Samsung claims to have built an AI chip with Naver that is eight times more efficient than the Nvidia H100. However, many details about the chips are not known.

Samsung has teamed up with Korean web giant Naver to develop its own AI chip. The companies say it is eight times more efficient than the market leader Nvidia H100. Details about the chip are scarce, but the device is intended primarily for inference. The chip is based on an FPGA and is optimized for the Naver HyperCLOVA X Great language model. It is not clear how scalable the features are to other models and applications.

prototype

At the moment, the chip is just a prototype, with no clear plan for its general availability. We also know little about the specifications, although Samsung claims that the efficiency is due to the use of LPDDR memory. Samsung has great expertise in memory production, so it’s not surprising that that’s where the focus of the new inference chip lies.

Naver and Samsung have been working together since 2022. The chip expertise comes from Samsung, while Naver builds AI applications. The two parties believe that tailored solutions focused on current AI needs are the future.