Added a new device to my @tiles.run cluster. Welcome to the lineup, M5 Pro with the 32 GB/1 TB spec.
Nov 25, 2025 at 5:19 AM
mzau/broke-cluster
A Poor Man's Apple Silicon LLM Cluster — tuned for MLX, scalable without shame.
EXO Labs reveals that they have been working with Apple for the past year on low-latency RDMA networking over TB5 which allows a cluster of 4 x M5 Ultra Mac Studios to scale to an aggregate memory bandwidth of 4.8TB/s
Megakernel: Matching Apple Silicon Efficiency at 2x the Throughput on a RTX 3090
The first megakernel for hybrid DeltaNet/Attention LLMs. 413 tok/s at 1.87 tok/J on a 2020 RTX 3090, matching M5 Max efficiency at 1.8x throughput.
AMD and Anthropic Announce Strategic Partnership to Deploy Up to 2 Gigawatts of AMD Instinct MI450 Series GPUs
News Highlights Anthropic to deploy up to 2 gigawatts of MI450 Series GPUs in AMD Helios rack-scale solutions, with deployment of the first…...
Tiles version 0.4.5 Alpha 9 has been released. tiles.run/download Adds peer-to-peer device linking for both offline and online networks, built with @iroh.computer, along with bug fixes for the auto-update system.
Download Tiles
tiles.runTiles version 0.4.19 Alpha 23 has been released. MTP is now opt-in, with a new --mtp flag for tiles run and persistent configuration under [llama]. Inference server warnings are now surfaced in the CLI, and fixed an issue where tiles update could install multiple versions. Release notes ↗
Shared chat session by @tiles.run | Tiles
chat.tiles.runTiles version 0.4.13 Alpha 17 has been released. Adds Linux support with llama.cpp, configurable inference runtime controls, capability-based P2P syncing with UCAN, improved device linking, and reliability improvements for tool calling and streaming. Release notes: tiles.run/share/YXQ6Ly9kaWQ6cGxjOnZreGY…
Shared chat session by @ankeshbharti.com | Tiles
www.tiles.runTiles version 0.4.2 Alpha 6 has been released. New onboarding flow for account and data setup, an OTA updater, and Harmony renderer support for gpt-oss-20b, enabling configurable reasoning effort and improved response quality. Release notes: tiles.run/changelog#0.4.2
Tiles Changelog
www.tiles.runTiles version 0.4.6 Alpha 10 has been released. tiles.run/download Adds peer-to-peer device sync, built with @iroh.computer's QUIC networking stack, and encryption at rest for local chats.
Download Tiles
tiles.runTiles version 0.4.4 Alpha 8 has been released. tiles.run/download Adds support for a fully offline, portable installer alongside the network installer, with the default gpt-oss-20b model bundled, and daemon process implementation for handling background tasks reliably.
Download Tiles
tiles.runTiles version 0.4.17 Alpha 21 has been released. The new default model is Gemma 4 12B, using Unsloth’s Q4_K_M GGUF. Added quantization tags to Modelfiles and automatic MTP decoding. Upgraded Pi to upstream v0.84.2, plus minor fixes for macOS notarization and GGUF model downloads. Release notes ↗
Shared chat session by @tiles.run | Tiles
chat.tiles.runNew project added to our project list! @tiles.run Local first AI using atproto for data sharing - seems esp. useful for labs interested in sovereign compute or working on sensitive data
Tiles: Own your AI
www.tiles.runTiles version 0.4.8 Alpha 12 has been released. tiles.run/download Adds Pi as an embedded agent harness, introduces chat sessions, and supports sharing chats publicly via ATProto, with an improved TUI featuring better help text and slash commands.
Download | Tiles
tiles.runGemma 4 now runs 2x faster with MTP GGUFs! Run locally on just 6GB RAM. ⚡️ MTP enables Google Gemma 4 run ~1.4–2.2× faster with no accuracy loss. Gemma 4 12B MTP can run at 162 t/s vs. 52 t/s without MTP. 31B reaches 101 t/s. GGUFs + Guide: unsloth.ai/docs/models/mtp
