DEEPX DX-M1 brings 25 TOPS edge inference to an M.2 slot
DEEPX's DX-M1 is an M.2 AI inference accelerator priced at $139, with vendor specifications of 25 TOPS INT8 compute and 2W–5W power consumption. It includes 4GB of LPDDR5 memory and 128GB of QSPI NAND Flash, supports PyTorch, TensorFlow, TensorFlow Lite and ONNX through DXNN SDK, and targets computer vision, robotics and visual inspection rather than large-model training or 27B- and 70B-class LLMs. DEEPX CEO Kim Keun-won claims it runs 25°C–40°C cooler than other chips, but the reports provide no independent thermal measurements.
The DX-M1 offers edge-AI developers an inexpensive, low-power inference option in a standard M.2 form factor, although its headline thermal and performance figures remain vendor claims.


韓國人工智慧晶片新創 DEEPX 近日在美國展示低發熱 AI 晶片 DX-M1,強調運作時溫度表現明顯優於一般晶片。 執行長 Kim Keun-won 接受 CNBC 訪問時表示,與其他晶片相比,DX-M1 溫差可達 25°C~40°C,甚至用「奶油放在上面都不易融化」形容卓越低溫特性。
DEEPX 補充,DX-M1 是主打推論運算的 AI 加速器,採用 M.2 規格設計,可直接裝在主機板的 M.2 插槽,售價 139 美元。 這款晶片提供 25TOPS 的 INT8 算力,功耗僅約 2W~5W,非常適合地端電腦視覺、物件偵測、即時影像分析、機器人導航與視覺檢測等應用場景。
#DEEPX#DX-M1#DFRobot#DXNN SDK#PyTorch#TensorFlow#TensorFlow Lite#ONNX