Local AI Infrastructure Notes (1/15) — Mac Mini M4 Home Server Architecture: ARM Unified Memory

Hardware selection rationale for local LLM inference, macOS server configuration, and storage separation architecture

Summary

  • The Mac Mini M4's ARM unified memory architecture makes it well-suited for local LLM inference (ollama), with extremely low power draw
  • Running macOS as a server requires explicit configuration of sleep, auto-restart, and update suppression — defaults are not production-safe
  • Routing projects to an external SSD separates OS and working data, enabling independent lifecycle management

Hardware Selection Rationale

The first decision in a home server build is hardware. The options are broad — Raspberry Pi, Intel NUC, repurposed desktop, Synology NAS.

When the primary workload is local LLM inference, the comparison axis shifts.

Running local models via ollama is GPU-memory-bound. On a standard x86 machine, that means a discrete GPU — and 16 GB VRAM cards carry significant cost. The Mac Mini M4 takes a structurally different approach.

ARM Unified Memory Architecture

The defining characteristic of the Mac Mini M4 is its memory architecture.

On a conventional PC, the CPU and GPU operate on independent memory pools (RAM and VRAM). Data transfers between these pools traverse the PCIe bus, and that transfer itself introduces latency.

In ARM unified memory architecture, the CPU, GPU, and Neural Engine all share a single memory pool. On a 16 GB model, the full 16 GB is available to LLM inference — there is no memory-copy overhead. The practical result: 7B–14B parameter models run at usable inference speeds without a discrete GPU.

Power Consumption Comparison

On a 24/7 server, power efficiency is a core operating cost variable.

Device Idle Power Load Power
Mac Mini M4 ~5–7 W ~30–40 W
Intel NUC ~10–15 W ~40–65 W
Standard desktop ~50–80 W 200 W+

Monthly electricity estimate (assuming 15 W average): - 15 W × 24 h × 30 days = 10.8 kWh - Approximate residential cost varies by region and tariff — verify against your specific rate plan

Upfront Cost Structure

Based on the Mac Mini M4 base model (16 GB RAM, 256 GB SSD):

Item Cost
Mac Mini M4 unit ~$550–620 USD (verify at time of purchase)
External SSD (1 TB) ~$70–100 USD (verify at time of purchase)
Total ~$620–720 USD

If a Mac Mini is already on hand, the incremental cost is limited to the external SSD. Against a comparable dedicated server build (Intel NUC + RAM + SSD, new), total cost is similar or lower. Prices shift over time — confirm current pricing before purchase.

macOS Server Configuration Checklist

macOS defaults are tuned for personal workstation use. Sleep mode, automatic updates, and FileVault are enabled by default — none of these are appropriate for always-on server operation. The following must be set explicitly.

Terminal configuration:

sudo pmset -a sleep 0
sudo pmset -a disksleep 0
sudo pmset -a displaysleep 0

sudo pmset -a autorestart 1

sudo pmset -a womp 1

Omitting disksleep causes the external SSD to enter standby independently, resulting in unmount events. sleep and disksleep must always be set together.

System Settings:

  • Enable auto-login: System Settings → Users & Groups → Automatic Login. If the machine stops at the login screen after a restart, services will not come up.
  • Disable automatic software updates: System Settings → General → Software Update → disable automatic updates. macOS can enter a "restart pending" state after a background update, causing unplanned downtime. Apply updates manually at a controlled time.
  • Review FileVault: With FileVault enabled, setting autorestart 1 causes the system to halt at the decryption password prompt after a power loss — unreachable remotely, requiring physical intervention. Evaluate the security vs. operational tradeoff explicitly before deciding.

Storage Separation Architecture

Rather than concentrating everything on the internal SSD, split storage by function.

Separation structure: - Internal SSD: macOS system + system services + ollama model cache - External SSD: all project directories (including git repositories)

Benefits of this structure:

  1. OS–data independence: On macOS reinstall or hardware replacement, transferring the external SSD preserves all project data intact.
  2. Simplified backup: Backup policy scopes to the external SSD as a unit.
  3. Capacity flexibility: A 256 GB or 512 GB internal drive cannot comfortably hold both model files and project files simultaneously. An external SSD removes that ceiling.

Over Thunderbolt 4, external SSD read/write speeds are indistinguishable from internal in practice.

