395, 128 GB</td> </tr> <tr> <td><strong>TinyPilot Voyager 3 KVM</strong></td> <td>Remote KVM management</td> </tr> </tbody> </table> <p>If you have hardware requests or feedback on this wishlist, please reach out to the ExCL team.</p> <hr /> <h2 id="-summary">📌 Summary</h2> <p>May’s meeting highlighted practical tooling improvements for day-to-day development workflows, a successfully completed infrastructure maintenance window, and a hardware wishlist in progress — feedback from users is welcome.</p>{"name"=>nil, "avatar"=>nil, "bio"=>nil, "location"=>nil, "email"=>nil, "googlescholar"=>nil, "orcid"=>nil, "bitbucket"=>nil, "github"=>nil, "linkedin"=>nil}

May 2026 ExCL meeting slides.

2026-03-19T00:00:00-04:002026-03-19T16:35:29-04:00https://www.excl.ornl.gov/excl/march-meeting-2026<p>March 2026 ExCL meeting slides.</p> <!--more--> <object data="/assets/presentations/2026-03-march-excl-meeting.pdf" type="application/pdf" width="560" height="700"><p>It appears that you don't have a PDF plugin for this browser. You can <a href="/assets/presentations/2026-03-march-excl-meeting.pdf">click here to download the PDF file.</a></p></object> <hr /> <p>Below is a LLM-generated summary from the meeting slides.</p> <hr /> <h1 id="-excl-monthly-update--march-2026">🚀 ExCL Monthly Update — March 2026</h1> <p>This month’s updates focus on <strong>storage usability, CI improvements, shared infrastructure, and data management best practices</strong>.</p> <hr /> <h2 id="-improved-project-storage-navigation">📂 Improved Project Storage Navigation</h2> <p>We’ve made a small but impactful usability improvement:</p> <ul> <li>You can now <strong>read project folder names under <code class="language-plaintext highlighter-rouge">/auto/projects</code></strong></li> <li>Permissions remain unchanged, but <strong>directory discovery is much easier</strong></li> <li>Tools like <code class="language-plaintext highlighter-rouge">vim</code>, <code class="language-plaintext highlighter-rouge">yazi</code>, and tab completion now work more effectively</li> </ul> <p>💡 <em>Note:</em> The first access still triggers automount—after that, navigation is seamless.</p> <hr /> <h2 id="-new-shared-storage-for-apptainer-images">📦 New Shared Storage for Apptainer Images</h2> <p>To better support large container workflows:</p> <ul> <li> <p>New shared location:</p> <div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code>/auto/projects/apptainer </code></pre></div> </div> </li> <li>Designed for <strong>large, reusable Apptainer images</strong></li> <li>Current use cases include: <ul> <li>Unreal Engine containers</li> <li>AMD Vitis environments</li> </ul> </li> <li>Storage policy: <ul> <li>Default cap: <strong>1 TB</strong></li> <li>Adjustable as needed</li> </ul> </li> </ul> <hr /> <h2 id="️-ci-on-excl-slurm--gitlab-recommended">⚙️ CI on ExCL: Slurm + GitLab (Recommended)</h2> <p>Running CI through Slurm is now the <strong>standard and recommended approach for jobs running across multiple ExCL systems</strong>.</p> <h3 id="key-improvements">Key Improvements</h3> <ul> <li>✅ <strong>CI template now checks <code class="language-plaintext highlighter-rouge">sbatch</code> return codes</strong></li> <li>⚠️ Standard error is treated as a <strong>warning</strong>, not immediate failure</li> <li>🧪 New <strong>system-device-smoke-test</strong>: <ul> <li>Validates CUDA, HIP, OpenMP, OpenCL</li> <li>Runs across major ExCL systems</li> <li>Uses shared IRIS setup scripts:</li> </ul> </li> </ul> <div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="nb">source</span> /auto/software/iris/setup-system.source </code></pre></div></div> <p>📌 This significantly improves <strong>reliability and portability of CI pipelines</strong> across heterogeneous nodes.</p> <hr /> <h2 id="-dvc-for-data-management-on-excl">📊 DVC for Data Management on ExCL</h2> <p>We now recommend <strong>DVC (Data Version Control)</strong> for managing datasets and pipelines.</p> <h3 id="why-dvc-works-well-on-excl">Why DVC Works Well on ExCL</h3> <ul> <li>Shared cache model: <ul> <li>Store data in a <strong>project-level shared cache</strong></li> <li>Avoid duplication across repos</li> </ul> </li> <li>Hybrid workflow: <ul> <li>Same cache can act as: <ul> <li>Local shared storage (on ExCL)</li> <li>Remote storage (outside ExCL)</li> </ul> </li> </ul> </li> </ul> <p>👉 See the <strong><a href="https://docs.excl.ornl.gov/quick-start-guides/dvc">DVC Quick Start Guide in ExCL Docs</a></strong> for setup instructions.