Domestic AI low-power semiconductor solutions will be unveiled at the CES exhibition held in Las Vegas, USA. DeepX, a domestic AI semiconductor company, will introduce global business cooperation cases for carbon emission reduction and AI server performance optimization solutions, and expand its business through attracting new customers and business development.

DeepX announced on the 19th that it will participate in 'CES 2024', the world's largest IT and home appliance exhibition in January, and introduce its AI server product 'DX-H1'.

DeepX's DX-H1 won a CES Innovation Award in the Computer Hardware and Components category.

DeepX's 'DX-H1' is an AI inference solution that maximizes performance, power, and cost efficiency. It was selected for the Innovation Award for the following reasons: △ It is the most innovative product in computer systems and solutions that configure systems, including desktops and laptops, that will expand the AI era; △ It is expected to have a carbon reduction effect as it is more than 10 times more power efficient than existing GPU-based solutions.

The current AI server market relies heavily on GPU solutions, which are identified as a major contributor to carbon emissions due to their high power consumption. This has led to a growing demand for low-power solutions in data centers to improve energy efficiency and minimize environmental impact.

According to TrendForce, the AI server market is expected to grow at a compound annual growth rate of more than 36% through 2027, and demand for low-power AI accelerator solutions such as the 'DX-H1' is expected to continue to increase. Companies seeking to simultaneously reduce carbon emissions and optimize performance are considering low-power AI accelerator solutions, including the DX-H1, as options.

DeepX is proud to have secured the original technology as a low-power solution. DeepX's core technologies include 'IQ8', an INT8 model compression technology, and 'Smart Memory Access', a technology that minimizes DRAM usage to less than a few tenths of that of GPUs.

This development reduces the weight of complex AI models without compromising accuracy. DeepX emphasized that even using LPDDR, a low-power memory solution rather than expensive HBM, it delivers high AI computational performance and provides power efficiency more than 10 times that of GPUs.

Additionally, the DX-H1 is compatible with existing GPU solutions, providing an interface that supports GPU-trained AI models. This reportedly allows customers to easily replace their existing systems with low-power solutions, facilitating performance upgrades.

Kim Nok-won, CEO of DeepX, emphasized, “AI technology will become as commonplace as the Internet is today, and for this, the spread of edge AI and edge AI semiconductors are essential.” He added, “Low-power, high-performance AI server solutions that can support this remotely will also be in the spotlight.”

He also said, “This will be a golden time for DeepX to lead innovation and growth in AI technology,” and “We will challenge ourselves to lead the AI semiconductor market by providing customers around the world with power and performance efficiency that is 10 times higher.”