Known Configuration Pitfalls

disksleep omission: Setting pmset sleep 0 without disksleep leaves the disk able to sleep independently, blocking external SSD access. All three values — sleep, disksleep, displaysleep — must be set together.

Unmanaged automatic updates: macOS automatic updates run overnight and can leave the system in a "restart required" state. Services continue running, but the next reboot applies the update and introduces unplanned downtime. Automatic updates must be disabled.

FileVault + autorestart conflict: With FileVault active, autorestart 1 causes the system to halt at the decryption prompt after a power loss. These two settings are mutually exclusive in practice.

Conclusion

Using the Mac Mini M4 as a home server is not a conventional approach. Linux is the standard substrate for server operation, and macOS has default behaviors that are actively inconvenient for always-on use.

The case for this configuration is narrow but clear: when local LLM inference is the core workload and the structural advantage of ARM unified memory is the deciding factor. Power efficiency and general-purpose capability support that decision.

The configuration checklist and storage separation structure documented here address the operational failure modes specific to this setup before they occur.

Series overview: Series index

๋Œ“๊ธ€

๋Œ“๊ธ€ ์“ฐ๊ธฐ

์ด ๋ธ”๋กœ๊ทธ์˜ ์ธ๊ธฐ ๊ฒŒ์‹œ๋ฌผ

Agent Memory Engine (2/10) — Building an AI Agent Memory System with SQLite Alone

← 1/10 SQLite Memory Engine f… ๐Ÿ“š Series Index 3/10 memcore Completed → A memory engine that runs without daemons, without dependencies, anywhere ํ•ต์‹ฌ ์š”์•ฝ This post covers how to design an AI agent memory engine (memcore) on a single-file SQLite backend. - Structure: 25 modules, ~6,464 lines, 22 DB tables - Principles: no daemon, single-file portability, no required external dependencies, host-agnostic - Core techniques: 4-layer hybrid search (topic · vector · FTS5 · LIKE), U-tag dialectic-based confidence evolution, ingestion-layer bias rejection rules What This Post Covers One approach to building AI agent memory without a server or cloud vector DB. Specifically: (1) the limitations of text-file-based memory, (2) vector DB alternatives compared and selection criteria, (3) how to layer search tiers, (4) how to evolve confidence scores over time, and (5) how to block biased memory accumulation at the ingestion stage. Design Principles: Why Single-File SQLite Text-file-based ...
Read more »

"ML Foundations (9/9) — PyTorch vs TensorFlow, and the Road to Local LLMs"

์ด๋ฏธ์ง€
← 8/9 Deep Learning Architec… ๐Ÿ“š Series Index (series end) The final part. So far we've focused on models. Now we focus on the tools that actually run them . We'll lay out the philosophical difference between PyTorch and TensorFlow, and trace how Hugging Face transformers , llama.cpp , MLX , and Ollama built the bridge to running large language models on your own machine. By the end you should have the full mental model of "download a pretrained LLM and serve it locally." 0. Learning Objectives Explain the eager-vs-graph difference between PyTorch and TensorFlow. Explain, in graph terms, how autograd automates the backward pass. Use Hugging Face transformers ' AutoModel , AutoTokenizer , and pipeline abstractions. Describe llama.cpp's GGUF quantization, INT4 inference, and CPU-first flow. Describe Apple MLX's use of unified memory and how it differs from PyTorch. Run a local LLM with Ollama and call it through an OpenAI-compatible API. ...
Read more »

"RAG Core Study (14/26) — Evaluation Sets with RAGAS & DeepEval"

์ด๋ฏธ์ง€
Series overview: Series index ๐Ÿ“š Series Index Without an evaluation set, every RAG improvement is a story, not evidence. RAG systems fail in more than one place: retrieval, context selection, answer generation, and grounding. That means "looks better" is not a metric. Part 14 explains how to build a golden dataset , how RAGAS and DeepEval fit into the workflow, and why teams must separate retrieval quality from answer quality if they want tuning decisions to hold up. 0. Prerequisites Part 13 reranking — system quality now depends on multi-stage ranking. Part 1 grounding — a correct-looking answer is not enough. Part 12 Hybrid — multiple retrieval settings need comparison on fixed data. 1. Learning Objectives Build a small but useful golden dataset for RAG. Distinguish retrieval metrics from answer metrics . Use RAGAS and DeepEval in the right roles. Avoid the common traps in judge-model-based evaluation. 2. ํ•ต์‹ฌ ์š”์•ฝ An evaluation set for RAG is ...
Read more »