</p> <hr /> <h2 id="-documentation-best-practices-highlight">📚 Documentation Best Practices Highlight</h2> <p>We highlighted an excellent resource:</p> <p><strong><a href="https://cass.community/events/hpcbp-097-onegoodtutorial">“Minimum Viable Documentation Product (MVDP)”</a></strong> from the Consortium for the Advancement of Scientific Software</p> <h3 id="key-takeaway">Key Takeaway</h3> <p>Follow a <strong>documentation checklist</strong>:</p> <ul class="task-list"> <li class="task-list-item"><input type="checkbox" class="task-list-item-checkbox" disabled="disabled" />🗺️ <strong><a href="https://onegoodtutorial.org/in-depth/synopsis/">Synopsis</a>:</strong> 1–3 sentence summary of your project</li> <li class="task-list-item"><input type="checkbox" class="task-list-item-checkbox" disabled="disabled" />📑 <strong><a href="https://onegoodtutorial.org/in-depth/tutorial-planning/">Tutorial</a>:</strong> ✨ Show people what your software can do! ✨</li> <li class="task-list-item"><input type="checkbox" class="task-list-item-checkbox" disabled="disabled" />👩‍🔧 <strong><a href="https://onegoodtutorial.org/in-depth/contact-information/">Contact Information</a>:</strong> How to ask a human about your software</li> <li class="task-list-item"><input type="checkbox" class="task-list-item-checkbox" disabled="disabled" />🚀 <strong><a href="https://onegoodtutorial.org/in-depth/installation-instructions/">Install Instructions</a>:</strong> How to install your software</li> <li class="task-list-item"><input type="checkbox" class="task-list-item-checkbox" disabled="disabled" />📜 <strong><a href="https://onegoodtutorial.org/in-depth/software-citation/">Citation Instructions</a>:</strong> How to cite your software</li> <li class="task-list-item"><input type="checkbox" class="task-list-item-checkbox" disabled="disabled" />🙌 <strong><a href="https://onegoodtutorial.org/in-depth/contribution-statements/">Contribution Statement</a>:</strong> How users can contribute to your project</li> <li class="task-list-item"><input type="checkbox" class="task-list-item-checkbox" disabled="disabled" />📚 <strong><a href="https://onegoodtutorial.org/in-depth/reference-material/">Reference Material</a>:</strong> Precise specifications of APIs, etc.</li> <li class="task-list-item"><input type="checkbox" class="task-list-item-checkbox" disabled="disabled" />⚖️ <strong><a href="https://onegoodtutorial.org/in-depth/licensing-statements/">Licensing Statement</a>:</strong> The legal status of your code</li> <li class="task-list-item"><input type="checkbox" class="task-list-item-checkbox" disabled="disabled" />🙏 <strong><a href="https://onegoodtutorial.org/in-depth/acknowledgments/">Acknowledgments</a>:</strong> Credit your funders</li> </ul> <p>🔗 Resource: <a href="https://onegoodtutorial.org">https://onegoodtutorial.org</a></p> <hr /> <h2 id="️-system--infrastructure-updates">🖥️ System &amp; Infrastructure Updates</h2> <ul> <li>⏳ Migration of <strong>MI100 / V100 system</strong> delayed until after SC26 submissions</li> <li>🔄 Work ongoing to restore <strong>Groq inference hardware</strong></li> <li>🆕 <strong>Tenstorrent Blackhole</strong> acquisition in progress</li> <li>🔧 Scheduled downtime: <ul> <li><strong>Tuesday, March 24, 9–11 AM</strong> <ul> <li>OS updates + system reboots</li> <li>No NVIDIA driver updates (unless requested)</li> </ul> </li> </ul> </li> </ul> <hr /> <h2 id="-open-topics--discussion-areas">💬 Open Topics &amp; Discussion Areas</h2> <p>We’re actively gathering feedback on:</p> <ul> <li>Legacy system usage (e.g., <em>Leconte PowerPC + V100</em>)</li> <li>Hudson access policies. Request to limit system to Slurm-only access.</li> <li>Cloud GPU access options</li> </ul> <hr /> <h2 id="-summary">📌 Summary</h2> <p>This month’s updates continue to push ExCL toward:</p> <ul> <li><strong>More usable shared infrastructure</strong></li> <li><strong>Scalable data workflows (DVC)</strong></li> <li><strong>Robust CI across heterogeneous systems</strong></li> <li><strong>Better documentation practices</strong></li> </ul> <p>If you have feedback or want help adopting any of these tools, reach out!</p>{"name"=>nil, "avatar"=>nil, "bio"=>nil, "location"=>nil, "email"=>nil, "googlescholar"=>nil, "orcid"=>nil, "bitbucket"=>nil, "github"=>nil, "linkedin"=>nil}

March 2026 ExCL meeting slides.

