# Eyjafjallajökull 🔥❄️ *High-Performance Local AI Lab by Magnús Smárason* Eyjafjallajökull is a custom-built AI lab workstation designed for deep learning, model fine-tuning, synthetic data generation, and AI infrastructure prototyping. Built around dual NVIDIA RTX GPUs and a 24-core Intel CPU, the system is optimized for local-first workflows, data sovereignty, and rapid experimentation with large models. ## 🔧 System Overview ### 🖥️ Host Machine - **Hostname**: `MAGGI` - **OS**: [EndeavourOS](https://endeavouros.com) (Arch-based rolling release) - **Kernel**: `Linux 6.13.6-arch1-1` - **Chassis**: Custom Desktop Workstation - **Motherboard**: ASRock Z790 Taichi Lite ### 🧠 CPU - **Model**: Intel® Core™ i9-14900K - **Cores/Threads**: 24 cores / 32 threads - **Max Frequency**: Up to 6.0 GHz - **Cache**: 32 MB L2 / 36 MB L3 - **Virtualization**: VT-x ### 🧬 Memory - **RAM**: 128 GB DDR5 - **Swap**: 8.8 GB - **Available at boot**: ~112 GiB ### 💽 Storage - **1 TB NVMe SSD (OS)** – Mounted at `/` - **1 TB NVMe SSD (Data/Experiments)** – Mounted separately - **4 TB HDD (Cold Storage)** – Mounted at `/mnt/data-slow` --- ## 🎮 GPU Configuration – Dual RTX Setup ### 🔹 NVIDIA RTX 5090 *(Primary)* - Ada Lovelace Architecture - ~32 GB GDDR7 - Estimated ~18,000+ CUDA cores ### 🔹 NVIDIA RTX 4090 *(Secondary)* - 24 GB GDDR6X - 16,384 CUDA cores > Combined: Over 70 TFLOPs FP32 performance – ideal for LLMs, diffusion models, reinforcement learning, and multi-GPU training. --- ## 🌐 Networking - **Ethernet (enp0s31f6)** – Active - **Wi-Fi (wlan0)** – Connected - **Docker Networks** – Multiple bridge interfaces for container orchestration --- ## 🧰 Running Services - `Docker`, `containerd` - `Gitea Actions Runner` (Local CI/CD) - `RealtimeKit`, `Avahi`, `Polkit`, `systemd` stack - Fully containerized experimentation environment --- ## ⚙️ System Details ```bash CPU: Intel Core i9-14900K (24 cores, 6.0 GHz) RAM: 128 GB DDR5 GPUs: RTX 5090 + RTX 4090 (dual GPU setup) Disk: 2 TB NVMe + 4 TB HDD OS: EndeavourOS (Arch) Kernel: Linux 6.13.6-arch1-1