Christian Meesters<p>Never again, I will attend a <a href="https://fediscience.org/tags/HPC" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#<span>HPC</span></a>-related Hackathon without a wooden platter. Something at least I can bang my head against, when people tell their HPC anecdotes.</p><p>It will never cease to amaze me to learn how admins throttle usability and thereby hinder scientific process.</p><p>- nodes with multiple GPUs, you only may reserve for whole nodes, even for single-GPU jobs<br>- this or that default setting which has been configured away.<br>- defaults, which are preconfigured and which users (or a user tool like a workflow engine) may not set. Even when requesting equal or less resources.<br>- compute nodes without internet access<br>- no standard env variables such TMP or TEMP</p><p>I could go on for many pages. Could write a paper, even. Yet, never did I take any note. </p><p>Now, I have written this before but: My dear fellow HPC-admins, if you do not strive for good service, people will use other resources. Potentially, some basement server. It is the year 2025 and saving energy is a thing. Your non-service directly translates to many MW of wasted energy.</p>