2026-02-19T00:00:00-05:002026-03-03T11:52:28-05:00https://www.excl.ornl.gov/excl/february-meeting-2026<p>February 2026 ExCL meeting slides.</p> <!--more--> <object data="/assets/presentations/2026-02-february-excl-meeting.pdf" type="application/pdf" width="560" height="700"><p>It appears that you don't have a PDF plugin for this browser. You can <a href="/assets/presentations/2026-02-february-excl-meeting.pdf">click here to download the PDF file.</a></p></object> <hr /> <p>Below is a LLM-generated summary from the meeting slides.</p> <hr /> <h1 id="excl-monthly-update--february-2026">ExCL Monthly Update – February 2026</h1> <p><strong>February 19, 2026 – Oak Ridge, Tennessee</strong></p> <p><em>Experimental Computing Laboratory (ExCL), Advanced Computing Systems Research Section</em></p> <p>The February ExCL Monthly Meeting highlighted significant infrastructure upgrades, documentation improvements, and ongoing efforts to expand our heterogeneous computing capabilities in support of growing AI and accelerator-driven workloads.</p> <p>Below is a summary of key updates.</p> <hr /> <h2 id="-new-system-deployment-maxwell-formerly-hyp-100">🚀 New System Deployment: Maxwell (Formerly hyp-100)</h2> <p>Maxwell is being redeployed from ORC to ExCL and will soon be integrated into the ExCL environment (including home directories and shared resources).</p> <h3 id="system-specifications">System Specifications</h3> <ul> <li><strong>2× AMD EPYC 7763 (64 cores each)</strong> <ul> <li>Expected to run hyperthreaded (256 cores total)</li> </ul> </li> <li><strong>1 TB Main Memory</strong></li> <li><strong>4× NVIDIA A100 (40GB, no NVLink)</strong></li> <li><strong>4× AMD MI100 (32GB)</strong></li> </ul> <p>This system is particularly well suited for:</p> <ul> <li>General compute workloads</li> <li>Large language model (LLM) experimentation</li> <li>Heterogeneous accelerator research</li> </ul> <p>Configuration is non-trivial, so the current OS release will likely remain in place during integration.</p> <hr /> <h2 id="-thinlinc-updates-and-licensing">🖥 ThinLinc Updates and Licensing</h2> <p>ThinLinc has undergone both licensing and version updates.</p> <h3 id="key-changes">Key Changes</h3> <ul> <li>Standard license reverted from <strong>10 → 3 concurrent users</strong></li> <li>Upgraded to <strong>ThinLinc 4.20.0</strong></li> <li>Secured <strong>90-day licenses for 10 concurrent users</strong> while permanent licensing is pursued</li> </ul> <h3 id="important-reminder">Important Reminder</h3> <p>ThinLinc sessions are persistent (similar to <code class="language-plaintext highlighter-rouge">tmux</code>). Please <strong>log off when not actively using your session</strong> to avoid consuming licenses unnecessarily.</p> <p>Idle sessions may be terminated if needed.</p> <hr /> <h2 id="️-thinlinc-xfce-configuration-reset">⚠️ ThinLinc XFCE Configuration Reset</h2> <p>If your XFCE desktop environment becomes unstable or misconfigured, reset it using:</p> <div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="nb">rm</span> <span class="nt">-rf</span> ~/.config/xfce4 <span class="nb">rm</span> <span class="nt">-rf</span> ~/.cache/sessions <span class="nb">rm</span> <span class="nt">-rf</span> ~/.local/share/xfce4 <span class="nb">rm</span> <span class="nt">-rf</span> ~/.config/xfce4/panel </code></pre></div></div> <p>This restores a clean session state.</p> <hr /> <h2 id="-apptainer-deployment-expands">📦 Apptainer Deployment Expands</h2> <p>Apptainer has been successfully deployed and tested on:</p> <ul> <li><strong>Affirmed</strong></li> <li><strong>Justify</strong></li> <li><strong>Pharaoh</strong></li> <li><strong>Secretariat</strong></li> <li><strong>Faraday</strong></li> </ul> <p>As additional systems are redeployed, Apptainer will be installed where appropriate to support containerized workflows.</p> <p>Documentation is available in the ExCL User Docs.</p> <hr /> <h2 id="-documentation-updates">📚 Documentation Updates</h2> <p>Several improvements have been made to ExCL documentation:</p> <ul> <li><a href="https://docs.excl.ornl.gov/quick-start-guides/python#python-virtual-environments-using-uv"><strong>Python Virtual Environments Using <code class="language-plaintext highlighter-rouge">uv</code></strong></a></li> <li><a href="https://docs.excl.ornl.gov/quick-start-guides/marimo"><strong>Marimo Quick Start Guide</strong></a></li> <li><a href="https://docs.excl.ornl.gov/quick-start-guides/jupyter-quick-start#using-foxyproxy-recommended"><strong>Jupyter Notebook Access via FoxyProxy (Recommended)</strong></a></li> </ul> <p>These updates streamline environment management and interactive development workflows.</p> <hr /> <h2 id="-live-demonstration">🎥 Live Demonstration</h2> <p>The meeting included a demonstration covering:</p> <ul> <li>ThinLinc usage</li> <li>FoxyProxy configuration</li> <li>Python environment setup with <code class="language-plaintext highlighter-rouge">uv</code></li> <li>SSH port forwarding</li> </ul> <p>These tools continue to form the recommended workflow for interactive ExCL usage.