https://theloginnode.com/posts/ Recent content in Posts on The Login Node Hugo -- gohugo.io en © 2026 Will Paik Sun, 14 Jun 2026 00:00:00 +0000 https://theloginnode.com/posts/hpc-special-topics-02/ Sun, 14 Jun 2026 00:00:00 +0000 https://theloginnode.com/posts/hpc-special-topics-02/ V100 and A100 both scale past 95% efficiency across 8 GPUs, but A100 delivers 2.4 to 2.7x more throughput per GPU. This post covers measured PyTorch DDP scaling results on 8xV100 SXM2 and 8xA100 SXM4, using ResNet-152 and ViT-B/16 with fp16 and bf16, and explains what the numbers actually mean for system selection. https://theloginnode.com/posts/hpc-from-scratch-06/ Sun, 07 Jun 2026 00:00:00 +0000 https://theloginnode.com/posts/hpc-from-scratch-06/ Without accounting, Slurm treats every user the same regardless of how much they have already consumed. Episode 6 of HPC From Scratch builds the full accounting layer: account hierarchy with sacctmgr, QOS policies with wall-time limits, and fair share scheduling with exponential decay. The result is a scheduler that adapts to actual usage and prevents any single user from dominating the queue. https://theloginnode.com/posts/hpc-from-scratch-05/ Wed, 20 May 2026 00:00:00 +0000 https://theloginnode.com/posts/hpc-from-scratch-05/ Getting Slurm running from source is more involved than installing from a package manager, but it gives you full control over version and build options. Episode 5 covers the complete setup: MUNGE for authentication, slurmctld on the controller, slurmd on compute nodes, and verifying everything works with a real batch job. https://theloginnode.com/posts/hpc-from-scratch-04/ Tue, 05 May 2026 00:00:00 +0000 https://theloginnode.com/posts/hpc-from-scratch-04/ A cluster where each node has its own home directory and user list is not really a cluster. Episode 4 fixes that by installing NFS for shared storage and FreeIPA for centralized authentication, so every node sees the same files and every user logs in with the same credentials from anywhere. https://theloginnode.com/posts/hpc-from-scratch-03/ Tue, 21 Apr 2026 00:00:00 +0000 https://theloginnode.com/posts/hpc-from-scratch-03/ Building a home HPC cluster does not require a perfect network on day one. Episode 3 covers setting up the login node on WiFi, planning a clean IP scheme for the cluster subnet, and getting SSH working reliably across all nodes while keeping the setup simple enough to upgrade later. https://theloginnode.com/posts/hpc-from-scratch-02/ Wed, 25 Mar 2026 00:00:00 +0000 https://theloginnode.com/posts/hpc-from-scratch-02/ Buying more RAM does not always mean your jobs get more of it. Episode 2 of HPC From Scratch covers the hardware upgrades made to the cluster nodes and explains how iGPU shared memory quietly reduces available system memory on compute nodes, with no error message to warn you. https://theloginnode.com/posts/hpc-from-scratch-01/ Fri, 13 Mar 2026 00:00:00 +0000 https://theloginnode.com/posts/hpc-from-scratch-01/ University clusters have waitlists and cloud HPC gets expensive fast. This series documents building a real 6-node HPC cluster from consumer hardware for $1,264. Episode 1 covers hardware selection, how to assign node roles, and the decisions that actually matter before you buy anything. https://theloginnode.com/posts/hpc-special-topics-01/ Mon, 16 Feb 2026 00:00:00 +0000 https://theloginnode.com/posts/hpc-special-topics-01/ Running rclone copy with default settings works, but it is usually much slower than it needs to be. This post covers how to configure Rclone for HPC workflows, benchmark real transfer speeds across cloud storage providers, and tune the parameters that actually affect throughput on a cluster network. https://theloginnode.com/posts/hpc101-04/ Wed, 28 Jan 2026 00:00:00 +0000 https://theloginnode.com/posts/hpc101-04/ Job failures on HPC clusters are frustrating, especially without clear feedback. This post breaks down how to use sacct, seff, and Slurm log files to figure out exactly what went wrong and how to prevent it next time. https://theloginnode.com/posts/hpc101-03/ Sun, 18 Jan 2026 00:00:00 +0000 https://theloginnode.com/posts/hpc101-03/ Shared HPC clusters do not give you root access, and the system Python is not yours to modify. This post shows how to set up isolated Python environments using Conda and venv so you can install whatever you need without permission errors or breaking someone else’s workflow. https://theloginnode.com/posts/linux101-01/ Fri, 09 Jan 2026 00:00:00 +0000 https://theloginnode.com/posts/linux101-01/ Most people’s first experience with the Linux terminal is copying a command from somewhere and hoping it works. This post takes a different approach: explain what is actually happening, cover the commands you will use every single day, and make the terminal feel like a tool instead of a trap. https://theloginnode.com/posts/hpc101-02/ Fri, 02 Jan 2026 00:00:00 +0000 https://theloginnode.com/posts/hpc101-02/ Copying files to a remote cluster is easy to get wrong. A basic scp works fine until your dataset is large and the connection drops halfway through. This post covers scp for quick transfers, rsync for large or resumable jobs, and git for code that belongs in version control anyway. https://theloginnode.com/posts/hpc101-01/ Sat, 27 Dec 2025 00:00:00 +0000 https://theloginnode.com/posts/hpc101-01/ Every HPC user starts somewhere, and it usually involves staring at a login prompt with no idea what to do next. This post covers the three things you need on day one: SSH access, loading software with the module system, and submitting your first Slurm batch job. https://theloginnode.com/posts/hello-world/ Tue, 16 Dec 2025 00:00:00 +0000 https://theloginnode.com/posts/hello-world/ The first post on The Login Node. Will Paik explains why he started the blog, what direction he is planning to take it, and how he is approaching it.