跳到正文
The Elec· KIMGEON·· 13 小時前AI 評分50

AMD 為 Unsloth Desktop 加入 NPU 支援

AMD Adds NPU Support to Unsloth Desktop

AI 導讀

AMD 為 Unsloth Desktop 加入基於 XDNA 2 架構的 NPU 支援,讓該軟件首次支援 NPU。搭載 AMD XDNA 2 NPU 的 PC 可在本機執行受支援的 AI 模型;Unsloth Desktop 的 API 可整合超過 100 個 AI 智能體應用程式及開發環境,包括 Cline、Open WebUI 和 Visual Studio Code。

正文

(Photo: AMD)

(Photo: AMD)

(Photo: AMD)
(Photo: AMD)

AMD has collaborated with artificial intelligence (AI) software company Unsloth to demonstrate technology for running large AI models on neural processing units (NPUs) in PCs.

AMD said on Nov. 19, local time, that it had added support for NPUs based on its XDNA 2 architecture to Unsloth Desktop, Unsloth's AI inference software. This marks the first time Unsloth Desktop has supported NPUs.

PCs equipped with AMD XDNA 2 NPUs can run supported AI models locally without relying on cloud servers. The technology integrates AI model execution software Lemonade and FastFlowLM (FLM) to process AI inference workloads on the NPU.

Unsloth Desktop's application programming interface (API) also enables integration with more than 100 AI agent applications and development environments. Examples include the AI coding tool Cline, Open WebUI and Visual Studio Code. Each application can access AI models running locally on the PC through the API.

AMD also demonstrated a 35-billion-parameter Qwen 3.6 mixture-of-experts (MoE) AI model running on an NPU. Running the model at its full context length requires approximately 28 gigabytes (GB) of memory allocated to the NPU. Under Windows' default shared memory allocation settings, 64GB of system memory is required.

AMD said using the NPU can reduce computational workloads on the central processing unit (CPU) and graphics processing unit (GPU) while improving power efficiency. The company said the technology enables AI agents to run continuously at low power consumption, even on battery-powered laptops.

來源:The Elec · thelec.net