</p> <hr /> <h2 id="-excl-microelectronics--neuromorphic-lab-3700a102">🔬 ExCL Microelectronics &amp; Neuromorphic Lab (3700/A102)</h2> <p>Progress continues on the ExCL microelectronics and neuromorphic laboratory space:</p> <ul> <li>ExCL VLAN networking completed</li> <li>Device relocation from 5100/227 being scheduled</li> <li>Infrastructure adjustments ongoing</li> </ul> <p>(We are also tracking down the mysteriously relocated stools.)</p> <hr /> <h2 id="-new-gpu-deployments-ai-inference-focus">🎮 New GPU Deployments (AI Inference Focus)</h2> <p>To support increasing AI inference workloads, we are preparing to deploy:</p> <ul> <li><strong>NVIDIA GeForce RTX 5090</strong></li> <li><strong>AMD Radeon RX 9070 XT</strong></li> </ul> <p>These are consumer-grade but high-power GPUs intended to supplement current infrastructure.</p> <p>We are also:</p> <ul> <li>Re-engaging with <strong>Grok</strong> regarding accelerator support</li> <li>Holding discussions with <strong>Tenstorrent</strong></li> </ul> <p>Long-term, these accelerators may help shift workload away from traditional GPUs.</p> <hr /> <h2 id="-slurm-reminder-gpu-jobs">🧠 Slurm Reminder (GPU Jobs)</h2> <p>If your job requires GPUs, ensure you request them explicitly via Slurm:</p> <div class="language-bash highlighter-rouge"><div class="highlight"><pre class="highlight"><code><span class="nt">--gpus</span><span class="o">=</span>2 </code></pre></div></div> <p>Failure to request GPU resources may result in improper scheduling or runtime errors.</p> <hr /> <h2 id="looking-ahead">Looking Ahead</h2> <p>ExCL continues to expand its heterogeneous computing capabilities across CPUs, GPUs, and emerging accelerator platforms. With growing AI and inference demands, our infrastructure roadmap remains focused on flexibility, containerization, and performance experimentation.</p> <p>If you have questions, project ideas, or would like systems deployed with specific configurations (e.g., Apptainer), please reach out.</p>{"name"=>nil, "avatar"=>nil, "bio"=>nil, "location"=>nil, "email"=>nil, "googlescholar"=>nil, "orcid"=>nil, "bitbucket"=>nil, "github"=>nil, "linkedin"=>nil}

February 2026 ExCL meeting slides.

2026-01-15T00:00:00-05:002026-01-15T14:49:37-05:00https://www.excl.ornl.gov/excl/january-meeting-2026<p>January 2026 ExCL meeting slides.</p> <!--more--> <object data="/assets/presentations/2026-01-january-excl-meeting.pdf" type="application/pdf" width="560" height="700"><p>It appears that you don't have a PDF plugin for this browser. You can <a href="/assets/presentations/2026-01-january-excl-meeting.pdf">click here to download the PDF file.</a></p></object> <hr /> <p>Below is a LLM-generated summary from the meeting slides.</p> <hr /> <h2 id="highlights-from-the-january-2026-excl-monthly-meeting">Highlights from the January 2026 ExCL Monthly Meeting</h2> <p><strong>1. Modernized Infrastructure &amp; Growth</strong></p> <ul> <li>ExCL now supports <strong>98 users across 34 groups</strong>, including <strong>66 external users from 29 organizations</strong>.</li> <li>Significant contributions: <strong>11 papers, 2 journals, 27 presentations</strong> citing ExCL in the past year.</li> <li>Focus on <strong>reliability, storage stability</strong>, and <strong>user self-service</strong> to accelerate research.</li> </ul> <p><strong>2. TRC Lab A102: Expanding Capabilities</strong></p> <ul> <li>Dedicated space for <strong>embedded systems, microelectronics, AI, and neuromorphic computing</strong>.</li> <li>Equipped with <strong>Zenith systems, AI-testbeds, neuromorphic testbeds</strong>, and photonic-ready infrastructure.</li> <li>Future vision: <strong>Quantum, photonics, and automated AI-driven experimentation</strong>.</li> </ul> <p><strong>3. Preemptable Slurm Queues</strong></p> <ul> <li>Added <strong>nvidia-long</strong> as a <strong>preemptable GPU queue</strong> to improve utilization.</li> <li>Jobs can be requeued with <code class="language-plaintext highlighter-rouge">--requeue</code> for long-running tasks.</li> <li>Documentation: https://docs.excl.ornl.gov/quick-start-guides/slurm.</li> </ul> <p><strong>4. PYX Discussion: Accelerating AI/HPC Research</strong></p> <ul> <li>Introducing <strong>pyx</strong>, a private GPU-aware package registry.</li> <li>Benefits: <ul> <li><strong>Auto-detect hardware</strong> and serve pre-built wheels.</li> <li><strong>Shared build cache</strong> for faster installs.</li> <li><strong>Time-to-Science</strong> reduced from 15+ minutes to &lt;10 seconds.</li> </ul> </li> <li>Learn more: https://docs.pyx.dev/ and https://talkpython.fm/episodes/show/520/pyx-the-other-side-of-the-uv-coin-announcing-pyx.</li> </ul> <p><strong>5. Other Ongoing Efforts</strong></p> <ul> <li>Networking for <strong>ExCL microelectronics and neuromorphic lab</strong> to complete by Jan 20.</li> <li>Upcoming GPU deployments: <strong>Nvidia GeForce RTX 5090</strong> and <strong>AMD Radeon RX 9070 XT</strong>.</li> </ul> <p><strong>Stay Connected</strong></p> <ul> <li>Check out the inaugural <strong>ExCL Newsletter</strong> for research highlights and best practices.</li> <li>For questions or collaborations, reach out to the ExCL team.</li> </ul>{"name"=>nil, "avatar"=>nil, "bio"=>nil, "location"=>nil, "email"=>nil, "googlescholar"=>nil, "orcid"=>nil, "bitbucket"=>nil, "github"=>nil, "linkedin"=>nil}

January 2026 ExCL meeting slides.

