Asus launches the UGen300 USB AI Accelerator. This device brings an NPU to your PC via USB, although the computing power of the thing is not exactly spectacular.
For those who have a laptop or desktop without an NPU and feel envious of people who say they have an ‘AI PC’, Asus is launching the UGen300 USB AI Accelerator. It is a small box containing a Hailo-10H accelerator. The Hailo chip is an NPU with a computing power of 40 TOPS. Hailo is an Israeli company specializing in such chips, with whom Asus is now partnering.
Dedicated RAM
The external accelerator measures 10.5 cm x 5 cm x 1.8 cm and consumes 2.5 watts of power, supplied via USB-C. The device itself has 8 GB of (LPDDR4) RAM on board. Asus does not shy away from big words, speaking of ‘unrivaled edge Gen AI performance’. We suspect this claim comes from the marketers rather than the engineers, because as far as AI performance is concerned, the Asus UGen300 USB AI Accelerator mainly provides a modest upgrade.
The external AI performance is comparable to that of a built-in NPU in recent Windows laptops. 40 TOPS is sufficient to efficiently power targeted AI functionality: think of background blur in Teams, image recognition, or a specific function in a photo editing application. 8 GB of RAM and 40 TOPS are by no means enough to run AI models with capabilities reminiscent of what is available via the cloud today.
Niche product
This makes the Asus UGen300 USB AI Accelerator primarily a niche product. The main advantage of a built-in NPU is efficient support for light AI workloads. Usually, the CPU and GPU can also handle these workloads, albeit less efficiently. The extra 40 TOPS can certainly offer added value, but you need an extra piece of hardware for it. We can’t think of many scenarios where that would be worth the effort.
If you want to provide a laptop on your desk with extra computing power for AI, there are also more powerful options such as an external GPU. This immediately supports heavier workloads. Although Asus has not communicated a price, you can assume that the Asus UGen300 USB AI Accelerator is much cheaper compared to an eGPU.
M.2 version
The product may not be broadly relevant, but it is interesting. With the UGen300 USB AI Accelerator, Asus is bringing an accessible piece of hardware to the market that allows any computer to get a bit of hardware-based acceleration for AI workloads.
Perhaps even more relevant is the M.2 version of the Hailo-10H. Asus indicates that it will also offer the Hailo chip in that format. In systems where an M.2 slot is available, you can integrate extra TOPS on the inside.
The Asus UGen300 USB AI Accelerator is compatible with Windows, Linux, and Android. The chip supports frameworks including PyTorch, TensorFlow, and ONNX. Asus indicates that a UGen Utility will be released soon, giving users quick access to pre-trained and compatible models. The accelerator itself will hit the market this quarter.