The power consumption is so low that the cream does not break down! AI chip DX-M1 released: 2W power consumption for 933 yuan

Fast Technology reported on November 9 that South Korean AI chip startup DEEPX recently demonstrated an AI chip DX-M1 that mainly pushes down heat generation in the United States. In an interview, its CEO declared that the temperature performance of the chip is significantly better than that of ordinary chips, with a temperature difference of 25° C to 40° C, and even if you put the cream on it, it will not melt easily.

As an AI accelerator featuring inference calculations, the DX-M1 is designed with the standard M.2 specification. This means that the user does not need a complicated expansion card and can plug it directly into the M.2 slot on the motherboard of the computer.

On core parameters, the DX-M1 provides INT8 computing power of 25 tops, while power consumption is only between 2W and 5W. This extremely low power control makes it ideal for scenarios where heat dissipation and noise requirements are extremely high, such as local computer vision, object detection, real-time image analysis, and robot navigation.

In terms of hardware configuration, the DX-M1 has built-in 4GB LPDDR5 memory and 128GB NAND flash memory, and interacts with the host through the PCIe Gen 3 interface. It is capable of supporting mainstream AI frameworks such as PyTorch, TensorFlow, TensorFlow Lite and ONNX, and comes with a dedicated DXNN software development kit.

To be clear, DEEPX positions the chip as a pure edge AI acceleration card. Officially, it is not used to train large-scale models, nor is it suitable for large-scale large language models at the 70b or 27b parameter level, but instead specializes in end-side AI inference tasks with low power consumption, low noise, and stable heat dissipation.

The product is currently priced at $139 (FastTech Note: about RMB 933).

Source: MyDrivers (Fast Technology). Content is presented with source attribution. Read the original report.