[Edaily Reporter Park Jeong-soo] #NOTA is showing strong gains. This is believed to be driven by buying interest following news that the company has implemented large language model (LLM) inference optimization technology based on heterogeneous computing, which utilizes both GPUs and NPUs in an AI PC environment.
According to MP Doctor on the 4th, as of 9:10 a.m., Nota is trading at 39,650 won, up 13.61% (4,750 won) from the previous trading day.
On this day, Nota announced that it has implemented optimization technology for large language model (LLM) inference based on heterogeneous computing. Heterogeneous computing is a method in which processors with different characteristics—such as CPUs, GPUs, and NPUs—share computational tasks by leveraging their respective strengths. In AI PCs, the key factor determining actual performance is not the performance of a single chip, but rather how efficiently various computational devices are utilized.
NOTA applied "Disaggregated Inference" technology to Intel Luna Lake-based AI PCs, dividing the LLM operation process into input processing and response generation stages and assigning the optimal computing device to each. Accordingly, the design assigns input processing to the GPU and response generation to the NPU.
Performance evaluations showed that the Disaggregated Inference approach reduced energy consumption per token by approximately 32% compared to a single-GPU configuration, while improving generation throughput by about 12%. Additionally, compared to a single-NPU configuration, it reduced the first-response latency by approximately 89%.
The company explained that the core of this technology lies not in simply utilizing the GPU and NPU in parallel, but in optimally allocating computations according to the specific characteristics of the AI model’s tasks. This implies that user experience can vary depending on how the hardware is utilized, even on the same AI PC.
Recently, the adoption of heterogeneous computing and decoupled inference technologies has been expanding across the global AI industry. At Computex 2026, major companies such as Intel and NVIDIA unveiled AI PCs combining CPUs, GPUs, and NPUs, while in the data center sector, companies like NVIDIA and Amazon Web Services (AWS) are implementing architectures that decouple AI computations into distinct stages.
NOTA emphasized that through this technological implementation, it has demonstrated its capability to not only streamline AI models but also optimize execution environments and hardware utilization. In particular, the company expects this to provide a distinct competitive advantage in on-device AI environments, where AI must be run within limited power and computational resources.