-
Doubling Memory
Upgrade the home server to improve large language model performance by increasing RAM to 64 GiB and enabling XMP for better speed. Despite installation challenges, this optimization aims to improve model efficiency using different techniques. Read More
-
Bonsai 27 B on old GPU’s
Primarily using Qwen3:8b for analysis on a 3080 Ti GPU, exploring larger models with improved performance. After significant installation challenges with Bonsai:27B, I achieved satisfactory results, balancing memory requirements and processing speed across both GPUs. Read More
-
1080 Ti Returns
Upgraded my server with an old 1080 Ti GPU, enhancing its capacity by nearly doubling the available VRAM. This change cut processing time for a batch job from 47 days to about 37 days, improving overall efficiency. Read More
-
Speeding Up Ollama
Discuss optimizing performance while managing tasks with limited GPU resources. I implemented a continuous job queue and a web interface to prioritize tasks with the Hermes Agent. Future enhancements include automating document processing and utilizing different models for varied tasks. Read More
-
Splitting WHM Accounts
A challenging day managing subdomains and SSL certificates on a cPanel account. I document the process of converting a domain, handling automation issues, file deletions, and ultimately finding efficient ways to create subdomains while addressing SSL configuration challenges. Read More
-
The Backup That Wasn’t Catching Up
Backblaze reported reaching the daily storage cap. Old backup versions accumulated because of default settings in Backblaze’s B2 bucket, leading to excessive storage usage. I adjusted settings to manage file retention more effectively, underscoring the importance of understanding backup systems. Read More
-
When Security Got Heavy
A failed Forgejo push exposed a three-hour server outage. Recent installs like Forgejo and Nextcloud seemed suspicious at first, but diagnostics pointed to an aggressive ImunifyAV scanner. After mitigation, I resized the server and planned safer monitoring. Read More
-
Unsupported Should Not Mean Blind
After a DigitalOcean monitoring failure filled my server logs, the final support answer was “unsupported.” This follow-up explains why disabling disk metrics is not enough, and why safer filesystem collection modes could help unusual server environments. Read More
-
Solving SSL Before Press
Days before a reporter visited the Restoring the Signal exhibit, I found myself untangling a bizarre SSL certificate problem involving DNS, AutoSSL, browser caching, and cPanel—while also preparing rotary phones, TTY demonstrations, flyers, and interactive storytelling experiences. Read More
-
Death by Metrics
For days, my server showed elevated CPU usage while remaining mostly functional. Then websites, WHM, and cPanel suddenly failed. Through SSH, I traced the problem to a runaway DigitalOcean monitoring agent interacting with CloudLinux CageFS, generating nearly 95 GB of endless filesystem metric errors. Read More