• AMD Ryzen AI Max+ 395 Processor — Flagship Ryzen AI processor with 60 FP16 TFLOPS GPU performance and up to 50 TOPS NPU for local AI inference and development
• 128GB LPDDR5x Unified Memory — Large unified memory pool for running up to 200B parameter models locally without cloud dependency
• Up to 3x Faster Than Apple M4 Pro — Outperforms Apple M4 Pro on image generation workloads including Stable Diffusion XL, Flux, and Hunyuan 3D
• Up to 14% More Tokens/Sec Than NVIDIA DGX Spark — Faster LLM inference vs DGX Spark on GPT OSS 120B, Qwen, and GLM models at a lower price point
• Windows & Linux OS Support — Run AI workflows on either OS — an advantage over NVIDIA DGX Spark which supports Linux only
• Up to 50 TOPS NPU — Dedicated neural processing unit for accelerated AI workloads alongside the GPU
• Pre-Configured AI Software Stack — Ships ready to build with AMD ROCm, Ollama, LM Studio, PyTorch, vLLM, and more
• AMD Ryzen AI Developer Center App — Central hub for playbooks, tools, software sync, and developer resources

The AMD Ryzen AI Halo is AMD’s purpose-built local AI developer platform — designed to go from power-on to running AI models in minutes without cloud dependency or complex configuration. Powered by the Ryzen AI Max+ 395 processor with 128GB of LPDDR5x unified memory, it runs models up to 200B parameters locally, delivering predictable cost-per-token versus cloud API billing. AMD ROCm software provides class-leading generative AI performance with Day 0 model and framework support.

In independent testing, the Ryzen AI Halo delivers up to 14% more tokens per second than NVIDIA DGX Spark across LLM benchmarks — at $700 less. It outperforms Apple M4 Pro by up to 7.3× on AI image generation tasks including Flux, Stable Diffusion XL, and Hunyuan 3D. Over a 3-year modeled scenario, local inference on the Ryzen AI Halo costs up to 6× less than equivalent cloud API usage. Available in Windows 11 Pro and Linux OS configurations.

Ideal For

• AI Developers & Researchers — Local LLM inference, fine-tuning, and model development without cloud costs
• Generative AI & Image Generation — Run Stable Diffusion XL, Flux, Hunyuan 3D, and ComfyUI workflows locally at full speed
• LLM Inference at Scale — Support up to 200B parameter models on 128GB unified memory with AMD ROCm
• AI Agent Development — Real-time reasoning, persistent memory, and automation without token costs
• Privacy-First AI Workloads — All inference runs locally — no data leaves the machine

Variant Options

• Windows 11 Pro — AMD Ryzen AI Max+ 395, 128GB LPDDR5x, Windows 11 Pro
• Linux OS — AMD Ryzen AI Max+ 395, 128GB LPDDR5x, Linux OS

Specifications

• Processor — AMD Ryzen AI Max+ 395
• Memory — 128GB LPDDR5x (unified CPU+GPU)
• Max Model Size — Up to 200B parameters
• GPU Performance — 60 FP16 TFLOPS
• NPU Performance — Up to 50 TOPS
• OS Options — Windows 11 Pro / Linux
• Software — AMD ROCm, Ollama, LM Studio, PyTorch, vLLM, LoRA, QLoRA, ComfyUI