"ML Foundations (8/9) — Deep Learning Architectures: CNN, RNN, Attention"

์ด๋ฏธ์ง€
← 7/9 Deep Learning Training ๐Ÿ“š Series Index 9/9 PyTorch vs TensorFlow → What did we keep adding on top of an MLP? Vision went CNN, sequences went RNN/LSTM, and eventually both converged on Attention and the Transformer. Each architecture is easier to remember if you read it as what it gave up to gain something else . This part is the one-line summary of the decisive inflection points: LeNet → AlexNet → ResNet → LSTM → Attention → Transformer. 0. Learning Objectives Explain why CNN's convolution, pooling, and weight sharing match images so well. Trace what got unblocked from LeNet → AlexNet → VGG → ResNet as depth grew. State BPTT for RNNs and the vanishing/exploding gradient problem. Write the LSTM gate equations (forget, input, output) with intuition. Write the attention formula in query/key/value form and the scaled dot-product variant. State the precise reasons Transformers replaced RNNs (parallelism + long-range dependencies). 1. ํ•ต์‹ฌ ์š”์•ฝ CNN : weight sharin...
Read more »

"ML Foundations (7/9) — Deep Learning Training: Optimizers, Regularization, Initialization"

← 6/9 Neural Networks ๐Ÿ“š Series Index 8/9 Deep Learning Architec… → Building an MLP, as we did in Part 6, is not the same as getting it to train well . The right optimizer, regularization, initialization, and learning rate — when these line up, deep networks converge. When they don't, the network refuses to learn at all. This part is the catalog of those crafts, with formulas, paper citations, and code in one place. 0. Learning Objectives Compare and write the update rules for SGD, Momentum, Nesterov, and Adam. Explain how Dropout, BatchNorm, and LayerNorm work and where they belong in a model. Derive the variance formulas for Xavier and He initialization and match them to activations. Implement step, cosine, and warmup learning-rate schedules in PyTorch. Explain why gradient clipping is effectively required for RNNs and Transformers. Diagnose the most common training failures (NaN, plateau, overfitting) and apply first-line fixes. 1. ํ•ต์‹ฌ ์š”์•ฝ SGD : \(w \leftarro...
Read more »

OpenClaw to Hermes Migration (2/13) — What to Preserve, Partially Port, or Discard

← 1/13 Current Structure Inve… ๐Ÿ“š Series Index 3/13 What Moves to Hermes → A code review record classifying operational scripts into three tiers: preserve, partial-preserve, and discard. What This Post Covers An asset classification framework (preserve / partial-preserve / discard) for migrating a legacy agent system to a successor A concrete case applying the Strangler Fig + Subset Migration strategy to a memory pipeline Classification results evaluated across LOC, external dependencies, and replaceability axes A method for slimming down a bloated single file (1,310 LOC) using the migration as a natural entry point Problem Definition: Port Everything or Cut Deep? The OpenClaw inventory comprises multiple agents, multiple scheduler/daemon processes, and multiple operational scripts. When moving to the successor system, Hermes, the choice is not binary. A full port carries accumulated technical debt along with it; a minimal port discards hard-won domain logic. Code-revie...
Read more »

AI Agents I Built (5/7) — Building an Automated Blogger API Publishing System

← 4/7 My Life Organizer Agent ๐Ÿ“š Series Index 6/7 Dual → Designing a Markdown-Based Auto-Publish Pipeline with Blogger API v3 + OAuth 2.0 Key Summary Blogger API v3 + OAuth 2.0 enables a pipeline that converts Markdown files into publishable HTML posts blogger_publish.py handles the full flow — frontmatter parsing → body cleanup → HTML conversion → CSS injection → API post — in a single script For bulk publishing, quota management (delay, retry, TOC suppression) is the critical design concern Platform Selection Criteria API access is mandatory for AI agent-driven content publishing. A comparison of platforms: Platform API Cost AdSense Decision WordPress.com REST API Paid plan required Paid only Cost overhead Ghost REST API Self-hosted or paid Manual setup Ops overhead Blogger v3 API Free Native integration Adopted Blogger is free, exposes a REST API, and integrates natively with AdSense. For personal blog automation, this combination i...
Read more »