Asus unveils "world's first USB edge AI accelerator for both classic AI and generative AI"

Expansion cards for PC systems are nothing new, with dedicated graphics cards being a must for demanding gamers. We've previously reported on expansion cards for accelerating AI models, although these are probably no longer strictly necessary for many modern PCs thanks to NPUs integrated into APUs. Nevertheless, such expansion cards can still be useful for increasing the AI ​​performance of existing systems. Asus has now announced one such card, the UGen300.

This is a USB expansion card that communicates via USB 3.1 Gen 2 and so requires no installation into a PCI slot nor opening of the PC case. The device is equipped with a Hailo 10H NPU, which is said to achieve a performance of 40 TOPS (INT4), and its typical power consumption is specified at 2.5 watts, making it very low to practically negligible. 8 GB of LPDDR4 RAM is on board, which is unlikely to be sufficient for hardcore AI enthusiasts since dedicated AI builds usually have well over 100 GB of RAM.

It is reportedly compatible with both x86 and ARM platforms, Windows, Linux and Android devices, as well as various frameworks such as TensorFlow, PyTorch, and ONNX. According to Asus, over 100 pre-trained models are available for use. A typical application could be advanced image or pattern recognition and interpretation. No information on price or availability is available at this time.