2026-01-08T00:00:00-05:002026-01-08T17:41:16-05:00https://www.excl.ornl.gov/excl/sc25<!--more--> <h2 id="-best-paper-recognition-at-waccpd">🏆 Best Paper Recognition at WACCPD</h2> <p>SULI Intern Tatiana Melnichenko (University of Tennessee, Knoxville) was part of the team awarded Best Paper at the Twelfth Workshop on Accelerator Programming and Directives (WACCPD 2025) workshop. She presented the work to a packed room, and highlighted ExCL’s role in enabling the evaluation of Mojo’s vendor-neutral programming capabilities across NVIDIA H100 and AMD MI300A GPUs for science workloads. Tatiana will be returning for a part-time internship in the spring, continuing her contributions with ORNL researchers.</p> <p><a href="https://doi.org/10.1145/3731599.3767573">Mojo: MLIR-based Performance-Portable HPC Science Kernels on GPUs for the Python Ecosystem</a>, William Godoy, Tatiana Melnichenko, Pedro Valero-Lara, Wael Elwasif, Philip Fackler, Rafael Ferreira Da Silva, Keita Teranishi, Jeffrey Vetter, <em>SC ‘25: Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis</em>, 2025.</p> <h2 id="-acm-undergraduate-research-competition-finalists">🎓 ACM Undergraduate Research Competition Finalists</h2> <p>Tatiana’s <a href="https://sc25.conference-program.com/presentation/?id=post151&amp;sess=sess529">poster</a> was selected as a Top 5 finalist in the ACM Undergraduate Research Competition, underscoring the strength of internship engagement in ExCL-supported projects.</p> <h2 id="-3rd-place-acm-src-poster-award">🥉 3rd Place ACM SRC Poster Award</h2> <p>Matthew Chung, mentored by Narasinga Rao, earned 3rd place in the ACM Undergraduate Research Poster Competition for his <a href="https://sc25.conference-program.com/presentation/?id=post230&amp;sess=sess529">poster</a> on a Heterogeneous Eigen solver using IRIS and conducted on ExCL systems.</p> <h2 id="-sc-best-poster-nomination">🧠 SC Best Poster Nomination</h2> <p>A <a href="https://sc25.conference-program.com/presentation/?id=post267&amp;sess=sess527">poster</a> on Julia with an Intelligent Runtime System, presented by Narasinga Rao, Pedro Valero Lara, William Godoy, Keita Teranishi, and Jeffrey Vetter, was nominated for Best Poster at SC25. The work explored dynamic optimization strategies for heterogeneous systems.</p> <h2 id="-chathpc-and-chatport-papers">📄 ChatHPC and ChatPort Papers</h2> <p>Two papers—ChatHPC and ChatPort—were presented on how AI capabilities can leverage and speed up the HPC software development process. Researchers used ExCL capabilities to fine-tune CodeLlama with libraries that are critical to DOE HPC mission (Kokkos, OpenMP, ADIOS2, IRIS, TAU).</p> <ul> <li><a href="https://doi.org/10.1145/3712285.3759787">ChatHPC: Building the Foundations for a Productive and Trustworthy AI-Assisted HPC Ecosystem</a>, Pedro Valero Lara, Aaron Young, Jeffrey S. Vetter, Zheming Jin, Swaroop Pophale, Mohammad Alaul Haque Monil, Keita Teranishi, William F. Godoy, <em>SC ‘25: Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis</em>, 2025.</li> <li><a href="https://doi.org/10.1145/3731599.3767398">Enhancing ChatPORT with CUDA-to-SYCL Kernel Translation Capability</a>, Zheming Jin, Swaroop Pophale, Keita Teranishi, <em>SC-W’25: Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis Workshops</em>, 2025.</li> </ul> <h2 id="-short-talks-at-workshops">🎤 Short talks at workshops</h2> <p>Philip W. Fackler gave a short talk titled <em>Speeding Up Phonon Dynamic Structure Factor Calculations in Phonopy with GPU-Accelerated Computing</em> as part of the <a href="https://wordpress.cels.anl.gov/xloop-2025/workshop-schedule/">XLOOP 2025 workshop</a>.</p> <p>William Godoy gave a talk titled <em>A first look at Mojo’s MLIR-based Performance Portable GPU Programming for Python Users</em> at the <a href="https://hppss.github.io/SC25/schedule.html#a-first-look-at-mojos-mlir-based-performance-portable-gpu-programming-for-python-users">Python HPPSS workshop</a>.