{"slug": "strix-halo-tuned-rust-engine", "title": "Strix Halo tuned rust engine", "summary": "An open-source MIT-licensed project called Talus released a Rust-based inference engine for AMD's Strix Halo (Ryzen AI Max+ 395 with 128 GB), forked from Niko1221's C++/HIP engine Strata, with GPU kernels re-tuned for the gfx1151 target. Talus serves one model at a time through an OpenAI-compatible API, with a terminal UI for switching models and presets, and was benchmarked end-to-end through the server on an HP ZBook Ultra G1a capped at 70 W running headless Linux 7.0 across Qwen3.8-27B, Qwen3.8-Flash-Next, GLM-5.3-Flash, MiMo-V2.6-Flash and Laguna-S-2.1. The code keeps per-GPU tables and multi-device hooks for possible RDNA4 and multi-box support, and a built-in benchmark and validation tool including SWE-bench with anti-cheating is planned for the next release.", "body_md": "Talus is an open-source (MIT) inference engine for Strix Halo (Max+ 395/128G).\n\nThe engine is a fork of Strata, a C++/HIP engine by Niko1221. We ported Strata’s host and server to Rust, tried a few other popular models, and re-tuned the GPU kernels, which come from Strata and, through it, from ggml/llama.cpp, for gfx1151. It serves one model at a time through an OpenAI-compatible API. A terminal UI switches models and presets, and explains trade offs when changing settings.\n\nWhy?\n\nI started chasing the big MoE models that I could fit on this laptop, which spiraled into forking everycode base I could find for performance.\n\nThe Goods:\n\nHP ZBook Ultra G1a, Ryzen AI Max+ 395, 128 GB, firmware-capped at 70 W, headless Linux 7.0\n\nQwen3.8-27B, Qwen3.8-Flash-Next, GLM-5.3-Flash, MiMo-V2.6-Flash, Laguna-S-2.1.\n\nNumbers:\n\n(tok/s, end to end through the server, on an HP ZBook Ultra G1a capped at 70 W)\n\nHow we got here:\n\nWhat mattered for gfx1151:\n\nHonest Comparisons (same laptop, measured by us)\n\nCredits:\n\nStrata - (Niko1221): the base.\n\nggml / llama.cpp: kernels and formats.\n\nThe model teams: Qwen, Z.ai, Xiaomi MiMo, poolside.\n\nUnsloth: the GGUF quants\n\nz-lab: the DFlash drafters.\n\nGufo: the long-copy drafting idea.\n\nAMD ROCm.\n\nWhat’s next:\n\nTalus is tuned and tested only on Strix Halo today. The code keeps per-GPU tables amd multi-device hooks, so other AMD GPUs (RDNA4) and multi-box setups may come later. A built-in benchmark and validation tool (quality suites plus SWE-bench with anti-cheating) is coming in the next release.\n\nGithub: nsbradley88/talus-public\n\nHuggingFace: j2ipper", "url": "https://wpnews.pro/news/strix-halo-tuned-rust-engine", "canonical_source": "https://forum.level1techs.com/t/strix-halo-tuned-rust-engine/259719#post_1", "published_at": "2026-10-28 20:13:49+00:00", "updated_at": "2026-10-28 20:45:46.016322+00:00", "lang": "en", "topics": ["ai-infrastructure", "ai-tools", "large-language-models", "ai-chips", "developer-tools"], "entities": ["Talus", "Strix Halo", "AMD", "Ryzen AI Max+ 395", "Strata", "Niko1221", "ggml", "llama.cpp"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/strix-halo-tuned-rust-engine", "markdown": "https://wpnews.pro/news/strix-halo-tuned-rust-engine.md", "text": "https://wpnews.pro/news/strix-halo-tuned-rust-engine.txt", "jsonld": "https://wpnews.pro/news/strix-halo-tuned-rust-engine.jsonld"}}