</p> <h2 id="-bof-and-tutorials">📒 BoF and Tutorials</h2> <p>Philip W. Fackler and William Godoy presented at the Julia for HPC <a href="https://sc25.conference-program.com/presentation/?id=bof116&amp;sess=sess480">BoF</a> and half-day <a href="https://sc25.conference-program.com/presentation/?id=tut140&amp;sess=sess268">Tutorial</a>. ExCL resources were used for the readiness of the presented software at these events, including CI services for the <a href="https://github.com/JuliaORNL/JACC.jl">JACC.jl</a> performance portable Julia library developed at ORNL, also presented as a poster <a href="https://sc25.supercomputing.org/proceedings/posters/poster_pages/post185.html"><em>JACC: Easy CPU/GPU Performance Portability for Scientific Applications in Julia</em></a>.</p> <figure class="third "> <a href="/assets/images/Tatiana-WACCPD.jpg" title="Tatiana presented to a full house."> <img src="/assets/images/Tatiana-WACCPD-th.jpg" alt="Tatiana presented to a full house." /> </a> <a href="/assets/images/Tatiana-WACCPD3.jpg" title="Tatiana’s presentation featuring ExCL on the slide."> <img src="/assets/images/Tatiana-WACCPD3-th.jpg" alt="Tatiana’s presentation featuring ExCL on the slide." /> </a> <a href="/assets/images/Tatiana-best-paper.jpg" title="MLIR Best paper award."> <img src="/assets/images/Tatiana-best-paper-th.jpg" alt="MLIR Best paper award." /> </a> <a href="/assets/images/chathpc-for-kokkos.jpg" title="Pedro presenting ChatHPC."> <img src="/assets/images/chathpc-for-kokkos.jpg" alt="Pedro presenting ChatHPC." /> </a> <figcaption>Pictures from SC25 </figcaption> </figure>{"name"=>nil, "avatar"=>nil, "bio"=>nil, "location"=>nil, "email"=>nil, "googlescholar"=>nil, "orcid"=>nil, "bitbucket"=>nil, "github"=>nil, "linkedin"=>nil}

2025-10-16T00:00:00-04:002025-10-31T11:44:05-04:00https://www.excl.ornl.gov/excl/october-meeting-2025<p>October 2025 ExCL meeting slides.</p> <!--more--> <object data="/assets/presentations/2025-10-october-excl-meeting.pdf" type="application/pdf" width="560" height="700"><p>It appears that you don't have a PDF plugin for this browser. You can <a href="/assets/presentations/2025-10-october-excl-meeting.pdf">click here to download the PDF file.</a></p></object> <hr /> <p>Below is a LLM-generated summary from the meeting slides.</p> <hr /> <h2 id="highlights-from-the-october-2025-excl-monthly-meeting">Highlights from the October 2025 ExCL Monthly Meeting</h2> <h3 id="introduction">Introduction</h3> <p>On October 16, 2025, the Experimental Computing Laboratory (ExCL) held its monthly meeting in Oak Ridge, Tennessee. The meeting was presented by Steve Moulton, Aaron Young, and Jeffrey Vetter from the Advanced Computing Systems Research Section. Here are the key topics and updates discussed during the meeting.</p> <h3 id="new-hardware-list-and-state">New Hardware List and State</h3> <h4 id="hardware-for-fpga-and-software-development">Hardware for FPGA and Software Development</h4> <ul> <li><strong>AMD Versal VHK158 Evaluation Kit</strong>: FPGA development, currently in a box for user deployment.</li> <li><strong>AMD Xilinx VCU118 Evaluation Kit</strong>: FPGA development, similar status as the Versal kit.</li> <li><strong>Apple Studio M3 MAX</strong>: Software development and GitLab runners, not deployed yet.</li> <li><strong>GeForce RTX 5090 (Nvidia consumer grade)</strong>: Development and analysis of software using GPUs, awaiting deployment.</li> <li><strong>AMD Radeon RX 9070 XT</strong>: Development and analysis of software using GPUs, requires 775 watts and will be deployed in one of the Milans.</li> </ul> <h4 id="networking-and-remote-management">Networking and Remote Management</h4> <ul> <li><strong>Cisco C9216 Network Switches</strong>: Core networking, deployment requested for 3700 and K200.</li> <li><strong>Hubbell Rack Cabinet (lockable)</strong>: Secure enclosure for Cisco switch, deployment requested for 3700.</li> </ul> <h3 id="2026-hardware-requests-and-availability">2026 Hardware Requests and Availability</h3> <ul> <li><strong>AI/LLM Desktop (Framework Desktop)</strong>: LLM development, shipping in Q4.</li> <li><strong>NVIDIA DGX Spark (Grace Blackwell)</strong>: LLM development and performance analysis, purchase invitation received.</li> <li><strong>Tenstorrent Blackhole P100a</strong>: AI accelerator, awaiting availability.</li> </ul> <h3 id="trc-lab-a102">TRC Lab A102</h3> <p>The TRC Lab A102 will be used for embedded system and microelectronics development. It will be on the ExCL private network with access to ORNL workstations network if needed. RHACS compliance will be required for admission.</p> <h3 id="general-updates">General Updates</h3> <ul> <li><strong>Slurm Accounting</strong>: Setup and configuration updates.</li> <li><strong>Account and Project Request Flows</strong>: Improved with automation.</li> <li><strong>System Specific Notes</strong>: Now included in the message of the day (MOTD) shown on every system.</li> </ul> <h3 id="gentle-reminders">Gentle Reminders</h3> <ul> <li><strong>Login.excl.ornl.gov</strong>: Not for computation or compiling, only for ssh and thinlinc access.</li> <li><strong>VSCode Usage</strong>: Avoid running VSCode on the login node via Remote-SSH or ThinLinc.</li> </ul> <h3 id="questions-and-discussion">Questions and Discussion</h3> <p>The meeting concluded with a Q&amp;A session where various topics such as power measurement at the host, CPU, and APU levels were discussed.</p>{"name"=>nil, "avatar"=>nil, "bio"=>nil, "location"=>nil, "email"=>nil, "googlescholar"=>nil, "orcid"=>nil, "bitbucket"=>nil, "github"=>nil, "linkedin"=>nil}

October 2025 ExCL meeting slides.

2025-07-17T00:00:00-04:002025-10-31T11:44:05-04:00https://www.excl.ornl.gov/excl/july-meeting-2025<p>July 2025 ExCL meeting slides.</p> <!--more--> <object data="/assets/presentations/2025-07-july-excl-meeting.pdf" type="application/pdf" width="560" height="700"><p>It appears that you don't have a PDF plugin for this browser. You can <a href="/assets/presentations/2025-07-july-excl-meeting.pdf">click here to download the PDF file.</a></p></object> <hr /> <p>Below is a LLM-generated summary of the meeting.</p> <hr /> <h1 id="excl-monthly-meeting-highlights--july-2025">ExCL Monthly Meeting Highlights – July 2025</h1> <p>Welcome to the July highlights from the <strong>Experimental Computing Laboratory (ExCL)</strong>! This month’s meeting focused on planning for future upgrades, sharing updates on resources, and a recap of the cause of a recent outage.</p> <hr /> <h2 id="system-updates">System Updates</h2> <p>The <strong>Equinox system</strong>, one of our older GPU platforms, is being phased out due to hardware issues. Projects needing multiple GPUs should reach out so we can ensure resources are available.</p> <hr /> <h2 id="looking-ahead-new-capabilities">Looking Ahead: New Capabilities</h2> <p>We’re exploring exciting new hardware options to expand ExCL’s research capabilities, including:</p> <ul> <li>Next-generation AI and GPU systems</li> <li>FPGA and accelerator platforms</li> <li>Infrastructure upgrades to support growth</li> </ul> <hr /> <h2 id="community-input">Community Input</h2> <p>We continue to welcome feedback from our users — especially around system needs, project requirements, and upcoming resource opportunities.</p>{"name"=>nil, "avatar"=>nil, "bio"=>nil, "location"=>nil, "email"=>nil, "googlescholar"=>nil, "orcid"=>nil, "bitbucket"=>nil, "github"=>nil, "linkedin"=>nil}

July 2025 ExCL meeting slides.

2025-05-15T00:00:00-04:002025-10-31T11:44:05-04:00https://www.excl.ornl.gov/excl/may-meeting-2025<p>May 2025 ExCL meeting slides.</p> <!--more--> <object data="/assets/presentations/2025-05-may-excl-meeting.pdf" type="application/pdf" width="560" height="700"><p>It appears that you don't have a PDF plugin for this browser. You can <a href="/assets/presentations/2025-05-may-excl-meeting.pdf">click here to download the PDF file.</a></p></object> <hr /> <p>Below is a LLM-generated summary of the meeting based on the presentation slides.</p> <hr /> <h1 id="excl-monthly-meeting-highlights--may-2025">ExCL Monthly Meeting Highlights — May 2025</h1> <p>The Experimental Computing Laboratory (ExCL) at Oak Ridge National Laboratory held its monthly meeting on <strong>May 15, 2025</strong>, offering a variety of updates on system infrastructure, software improvements, and the official deployment of the powerful <strong>Mi300a</strong> system. Here are the key takeaways from this month’s meeting.</p> <hr /> <h2 id="️-infrastructure--system-status">🖥️ Infrastructure &amp; System Status</h2> <ul> <li> <p><strong><code class="language-plaintext highlighter-rouge">news</code> Shell Command</strong>: A new command-line utility <code class="language-plaintext highlighter-rouge">news</code> is now available to keep users informed of system status changes. The login banner will notify users of unread items. Eventually, this will integrate with the <code class="language-plaintext highlighter-rouge">news@excl.groups.io</code> mailing list.</p> </li> <li> <p><strong>Mi300a (“Faraday”)</strong>: Officially deployed and ready for use.</p> </li> <li> <p><strong>U280 FPGA</strong>: Deployed to <code class="language-plaintext highlighter-rouge">milan3</code>, with updated SLURM helper functions and environment setup.</p> </li> <li> <p><strong>ThinLinc</strong>: Updated to version <strong>4.18</strong>.</p> </li> <li> <p><strong>Account Automation</strong>: Improved reporting and automation reduce the burden of manual account management.</p> </li> <li> <p><strong>Security</strong>: Gateway machines will now <strong>auto-update and reboot weekly</strong> to improve vulnerability management.</p> </li> <li> <p><strong>TinyPilot Update</strong>: New run script menu enables <strong>remote system resets</strong> via REISUB.</p> </li> <li> <p><strong>Siemens Remix and Licensing</strong>: Deployed with no changes to user selections.</p> </li> </ul> <hr /> <h2 id="-spotlight-mi300a-system-faraday">⚡ Spotlight: Mi300a System (“Faraday”)</h2> <h3 id="what-is-it">What Is It?</h3> <ul> <li>Supermicro <strong>AS-4145GH-TNMR</strong> 4U air-cooled server.</li> <li>Contains <strong>4 APU nodes</strong>, each with: <ul> <li>912 CDNA3 GPU units</li> <li>96 Zen 4 CPU cores</li> <li>128 GB HBM3 (for a total of 512 GB)</li> </ul> </li> <li>Ubuntu 24.04 LTS + ROCm 6.4.0</li> <li>Specialized riser/backplane needed for hardware expansion</li> </ul> <h3 id="why-its-interesting">Why It’s Interesting</h3> <ul> <li><strong>High-capability AMD GPUs</strong> – ideal for AI and HPC workloads</li> <li><strong>Infinity Cache &amp; Interconnect</strong></li> <li><strong>Unified Memory Access</strong> – significant for memory hierarchy research</li> <li>Fast performance – e.g., <strong>LLVM built in 5 minutes</strong></li> </ul> <h3 id="how-to-access">How to Access</h3> <ul> <li>Log in via <code class="language-plaintext highlighter-rouge">login.excl.ornl.gov</code></li> <li>Hostname: <code class="language-plaintext highlighter-rouge">faraday</code></li> <li>Load ROCm module: <code class="language-plaintext highlighter-rouge">module load rocmmod</code></li> <li>Test setup with <a href="https://github.com/jungwonkim/amd-toy"><code class="language-plaintext highlighter-rouge">amd-toy</code></a></li> <li>Contact <code class="language-plaintext highlighter-rouge">help@excl.ornl.gov</code> if render group access is missing</li> </ul> <h3 id="documentation-links">Documentation Links</h3> <ul> <li><a href="https://docs.excl.ornl.gov/system-overview/faraday">ExCL Faraday User Docs</a></li> <li><a href="https://www.supermicro.com/en/accelerators/amd">Supermicro AMD Accelerator Systems</a></li> <li><a href="https://www.supermicro.com/datasheet/datasheet_H13_QuadAPU.pdf">Faraday Datasheet (PDF)</a></li> <li><a href="https://www.supermicro.com/manuals/superserver/4U/MNL-2754.pdf">Supermicro Hardware Manual (PDF)</a></li> </ul> <hr /> <h2 id="-power-monitoring">🔋 Power Monitoring</h2> <ul> <li><strong>CheckMK Integration</strong> for Mi300a power tracking is in progress.</li> <li>Documentation for power configuration will follow soon.</li> </ul> <hr /> <h2 id="-final-notes">📣 Final Notes</h2> <ul> <li>Questions, suggestions, and community discussions are welcome.</li> <li>Stay informed through the <code class="language-plaintext highlighter-rouge">news</code> shell command and the <a href="mailto:news@excl.groups.io">ExCL mailing list</a>.</li> </ul> <hr /> <p>The ExCL team continues to work on enhancing their infrastructure and documentation to better support users. The upcoming months promise exciting new deployments and important updates. Make sure to stay informed through the ExCL mailing list, monthly meetings, and their documentation updates!</p>{"name"=>nil, "avatar"=>nil, "bio"=>nil, "location"=>nil, "email"=>nil, "googlescholar"=>nil, "orcid"=>nil, "bitbucket"=>nil, "github"=>nil, "linkedin"=>nil}

May 2025 ExCL meeting slides.